Files

T

roboomp fb6a62c227 fix(shell): preserved negative pid operands in kill

Restricted -sigspec parsing to the option position and consumed the -- end-of-options marker, so negative PIDs (process groups) and post-marker operands are signaled rather than parsed as signals.

Added a process-group regression covering kill -TERM -- -<pgid> <pid>.

Fixes #6779

2026-07-27 09:27:57 +00:00

1.5 MiB

Raw Blame History

Changelog

[Unreleased]

Fixed

Fixed the bash tool's kill builtin rejecting numeric signals and multiple process operands, and changed its default signal from SIGKILL to the standard SIGTERM. Negative PID operands (process groups per kill(2)) and the -- end-of-options marker are now handled instead of being misparsed as signals (#6779).

[17.1.5] - 2026-07-27

Added

Added a configurable per-request timeout for the inspect_image tool (inspect_image.timeoutMs, default 5 minutes; set to 0 to disable) so a stalled vision-model provider fails fast with a clear error instead of blocking until manual abort (#4165).

Changed

Reduced default startup resident memory by constructing the default-off ComputerTool ArkType schema only on first parameter access, then reusing it across tool instances without changing validation or tool behavior (#6742 by @usr-bin-roygbiv).
Reduced startup CPU and memory by loading the bundled changelog only when needed, while preserving source, npm bundle, standalone binary, and native absolute-path fallback resolution.
Moved PTY log replay into the shared project launch broker, so normal CLI and Hub startup no longer load the xterm runtime while launch logs return validated rendered terminal rows.

Fixed

Fixed DeepSeek V4 Flash and Step 3.7 Flash models using hashline edit mode by default despite repeatedly misreading its range grammar; both now use the simpler replace-mode fallback unless explicitly overridden (#6671).
Fixed an Ask form appearing while the main prompt contains a draft hiding that text and consuming the next in-flight keystroke. The draft now remains visible and keeps receiving input until it is submitted or cleared; only then do form controls activate (#6737).
Fixed glob rejecting safe memory://root/<directory>/** patterns. Memory globs now resolve their directory prefix inside the project memory root while rejecting traversal and percent-encoded path separators across the complete glob path.
Fixed omp --resume <id> prompting to fork sessions from another existing directory instead of switching the process and cwd-scoped settings into the resumed session's recorded directory (#6752).
Fixed deferred CLI model roles resolving ambiguous bare model IDs to a preferred but unauthenticated provider instead of the authenticated provider selected by the eager path (#6727).
Fixed Windows sessions crashing with an unhandled EPIPE: broken pipe, write when an LSP server closed its stdin between filesystem mutations; LSP writes now observe asynchronous FileSink.write() failures and route them through the existing request/notification failure path.
Fixed the bash tool's stat builtin failing on native Windows with stat: unsupported on this platform (exit 1) for every invocation. The vendored uu-stat now ships a Windows-native backend that maps the GNU format directives onto std::fs::Metadata, the windows_by_handle metadata extensions (inode, hard-link count, and device via GetFileInformationByHandle), and the Win32 volume APIs for --file-system mode; Unix behavior is unchanged (#6723).
Fixed auto-retry wedging the session after an assistant-tail removal miss: when a context rebuild recreated the failed turn's message object, the identity-keyed cleanup logged assistant removal missed but the retry still scheduled continue(), which rejected the terminal assistant error message locally (Cannot continue from message role: assistant) before any provider request — auto_retry_end never fired, the TUI kept showing retry progress, and the in-flight prompt() hung until a manual follow-up. The retry path now strips a still-failed assistant tail positionally after the backoff, and a continuation that still fails locally closes the retry saga with a failed auto_retry_end (#5382).
Fixed native Anthropic web-search history being recursively truncated during session persistence or retained under a different user turn, preserving opaque replay bytes across reload and stripping them on reparent (#6703).
Fixed malformed or temporarily unreadable config.yml files being treated as empty settings and then overwritten by the next setting change, which could permanently erase broker tokens, model roles, and provider configuration. Invalid YAML is now moved to a timestamped .broken-* backup, read failures abort without touching the source, pending changes remain retryable with the last successfully loaded settings, atomic writes preserve symlink targets and handle Windows EPERM replacement, concurrent startup failures are fully observed and quarantine races fail closed, and omp config set/reset waits for persistence before reporting success.
Fixed mounted MCP tools being hard to invoke when server or plugin guidance names their original calls: sessions now include one bounded, exact original-name-to-xd:// route map for every live mounted MCP tool—including servers without initialize instructions—and refresh it as catalogs change without disabling schema virtualization.
Fixed inspect_image blocking indefinitely when the vision-model API stalls by combining the caller's abort signal with an AbortSignal.timeout() and surfacing a distinct timeout ToolError (separate from user-triggered abort) (#4165).
Fixed MiMo models using hashline edit mode by default despite needing the same replace-mode fallback as Kimi. (#3772)
Fixed omp refusing to start on Windows when no bash.exe is discoverable — most visibly with scoop-installed Git, whose manifest shims sh.exe/git.exe but never bash.exe, so PATH lookup missed it. Startup threw No bash shell found while merely building the bash tool description, even though bash tool commands always execute in the embedded brush-core shell and need no host bash. Shell discovery now also checks GIT_INSTALL_ROOT, scoop and per-user Git for Windows install roots, and sh.exe on PATH, then falls back to cmd.exe for the spawn-only paths (interactive PTY, ACP client terminals) instead of failing; the cmd fallback is never used to wrap user-shell commands — brush runs the POSIX line directly.
Added a selectable voice setting for /live realtime sessions (#6566).

[17.1.4] - 2026-07-26

Added

omp usage now surfaces auto-disabled credentials as red ✗ tombstone rows (identity, how long ago, the shortened upstream cause — e.g. Refresh token expired — and a re-login hint), including a provider section when no active credential remains. User-driven tombstones (replaced by newer credential, deleted by user) and API-key rows stay hidden. Requires a broker with GET /v1/credentials/disabled; older brokers degrade to no tombstone rows.
omp usage warns about Anthropic's ~30-day OAuth grant lifetime: accounts whose interactive login (authorizedAt) is within a week of the deadline get a yellow ⚠ re-login within <time> line, and past-deadline accounts a red one. Grants die server-side exactly ~30 days after login regardless of refresh rotation, so this is the only warning before the broker auto-disables the row.

Changed

Enabled Computer Use sessions now state the desktop-routing contract in the compact system prompt, retain their controller across model switches, expose effective native/function routing through /computer status, and emit structured lifecycle diagnostics without logging captured content.

Fixed

Fixed dragging an image whose path contains unescaped spaces (e.g. macOS screenshot names like Screenshot 2026-07-24 at 1.55.12 PM.png) into the terminal — the bracketed-paste image extraction route now has the same whole-text-as-path fallback as the clipboard keybind route, so both routes share identical detection and attach the image instead of inserting the raw path as literal text (#6578). The shared fallback only claims payloads that hold a single path: one carrying a second absolute-path anchor after unescaped whitespace (/tmp/a.png /tmp/b shot.png — dragging two files at once when either name has spaces) now pastes as text on both routes instead of being fused into one unresolvable path, which on the clipboard route previously attached nothing and swallowed the text behind an "Image not found" status.
Fixed transient reasonless request aborts that arrived after a tool call finished streaming ending the turn instead of entering recovery, which left edit calls and task subagents dead until the user manually resumed. The session now continues from the synthetic unexecuted tool result under the normal retry policy without replaying completed side effects (#6668).
Fixed prewalk silently dropping a same-model hand-off that only lowers the thinking level: the arm/switch guard compared model identity alone and discarded the resolved thinkingLevel, so a legal effort-downgrade target (e.g. prewalk: "@task" resolving to the same model at a cheaper effort) never applied and the session paid the plan/continue nudges for nothing. Prewalk now compares (provider, id, effective thinking level), applies effort-only hand-offs, and emits a notice on a genuine no-op instead of returning silently (#6659).
Fixed @czottmann/pi-automode failing legacy extension validation because the pi-ai compatibility shim omitted clampThinkingLevel, then failing every classified tool call because ctx.modelRegistry omitted getApiKeyAndHeaders. (#6648)
Fixed hide-secrets placeholders conflicting with hashline edit headers by replacing hash-delimited tokens with the unambiguous $$HASH$$ format (#6631).
Fixed the advisor silently swallowing its own quarantined turns: when an advisor called an ungranted tool (e.g. bash) its whole turn was discarded before dispatch, so its advice never reached the primary agent and the failure surfaced only in advisor diagnostics — every other non-recovering failure branch notifies the host UI, but quarantine re-primed silently with no bound. A persistently-quarantining advisor now surfaces a notifyFailure warning in the main UI (deduped, cleared on the next successful turn) and stops the unbounded silent re-prime loop (#6661).
Fixed the Docker natives-builder stage failing to build releases ≥ 17.1.1: the native audio stack added bindgen (miniaudio needs libclang) and a bundled-opus CMake build (needs cmake + make), none of which were installed in the slim builder image.
Fixed a configured modelRoles.default naming an extension-registered model (listed in enabledModels) silently running on a different in-scope provider's model. The startup model scope is resolved before extensions call registerProvider(), so the default role dropped out of scope and buildSessionOptions pinned options.model to the first scoped model — which marked the model "explicit" and suppressed the post-extension default-role re-resolution. A configured default that can't be found in the startup scope is now deferred so it re-resolves against the fully registered, still enabledModels-scoped catalog once extensions load (#6694).
Fixed Parakeet speech-to-text failing to load sherpa-onnx-node from Windows source workspaces when Bun installed the wrapper under packages/coding-agent/node_modules but hoisted its native platform package to the repository root (#6690).
Fixed omp usage duplicating org-less legacy accounts as "no usage data" rows whenever any sibling report carried an organization (mixed pools of pre-org-capture rows and fresh org-scoped logins): an org-less account is now covered by its own org-less report, while org-attributed sibling reports still never count as its coverage.
omp usage revalidates the broker credential snapshot before rendering: live usage reports were previously paired with a disk-cached account list up to an hour old, so a just-completed re-login (org-less row upserted to org-scoped) rendered as a phantom duplicate until the cache expired.
Fixed Advisor requests reaching Anthropic-compatible endpoints without a provider-facing session identity: the separately constructed advisor Agent never had a metadata resolver installed, so its outbound requests omitted the metadata.user_id session id that the main and subagent agents carry. Each advisor now emits its own advisorProviderSessionId via metadata.user_id, resolved live so a token refresh surfaces the current account_uuid, giving Main, subagent, and Advisor traffic distinct, stable session ids for proxy routing and attribution (#6625).
Todo progress now stays in sync when using Cursor models: the Cursor exec bridge mirrors the provider's server-owned todo list into session state, refreshes the interactive todo panel, and persists each snapshot to the session branch so the list survives reloads, rewinds, compaction, and session switches. Existing phase grouping is preserved for tasks the session already knows. Previously the list was in-memory only and the panel stayed stale, because Cursor resolves the todo tool remotely and never emits the local todo tool result that both paths key off.
Cursor todo calls the server refuses or rejects no longer leave the todo card spinning: the bridge settles every completed native todo call, not just the ones carrying a list. Local phases and the session branch are left untouched in that case, and the settling result deliberately carries no details.phases — echoing the current list back would let a call that changed nothing overwrite live panel state.
Fixed the todo renderer emitting mirrored label text verbatim. A Cursor snapshot carries provider-authored task content, phase names, and summary text, and the renderer interpolated all of it straight into terminal output, so a label holding ANSI/C0 sequences rewrote the terminal every time the list rendered or replayed. Every display path now goes through one sanitize-and-flatten-tabs helper — task labels, blocker notes, phase headers, the zero-task fallback, and the streaming call preview — while the raw values stay untouched as the lookup keys they are.
Fixed a server-resolved Cursor todo card animating for the rest of the session when the server packed the call's start and completion into one HTTP/2 chunk. The bridge's tool_execution_end is a synchronous callback fired mid-parse, while the streamed toolcall_start that creates the visible card is queued on the event stream and delivered a microtask later — the interactive controller handled the completion first, found no pending card, and dropped it, leaving the card that appeared moments later with nothing to settle it. An early completion is now held and attached the moment the streamed block creates its card, settling it without repeating the panel refresh or failure warning that already fired on first arrival. Card creation from cumulative message_update frames is also guarded by the turn's timeline map, so a call settled mid-stream can no longer be recreated as a second, permanently pending card by the next update re-listing the same block.
Cursor todo failures no longer render unsanitized provider text into the status line. The bridge forwards the server's error string verbatim, so an ANSI escape or other C0/C1 control reached the terminal intact and could repaint outside the row, tabs punched holes in the single-line warning, and a long message overflowed it. The detail is now stripped of control sequences, collapsed, and truncated at the render boundary; the persisted result keeps the full-fidelity error for the transcript.
Fixed disabling the Advisor from /settings updating the persisted setting without stopping the live Advisor runtime until the session restarted: SelectorController.handleSettingChange had no case for advisor.enabled, unlike other session-managed toggles (autoCompact, steeringMode, ...), so the change never reached session.setAdvisorEnabled.
Fixed bash.patterns deny/prompt rules matching only against the whole command string, so a dangerous command in any non-leading position of a compound line (e.g. cd /tmp && rm -rf /tmp/x, sleep 1 & rm -rf /tmp/x) silently bypassed a deny rule and, under approvalMode: yolo, executed with no prompt. deny/prompt rules now also match each command segment, split with a shell-aware tokenizer that honors every command boundary (&&, ||, ;, |, single &, subshells, newlines) and quoting; allow rules still require the whole command to match and never apply to compound lines (#6695).
Fixed omp config list printing credential settings in plain text. auth.broker.token, searxng.token, searxng.basicPassword and dev.autoqaPush.token were disclosed in both the human and --json output of a command that dumps every value without anyone asking for a specific credential. Credentials are now marked in the schema with a top-level credential flag, which also covers settings that have no settings-panel entry and so cannot use ui.secret. Human output shows dots; JSON omits value and marks the entry redacted rather than substituting a placeholder a consumer could write back. omp config get <path> is unchanged, since that is an explicit request for one value. The settings panel now derives masking from the same flag, so a credential cannot render as plain text on one surface and dots on the other. Only a credential that is actually set is redacted, so a fresh configuration still reports unset credentials as unset rather than implying every one of them is configured.
Fixed /new, /drop, /fork, and /move crashing or doing unnecessary work when invoked during vibe mode; interactive session transitions now show the existing exit-vibe warning and leave the session unchanged, and reset loops disable themselves instead of resubmitting into that unchanged session (#6607).
Fixed legacy pi extensions failing extension validation when importing estimateTokens from @earendil-works/pi-coding-agent (aliased to the legacy shim). Legacy pi re-exported estimateTokens from its coding-agent package root; in omp it lives in @oh-my-pi/pi-agent-core/compaction and the coding-agent barrel does not forward it, so the shim's export * from "../index" left it off the surface and a named import threw Bun's static "Export named 'estimateTokens' not found" error (e.g. omp plugin install pi-blackhole). The shim now re-exports it (#6583).
Fixed plan approval presenting a completed plan instead of the newest draft when the submitted title did not match the draft filename (#6569).
Fixed omp auth-gateway serve advertising only the compiled-in bundled catalog, so every model omp reaches through provider discovery (e.g. ids released after the build date) was invisible on /v1/models and returned Unknown model through /v1/chat/completions even though the same broker credential answered it in the TUI. The gateway now sources its catalog from ModelRegistry — the same component the TUI/CLI use (bundled + cached + discovered) — keeping the credential scoping and qualified/bare-id registration, and rebuilds it periodically so a long-lived serve tracks newly discovered models without a restart (#6615).
Fixed screenshot-relative pointer actions missing their visible targets when image transports that cannot preserve original detail silently downscaled a large computer screenshot; affected transports now establish the native coordinate frame below the verified image-resize threshold without changing the public capture defaults for other models (#6596 by @wolfiesch).
Corrected Windows shell resolution errors to identify the active global, project, overlay, or runtime source for shellPath, including profile and custom configuration directories, instead of directing every user to the retired settings.json file (#6579).
Fixed debug (js-debug/pwa-node) stateful commands misrouting after launch: a lazily-attached [worker N] child session (or the threadless root launcher) would steal the active-session focus from the stopped script child, so threads listed only the worker thread, post-launch breakpoints read back as pending/unbound, and there was no way to step/continue/evaluate the script's thread. Focus now follows stops rather than registrations, and threads aggregates every live thread across the session tree (#6663).
Fixed a turn-ending provider error being truncated to 8 lines in the transcript with no way to reveal the rest: AssistantMessageComponent now implements setExpanded, so Ctrl+O (tool-output expansion) reveals the full error body and the collapsed view shows a … +N more lines (Ctrl+O to expand) hint (#6555).
Fixed direct binary updates trusting an executable that only reported the expected version. The updater now selects one exact asset from the tagged GitHub release, requires its published SHA-256 digest and size, and verifies both while streaming the download before installation. GitHub release metadata requests use GITHUB_TOKEN or GH_TOKEN when available, allowing users behind an exhausted anonymous rate limit to authenticate.
Documented that the non-PTY shell's bundled jq command is backed by jaq, including its null-indexing divergence and portable filter syntax (#6614).
Fixed omp://tools/task.md and omp://tools/eval.md drifting from the 17.1.3 runtime: task.md claimed subagents force-disable async.enabled/bash.autoBackground.enabled (both are inherited from the parent since 17.1.0) and omitted the task tool's effort parameter, and eval.md omitted the still-working eval agent(model=…) per-call model selector (#6594).
Fixed advisor retry amplification after transient Codex SSE socket closures by limiting each advisor-level try to one provider transport attempt.
Fixed omp update aborting with npm error EEXIST on standalone binary installs whose directory coincides with the global npm/bun bin dir (for example npm prefix -g set to ~/.local, which the installer also targets). The install-target resolver classified the binary as npm/bun-managed from directory containment alone, so npm install -g tried to replace a regular file its symlink step would clobber; it now treats a plain executable (not a symlink) in a package-manager bin dir as the standalone binary and self-updates it in place (#6527 by @am423).
Fixed Codex subscription and proxy models being sent the unsupported native { type: "computer" } declaration based only on model ID. They now receive the callable function-tool fallback, including after switching from native OpenAI Responses history, while explicit endpoint metadata can still opt into the GA contract. Explicit native Codex replays preserve computer_call/computer_call_output pairing, normal CLI startup keeps the native desktop worker graph lazy, and packaged workers re-enter the single CLI host without the computer module claiming non-computer selectors.
Fixed isolated JavaScript eval subprocesses letting the global fatal-rejection handler race the cell rejection interceptor. A floated promise rejection is now folded into the owning cell result without killing its reusable worker process.
Fixed /context counting hidden, explicit-only skills (hide: true / disable-model-invocation) in the Skills category and subtracting that inflated estimate from the first system-prompt block, which reported System prompt: 0 tokens and inflated Skills usage. Accounting now counts only the skills actually rendered into the system prompt — mirroring buildSystemPrompt's filter, so hidden skills and all skills when the read tool is unavailable contribute zero (#6498).
Fixed pi-sprite failing plugin validation because the legacy Pi compatibility shims omitted createExtensionRuntime and terminal capability/image-deletion helpers used by the extension (#6506).
Fixed Escape waiting for an in-flight session_stop extension handler to exhaust its timeout; abort now cancels the active stop pass without reporting a false timeout or applying stale continuation context (#6489).
Fixed the agent not resuming after re-answering a past ask from the session tree. Committing a new answer via /tree branched a fresh sibling toolResult and rebuilt context, but nothing ever continued the agent — unlike a live ask, whose continuation is intrinsic to the streaming run loop — so the model never consumed the new answer and the session sat idle until a manual prompt. navigateTree now reports the commit (askReanswerCommitted) and the interactive /tree handler resumes the agent via resumeAfterAskReanswer() after its transcript rebuild, so the resumed turn never renders against the stale pre-rebuild UI. Plain leaf moves and the read-only reopenAsk probe stay idle (#6483).
Fixed Ctrl+C and fatal shutdown entering an ExtensionExitError rejection loop while an extension or hook was still loading (#6488).

[17.1.3] - 2026-07-24

Fixed

Fixed the in-process find builtin's -exec/-execdir children inheriting the omp process's real stdout/stderr, so commands like find … -exec ls -ld {} \; spammed their output straight into the terminal (corrupting the TUI) instead of the tool's captured output, and bypassed shell redirects. Exec children now stream stdout/stderr back through the shell's scope streams (matching xargs/ifne) and run with the shell's exported environment (env_clear + scope snapshot) instead of the host process environment.
Fixed ast_edit erroring with "lang is required" — an argument that no longer exists in the tool schema — when paths resolved to files of multiple languages (e.g. a crate directory with .rs + .toml). Mixed-language paths now rewrite per file: each file is parsed in its own inferred language, patterns are compiled per language, and a pattern that doesn't parse in some matched language simply skips those files.
Fixed the retain tool's TUI renderer crashing when streaming arguments temporarily expose a non-array items value (#6528).
Fixed Edit and Write tools failing with Settings not initialized in isolated sessions by using each tool session's settings for generated-file guards, with safe schema defaults for standalone guards and inline-image rendering (#6549).

[17.1.2] - 2026-07-24

Added

Added in-process moreutils-style shell builtins to the bash tool's embedded shell: ts (timestamp lines; -i/-s elapsed modes, -m monotonic clock, -r relative rewriting of RFC3339/syslog timestamps, %.S/%.s/%.T subsecond extensions), sponge (soak stdin fully before atomically writing the target, so foo file | ... | sponge file works; -a appends), ifne (run a command only when stdin is non-empty; -n inverts and passes non-empty stdin through), isutf8 (streaming UTF-8 validation with line/char/byte diagnostics; -q, -l, -i), combine (boolean and/not/or/xor on the lines of two files, - for stdin), and errno (errno name/number/description lookup with -l list and -s search; unix only). Like the uutils-backed builtins, they run in-process against the command's own stdio, resolve paths against the shell working directory, honor cancellation, and are disabled by PI_DISABLE_UUTILS_BUILTINS.
The bash tool prompt now lists the available shell builtins (mkdir through jq, rm/mv/ln, and the moreutils set) so the model relies on them without existence checks; the line is dropped when PI_DISABLE_UUTILS_BUILTINS disables the builtins and omits unix-only errno on Windows.
Sessions using a broker-backed auth store now report each completed request's token usage and cost to the auth broker (batched, 10s cadence) so the broker can track actual token burn per client install
Added a per-spawn effort parameter to the task tool ("lo" | "med" | "hi"): each selector maps onto the resolved model's supported thinking range (lowest, middle, and highest level — whatever the model tops out at, e.g. high, xhigh, or max) and overrides the agent's default selector, including auto. Omitting effort keeps the existing automatic per-prompt thinking classification.
Added searxng.engines setting for the SearXNG web search provider: a comma-separated list of engine names or bang shortcuts (e.g. ddg, br, startpage) sent as the API's engines= parameter. Shortcuts are resolved to canonical engine names via the instance's /config endpoint (cached per endpoint; entries pass through verbatim if /config is unreachable). Bang syntax in queries (!ddg foo) continues to pass through to the instance, and external bangs (!!g) are now stripped client-side since SearXNG answers them with an HTTP redirect even for JSON requests.
Web search queries now understand Google-style directives on every provider: site:/-site: (plus domain:/host: aliases), after:/before:/since:/until: date bounds, inurl:/intitle:/intext:/allin*:, filetype:/ext:, lang:, quoted phrases (including smart quotes), +term, -/NOT exclusions, and OR/| groups. A shared parser (web/search/query.ts) structures the query once per request; each provider maps constraints onto native API filters where the upstream supports them (Perplexity domain/date/language filters on both the API-key and ask paths, Tavily include_domains/exclude_domains + start_date/end_date, Exa domain lists + published-date bounds on API and MCP paths, Anthropic allowed_domains/blocked_domains, xAI filters.allowed_domains/excluded_domains, Parallel source_policy, Brave absolute freshness ranges, Firecrawl tbs=cdr date ranges, Jina X-Site, SearXNG language) and otherwise re-emits only the operator syntax its engine parses (full Google syntax for Gemini grounding, OpenAI, Kagi, and the credential-free scrapers — with scraper-hostile path-site:/inurl: operators demoted to plain keywords; conservative subsets for DuckDuckGo, Mojeek, Kimi, Z.AI, TinyFish, and Synthetic). The pipeline then applies a lenient post-filter to every response: constraints the engine ignored are enforced on the returned sources, and any constraint dimension that would eliminate every result is relaxed and reported to the model (Note: no results matched \site:...`; the constraint was relaxed`) instead of returning nothing. Directive-free queries are passed through byte-identical everywhere.
Eval and browser run code previews are now prettified for display: minified/compact agent-written cells are expanded by conservative streaming formatters (Python, JavaScript, Ruby, Julia) that indent block structure, split statements, normalize operator spacing (JS), and expand object literals wider than 100 columns onto one property per line — purely for rendering, the executed source is untouched. The formatters handle in-flight prefixes (unterminated strings/comments/blocks) without inventing closers, and layout decisions are prefix-stable so already-rendered lines never reflow while a call streams.

Changed

Large pastes saved via the large-paste menu now insert local://paste-N.md references (previously local://attachment-N), so the saved paste carries a markdown extension and a clearer name.
Raw SSE debug capture now trims over-budget events smartly instead of chopping off the tail: tool definitions inside data: payloads are compacted first (name kept, schema/description elided — often enough to keep the whole payload as valid JSON), and anything still over the 64k cap keeps its head and tail with a : omp-debug-elided chars=N comment marking the removed middle, so trailing fields like usage stay visible.
The web_search tool prompt now tells the model to never search for content that is programmatically accessible or has a known URL (GitHub, known arXiv papers, Wikipedia pages, official docs) and to read the URL directly instead.

Fixed

Fixed todo calls that omit op hard-failing validation ("op must be operation to apply (was missing)"): the tool now validates leniently and infers the op for unambiguous payloads (list → init, phase+items → append, bare items on an empty list → init); op stays required in the schema, and ambiguous op-less calls surface the schema error as a retryable tool error.
Fixed credential-free web search engines (SearXNG, DuckDuckGo, Google, Startpage, Ecosia, Mojeek, and the Public Web fan-out) returning zero results for queries with site: paths (e.g. site:github.com/owner/repo) or inurl: operators: scraper engines only match site: against a bare domain and DuckDuckGo ignores inurl: entirely, so such queries silently emptied the result set and fell through to the next provider in the chain. A shared formatScraperQuery formatter now structurally demotes path-carrying site: and all inurl: values to plain search terms (covering OR-grouped and quoted directives) while preserving bare-domain site: filters, negated operators, and each engine's supported syntax; the pipeline post-filter still enforces the demoted constraints on returned sources.
Fixed ast_edit previews reading like applied edits to the model: the ⟨proposed⟩ badge was TUI-only, so the model-visible result (hashline header + -/+ rows, identical to applied edit output) carried no staged-proposal signal. The preview result now leads with a "Staged as a proposal — files NOT modified yet" notice naming xd://resolve/xd://reject, the injected resolve reminder names the source tool, and the ast_edit tool prompt documents the two-phase flow.
Fixed the hub launch ps/list response burying the active process behind every exited one and growing without bound in long-lived projects: the broker now lists non-terminal daemons first (oldest to newest) and caps exited/failed history at the 10 most recently exited, so the active launch is immediately visible and the response stays bounded. Broker recovery also preserves each already-terminal daemon's real exit time instead of overwriting it with the restart timestamp, so the history cap keeps the genuinely most-recently-exited processes after an idle-broker restart (#6517).
Fixed a tool call rendering twice in the transcript. A mid-stream rebuildChatFromMessages (from /shake, auto-compaction, or a settings toggle) preserves the live pendingTools components across the clear+replay, but that preservation assumed every pending-tool component was still dangling. When a tool's result had already landed in the session entries while its component still lingered in pendingTools, the replay reconstructed the completed block and the preserved live component was re-appended, so the same block appeared twice; resolved components are now dropped from preservation and owned by the replay (#6516).
Fixed --model default (and other bare role names) resolving to the bundled cursor/default catalog model instead of the configured modelRoles.default role. resolveCliModel's exact-match phase ran its unauthenticated catalog fallback before role resolution, so a bundled id colliding with a reserved role name shadowed a configured, runnable role — failing with No API key found for cursor on machines without Cursor credentials. The catalog fallback is now deferred so an authenticated exact model still wins, a configured role beats an unauthenticated catalog-only id, and the catalog id is still recovered when no role matches (#6508).

Removed

Removed the model parameter from task and agent(): explicit per-spawn model selectors and fallback chains are no longer supported; spawns always use the agent's configured model

[17.1.1] - 2026-07-24

Added

Added the /session pin subcommand and account picker to pin provider OAuth accounts for the current session
Added the disabled-by-default computer essential tool with configurable enablement, backend, display, and maximum width/height settings. Native desktop execution runs through a DesktopSession worker; observation uses read approval, input uses exec approval, and provider checks always prompt and fail closed.
Added the /computer slash command (on/off/status/toggle) to enable or disable the computer tool for the current session without persisting settings.
Exposed computer to models without native OpenAI computer-use support as a regular function tool with a typed GA action schema; the same native desktop backend and approval policy apply on both paths.
Hardened computer action ingress: action-specific fields, modifier/key arrays, coordinates, drag points, and scroll deltas fail closed before native input; numeric fields must be signed 32-bit integers and coordinates must be non-negative.

Changed

Replaced Chromium-backed /live media and external speech recorder/player subprocesses with the cross-platform native microphone, speaker, Opus, and WebRTC stack from @oh-my-pi/pi-natives.

Fixed

Fixed live-call attestation depending on the ChatGPT desktop app being installed: generateLiveAttestation now mints DeviceCheck tokens in-process through the @oh-my-pi/pi-natives deviceCheckGenerateToken binding instead of probing /Applications for the app's devicecheck.node addon, so the x-oai-attestation header works on hosts without the desktop app and drops the createRequire addon probing; the attestation provider is now wired up to @oh-my-pi/pi-ai for ChatGPT-OAuth Codex requests.
Fixed xd:// device execution failures rendering as write errors instead of using the mounted tool's own error renderer.
Fixed custom tools without bespoke renderers losing the default state-tinted card when mounted under xd://; dispatched calls now keep their label, arguments, status, output preview, and expansion affordance instead of dumping a bare result line into the transcript.
Fixed the clipboard image-paste keybind mangling copied URL text into a bogus path error on macOS (e.g. Image not found at /https/::i.can.ac:CE4Ek3.png for a copied https://i.can.ac/CE4Ek3.png). AppleScript's the clipboard as «class furl» coerces plain text into a file URL by treating the string as an HFS path (:↔/ swap), so readMacFileUrlsFromClipboard returned a garbage path that dead-ended in handleImagePathPaste instead of falling through to the text paste. The script now bails early via clipboard info for «class furl» unless the pasteboard actually carries a public.file-url representation, so URL/text clipboards paste as text.
Fixed spilled tool-output artifact descriptors leaking on error/abort paths. OutputSink.dump() was the only path that closed the spill Bun.FileSink, but the bash and Python executors re-throw on failure and their finally blocks never closed the sink, so a large-output command that errored leaked the artifact descriptor until an unrelated read (e.g. a SKILL.md load) hit EMFILE. OutputSink now exposes an idempotent dispose() that closes the sink exactly once, wired into every executor's finally (#6463).
Fixed the first submitted prompt stalling while the local tiny-title worker started: the interactive submit handler now paints the pending user row before starting title generation, and startup prewarms an idle, unref'd worker so the first submit reuses a live subprocess instead of paying spawn latency ahead of the first frame (#6462).
Fixed legacy Pi extensions failing validation when importing the upstream keyText keybinding helper (#6470).

[17.1.0] - 2026-07-24

Breaking Changes

Replaced the providers.webSearch and providers.image single-preference configuration options with providers.webSearchOrder and providers.imageOrder priority lists. Existing configurations migrate automatically on startup.

Added

Added dynamic multi-root workspace context support, allowing users to manage multiple workspace directories mid-session via /add-dir, /remove-dir, and /dirs slash commands, or seed them at launch using the --add-dir CLI flag.
Added /live, a Codex-authenticated real-time voice interface that streams microphone audio over WebRTC and routes coding tasks through the active agent session.
Added opt-in usage-aware model fallback for rationed coding plans, including a /usage command to view live quantitative usage data and automatic fallback chain traversal.
Added error.notify configuration to allow failed model turns to trigger distinct terminal or desktop notifications.
Added auto-following light and dark themes to HTML session exports, with a /export --themes option to bundle selected TUI themes.
Added owner-routed asynchronous job delivery, ensuring background bash and task results are injected directly into the owning subagent or agent session rather than the top-level session.
Added background-on-steer capability for auto-backgrounded bash commands, allowing incoming user or peer messages to immediately background running commands.
Added friendlyName support for hidden secrets, allowing model-visible placeholders to carry sanitized semantic labels, hashes, and case hints.
Added support for Jujutsu (jj) repositories in the statusline git segment, displaying the nearest bookmark or change ID and retrieving working-copy change counts.
Added block and unblock operations for tasks, introducing a blocked status for tasks waiting on external input to exclude them from incomplete-todo reminders.
Added a toggle-list editor in /settings for managing array-of-enum settings like search and image provider orders.
Added models.yml Bedrock Converse prompt-cache capability overrides for bundled and opaque inference profiles.
Added getServiceTiers() and setServiceTier() extension APIs to read and modify the live per-family service tier for session requests.
Added opt-in omp bench --cache for independent cold/warm prompt-cache benchmarking with stable-prefix controls.
Added tools.xdevDocs prompt-doc modes and the tools.xdevInlineDevices glob allowlist to control which mounted device documentation is inlined into the system prompt.
Added the opt-in read.renderMarkdown setting for formatted Markdown read previews.

Changed

Updated subagent behavior to inherit async.enabled and bash.autoBackground.enabled from parent sessions, and refined subagent run completion to wait for background jobs to settle.
Added ordered bash.patterns command approval rules to allow, prompt, or deny bash commands by pattern.
Updated Markdown file handling so all Markdown flavors (.markdown, .mdx, .mdc, etc.) respect the read.summarize.prose setting.
Upgraded xAI web search to use grok-4.5 at low reasoning effort instead of grok-4.3.
Improved search provider resilience by cascading and falling back through other configured search providers when the preferred provider fails.
Extended the bash tool's direnv and devenv auto-loading to all backends (including the ACP client terminal and interactive PTY) while honoring direnv's local allow list.

Fixed

Fixed a path traversal vulnerability in blob reference resolution by rejecting non-canonical hashes in parseBlobRef.
Fixed multiple edge cases in the secret obfuscation and redaction engine, including handling of context-sensitive regexes, placeholder key requirements in unwritable directories, friendly-name forgery vulnerabilities, and regex match boundaries straddling existing placeholders.
Fixed a first-use race condition in ArtifactManager where concurrent callers could allocate duplicate artifact IDs.
Fixed Vibe-mode session stability, resolving issues with workers disappearing across restarts, hanging during teardown, clobbering target tools during session switches, and resolving against incorrect models.
Fixed concurrent MCP configuration mutations losing updates by serializing read-modify-write operations under a per-file lock with atomic writes.
Fixed legacy extensions failing to load on npm/source-link installs due to transitive CommonJS dependency graph clobbering.
Fixed omp auth-gateway commands bypassing the process-scoped OAuth account pool configured via environment variables.
Fixed terminal transcript rendering issues where displaceable snapshots (like waiting polls and todo lists) spammed native scrollback.
Fixed the terminal title to reflect the active agent run state (working, waiting, or blocked) when tui.titleState is enabled.
Fixed the browser tool's open action ignoring timeouts during browser acquisition and leaking orphaned browser instances.
Fixed the write tool silently creating empty files when a read-tool selector was mis-dispatched as a write.
Fixed snapcompact archiving reproducing assistant reasoning (¶think: sections) into replayed frames for Anthropic-dialect models.
Fixed Linux socket-mode DAP launches hanging indefinitely on connection failures.
Fixed Plan Review annotations being discarded on dismissal and limited to headings.
Fixed Assistant-mode TTS playback aborting prematurely when an agent continued after a tool call.
Fixed absolute usage amounts rendering inconsistently across CLI, TUI, and ACP output surfaces.
Fixed MCP sessions dropping tools from servers that finished connecting after the initial startup window.

[17.0.9] - 2026-07-23

Added

Added per-call model selection to the task tool, including per-item batch selectors, fallback chains, and explicit reasoning suffixes.
Added Firecrawl keyless mode: explicitly selecting firecrawl as the web-search provider now works without FIRECRAWL_API_KEY by calling the Firecrawl REST API without an Authorization header; the automatic provider chain remains credential-gated (#4332).
Added mcp.renderMarkdownResults (enabled by default): non-JSON MCP text results render as Markdown in the terminal transcript; set it to false to keep raw text.

Changed

Adjusted retry fallback handling to recognize discovery-only and runtime extension providers, preventing spurious unknown-provider warnings.
Restored Auto QA's ask-the-user default: dev.autoqa defaults to true again, so the first xd://report_issue write pops the consent dialog instead of the feature being silently off. Denying consent (or dev.autoqa: false / PI_AUTO_QA=0) fully disables prompt injection; an explicitly configured dev.autoqa: true overrides a past denial. Also restored the #1224 guarantee lost in the xd:// device consolidation: the grievance row is inserted only after consent resolves to granted (or PI_AUTO_QA_PUSH=1), so nothing touches the local database while consent is unset or denied.

Fixed

Fixed Auto QA grievance recording silently dropping every report since the xd:// device consolidation: openAutoQaDb treated the database file path (~/.omp/autoqa.db) as a directory and tried to open autoqa.db/autoqa.db inside it, which fails on legacy installs (the flat file blocks the directory) and fresh ones alike (SQLite does not create parent directories). Also restored the busy_timeout pragma dropped in the same refactor (#2421). Renamed getAutoQaDbDir to getAutoQaDbPath to match what it returns.
Fixed the setup wizard hiding the selected row on short terminals (e.g. 24x80): the provider sign-in, theme, and web-search lists now fit their windows to the visible height, and decorative chrome (sign-in hint, theme mock preview) yields to the list when space is tight.
Fixed restored sessions replaying terminal aborted or errored assistant turns, which could repeatedly fail continuation from an assistant role; /retry now consults the persisted transcript so the failed turn remains retryable without re-entering provider context.
Fixed get_available_models and set_model RPCs racing background model discovery on cold start by awaiting the in-flight refresh before reading the registry. RPC/ACP clients that query the catalog or select a model immediately after session ready previously saw only statically-bundled models until discovery completed seconds later.
Fixed deferred --model <provider>/<pattern> CLI resolution failing on cold start with "Model not found" when the selector pointed at a discovery-backed provider (proxy / ollama / lm-studio / llama.cpp / litellm). The deferred retry now runs a cache-aware discovery pass before resolving, mirroring the default-role fallback's cold-cache race fix (issues #6114, #6162).
Fixed MCP tool calls that return a WWW-Authenticate challenge by preserving the structured metadata, completing the configured OAuth flow, and retrying the call once on the refreshed connection.
Fixed the Hindsight API token setting being absent from the Memory tab, so authenticated servers can be configured entirely in the TUI.
Fixed aborted-task follow-up hints pointing at history:// transcripts that cannot resolve: the hint now reports the transcript as unavailable when the agent ref retains no session file, while still-resumable agents keep their hub resume hint.
Fixed compiled binaries failing to load legacy Pi extensions with minified imports, pi-ai/compat, or transitive runtime dependencies. The compatibility loader now follows compact static imports, resolves transitive on-disk ESM imports and CommonJS requires with package conditions, and restores the legacy copyToClipboard and decodeKittyPrintable root exports used by pi-vimmode and pi-web-access.
Fixed a budget-aborted keep-alive subagent becoming an unkillable registration with no hub-level stop. A subagent force-stopped for exceeding its soft request budget is kept resumable (status idle, adopted by the lifecycle) so its context can be salvaged, but its async job row settles and is reaped after ~5 min — after which hub cancel <id> could only report Background job not found because it consulted the job manager alone. hub cancel now falls through to the agent registration: for an id the caller spawned that has no live job, it aborts any in-flight turn, disposes the session, and drops the registration (the interactive Agent Hub x and collab kill already did this; the model-facing hub did not). Cross-agent kills stay impossible and Main/advisor refs are never targeted (#6315).
Fixed Agent Hub fallback rows hiding routing provenance and the resolved provider/model (#6316).
Reduced format-on-write latency by avoiding cold language-server startup when diagnostics are disabled.
Rewrote the /guided-goal interviewer rubric around loop-engineering: deterministic success criteria, verification commands, attempt caps, scope boundaries, and stop conditions. Ready objectives must use the five-section structured markdown form.
Added task.isolation.apply (default true) to choose whether successful isolated task runs automatically apply their changes to the parent checkout or retain patch/branch artifacts for later integration.
Added opt-in RPC protocol v2 negotiation with bounded, lossless chunking for stdout objects up to 64 MiB, plus stable cursor-based message pages for histories that should not travel as one response. Legacy JSONL clients remain on protocol v1, while the bundled TypeScript and Python RPC clients negotiate, reassemble, and drain message pages automatically.
Fixed protocol v2 chunked framing materializing the whole base64 transport in memory: near-limit logical frames (~63 MiB) peaked around 686 MB RSS and over-ceiling frames allocated the full payload buffer before rejection. Chunk lines are now produced lazily from a single serialization, the 64 MiB ceiling is checked before any full-payload allocation, and RPC stdout writes honor backpressure line by line.
Fixed the bundled TypeScript and Python RPC clients throwing when a get_messages_page cursor went stale mid-walk (e.g. a background bash appending a message between pages): the high-level getMessages() drains now discard partial pages and fall back to the legacy snapshot on both session_busy and stale_cursor, driven by a new machine-readable code field on RPC error responses. Direct page calls remain strict.

[17.0.8] - 2026-07-22

Added

Added a /tree re-answer option for past ask tool results, allowing users to re-open the picker with original questions and branch the new answer as a sibling while keeping the original branch reachable.
Added configurable Hindsight client request deadlines via hindsight.requestTimeoutMs, reflectTimeoutMs, recallTimeoutMs, and retainTimeoutMs settings (and matching HINDSIGHT_*_TIMEOUT_MS environment variables).
Added omp-linux-musl-x64 and omp-linux-musl-arm64 release binaries for Alpine and other musl-based Linux distributions, with automatic musl selection in the installer and self-updater.

Changed

Optimized edit-tool previews, diff components, and intra-line word highlighting to compute line and word diffs natively, reducing synchronous diff times by 2-10x on large inputs.
Updated diff generation and rendering components to rely exclusively on native UTF-16 diff bindings, removing isWellFormed() guards and JS fallback code paths.

Removed

Removed npm diff dependency.
Fixed an issue where Ctrl+V clipboard paste was ignored while API-key and other modal prompts had focus.
Fixed scripts/install.sh incorrectly installing an x86_64 build on Apple Silicon when running under Rosetta.
Fixed the model picker hiding Codex models available through secondary configured ChatGPT/Codex OAuth accounts by unioning catalogs across all stored accounts.
Fixed GitHub Copilot 1M-context models disappearing from the model picker on restart with a "Could not restore model" warning.
Fixed --model <role> startup selection skipping configured fallback chains when the primary model is unavailable.
Fixed global model role updates clobbering concurrent or external edits to config.yml by merging only the changed role instead of persisting a stale in-memory snapshot.
Fixed terminal provider errors on continuation turns after failed tool results silently ending runs without persisting the error diagnostics.
Fixed repeated OpenRouter Gemini stream closures consuming the full retry budget by limiting recovery attempts before surfacing the error.
Fixed Agent Hub performance freezes when opening large read-only Advisor transcripts by collapsing synthetic inputs into compact summary rows and rendering Markdown lazily on expansion.
Fixed /agents incorrectly showing prewalk as disabled for the bundled task agent when enabled by its runtime default.
Fixed hub start waiting for the full timeout when a launched process exited or became ready quickly.
Fixed tools.maxTimeout failing to clamp default tool timeouts when no explicit timeout was provided by the agent.
Fixed MCP argument-shaping parity between direct and subagent tool calls, ensuring strict servers do not reject proxied calls with unrecognized keys.
Fixed a crash in extensions like pi-mcp-adapter caused by the TypeBox compatibility shim omitting Type.Unsafe.
Fixed omp models hanging after output by properly clearing managed extension timers and shutting down sessions before returning.
Fixed HTML session exports causing browser call stack overflows when rendering deeply nested conversation trees.
Fixed task agents ending prematurely on connection errors instead of entering the auto-retry path.
Fixed a startup race condition where the default model role was incorrectly overridden by an unrelated provider's default on a cold cache.
Fixed unqualified --model startup selection preferring unauthenticated provider catalog entries over configured providers.
Fixed provider stream failures being invisible in the main log by logging a warning with error details when a turn ends in a provider error.
Fixed parallel todo done calls losing completions due to asynchronous session events overwriting newer tool states.
Fixed omp crashing when git is not installed or missing from the system PATH.
Fixed /changelog commands reporting no entries in standalone binaries by embedding the release history as a fallback.
Fixed isolated branch merge-backs rejecting committed agent edits when the parent branch had unrelated uncommitted changes in the same file.
Fixed the 30-second Hindsight client timeout aborting healthy reflect operations by applying dedicated, longer deadlines.
Fixed Mnemopi consolidation redundantly re-storing cumulative session transcripts after incremental auto-retain.
Fixed turn-ending Codex rate-limit errors being hidden behind the Plan Review overlay.
Fixed prewalked subagents continuing to display their starting model after switching to the target model.
Fixed the Escape key aborting an ongoing agent turn instead of stopping text-to-speech playback.
Fixed project system prompts shortening working directories to ~, which could cause models to generate incorrect absolute paths for tool calls.
Fixed the TUI /usage matrix misaligning multi-account columns across quota windows.
Fixed near-miss xd:// write targets silently creating filesystem paths instead of throwing a corrective URI error.
Fixed the task tool rejecting valid batch calls with misleading validation errors when batching is disabled.
Fixed JS/TS debug launches timing out on WSL2 with mirrored networking by waiting for the adapter's listening banner and handling transport closures immediately.
Fixed post-compaction transcript rebuilds blocking the main thread by reusing settled message components and layout caches.
Fixed the fullscreen Plan Review overlay remaining interactive and appearing frozen during slow asynchronous operations by locking input and showing a submitting indicator.
Fixed dynamic model discovery refreshes dropping provider-level compatibility overrides from models.yml.
Fixed a startup crash that locked users out of the app when prewalk.enabled was set but the prewalk hand-off target had no configured API key.
Fixed in-progress aborts awaiting session_stop extension handlers whose results would be discarded.
Fixed /retry reporting "Nothing to retry" after a stream stalled or aborted mid-tool-call.
Fixed locally consumed extension commands triggering automatic title generation and exposing their command text to the title model.

[17.0.7] - 2026-07-21

Fixed

Fixed Portkey/gateway custom models whose ids start with @ (e.g. @modal/GLM-5-2-FP8) being rewritten to unrelated bundled wire ids (e.g. glm-5-2), which caused 400 responses requiring x-portkey-config or x-portkey-provider.

[17.0.6] - 2026-07-20

Fixed failed plan-mode exits leaving the session on the restored execution model while plan mode remained active and silently changing ambient xd:// tool presentation; rollback now restores the plan model, thinking level, and exact top-level-versus-mounted tool partition so exit can be retried safely (#6013).

Added

Added native Warp CLI-agent events for rich session status, tool approvals, and completion notifications (#5592 by @metaphorics).
Added Codex (ChatGPT subscription) support to generate_image. The tool now resolves a connected openai-codex OAuth credential and drives OpenAI's hosted image_generation tool through the ChatGPT backend (chatgpt.com/backend-api/codex/responses, chatgpt-account-id header) independent of the active chat model — so image generation works on a ChatGPT/Codex subscription with no metered OPENAI_API_KEY, even when the active model is Claude/Gemini/etc. A new providers.image: "openai-codex" option forces it; auto now auto-detects a connected subscription (priority: active GPT image tool > Codex subscription > Antigravity > xAI > OpenRouter > Gemini), and the openai preference falls back to it when no OPENAI_API_KEY/active GPT model is present.
Added an optional provider parameter to generate_image (auto | openai | openai-codex | antigravity | xai | gemini | openrouter) that overrides the providers.image setting for a single request — so "generate this using gemini / codex / xai" routes per-call without changing the global setting. Absent → the providers.image setting applies, unchanged; the named provider uses the same resolution semantics (falls back to auto-detect if it has no credentials). File: tools/image-gen.ts (imageProviderSchema, findImageApiKey preference arg).
Added OpenTelemetry log and metric export alongside the existing trace export. When OTEL_EXPORTER_OTLP_LOGS_ENDPOINT (or the shared OTEL_EXPORTER_OTLP_ENDPOINT) is set, omp registers a LoggerProvider and forwards every centralized-logger event as an OTLP log record (severity + attributes + active span context for log↔trace correlation, min level via OTEL_LOG_LEVEL, plus a structured agent run completed summary event). When OTEL_EXPORTER_OTLP_METRICS_ENDPOINT (or the shared endpoint) is set, it registers a MeterProvider with a PeriodicExportingMetricReader and records GenAI-semconv gen_ai.client.token.usage plus pi.omp.agent.* counters/histograms (runs, steps, chat/tool calls by name+status+finish reason, latencies, estimated cost, errors) from the agent run summary and per-chat usage hooks. Each signal honors its own OTEL_*_EXPORTER=none kill switch, the global OTEL_SDK_DISABLED, and declines non-http/protobuf protocols independently (#4604).
retry.fallbackChains wildcards now support id-prefixed targets and keys: a chain entry like "openrouter/google/*" re-prefixes the failing model's bare id (google-antigravity/gemini-x → openrouter/google/gemini-x), a plain "provider/*" entry falling back from an aggregator strips the vendor prefix when the target provider only knows the bare id (openrouter/google/x → google-vertex/x), and an id-prefixed key ("openrouter/google/*") scopes a chain to that provider's ids under the prefix.
The session tree selector (/tree, /branch) now supports Shift+Enter to summarize-and-switch in one step: it forks from the selected entry with a branch summary, with no extra prompt and regardless of branchSummary.enabled. Plain Enter keeps the current behavior (direct switch by default; the summary prompt only when branchSummary.enabled is on). (#5152)
Added the turn's local timestamp (YYYY-MM-DD HH:mm:ss, down to the second) to the per-turn token-usage row shown under assistant messages when display.showTokenUsage is enabled.

Changed

Reduced concurrent subagent update CPU by reconstructing recent output only at progress emission boundaries. (#5936)
Fixed docs/advisor-watchdog.md overstating advisor delivery for a normal yield: the severity table listed concern as unconditionally interrupting and the prose promised a self-ended run could always be steered/resumed. Documented the #4840 terminal-answer exception (concern becomes a passive card while blocker normally steers, #5628) plus the plan-mode and deferred-ACP constraints that preserve would-be steers until the user resumes (#5913).
Fixed subagent (task) sessions triggering an unnecessary tiny-model session-title generation call on todo init. Subagent sessions in a non-interactive host (print/RPC/ACP/eval/SDK/CI) have no operator-visible title and now skip the replan title refresh; interactive hosts keep it, since a live subagent focused from the Agent Hub renders its session name in the status line (#5910).
Fixed tab.scroll() timing out after a queued wheel event waits too long for a busy renderer's acknowledgement (#5905).
Made the hashline seen-line guard opt-in and off by default (see edit.enforceSeenLines), and stopped excluding column-clipped (>512-char) lines from a snapshot's seen set: a displayed line now counts as seen even when its display was column-truncated, so single-line edits on long lines found via read/grep apply without a separate full-width re-read.
Changed the default astGrep.enabled setting to false
Batched todo operations with real tool calls to prevent solo todo turns and extra round trips
Changed every bundled TTSR rule to warn without interrupting generation.
Renamed the system prompt's project-context section wrapper from <context> to <repo-rules> to stop it colliding with the task tool's context parameter under in-band XML tool dialects: models were closing <parameter name="context"> with a stray </context> (primed by the ambient section tag) and emitting sibling params as bare <tasks> elements, so tasks arrived missing.
Rendered read xd:// calls in the compact grouped read view instead of a full tool-execution card; other internal URLs (skill://, agent://, …) still render full so their resolved content stays visible.

Fixed

Fixed the interactive !/!! shell shortcut spawning fish as a login shell (fish -l -c …), which fired status is-login blocks in user config (agent/keychain setup, PATH mutation) on every command. fish is now started with -i instead — interactive shells source the same config.fish/conf.d files (so aliases and functions from #1816 keep working) without login-shell side effects. zsh behavior (-l -i) is unchanged.
Fixed the status-line tok/s badge ignoring vibe worker sessions: in /vibe mode the director is often idle while workers stream, so the badge showed a stale/zero rate while parallel work was actively generating tokens. The rate now aggregates the main session's live tok/s with every live vibe worker's tok/s, and falls back to the main session's own cached rate when no workers are streaming.
Fixed plan.defaultOnStartup being ignored by headless omp -p sessions, so the initial prompt now runs in plan mode and the persisted session remains in plan mode for later review (#6017).
Fixed resuming an active plan session replacing its journal-restored model with the current modelRoles.plan setting (#6015).
Fixed --model <role> resolving a bare configured modelRoles key.
Browser tool selectors now accept bare snapshot refs (tab.click("e501"), @e501) everywhere aria-ref=e501 works — previously the tab-worker backend fell through to a CSS tag selector that could never match, burning the 2s zero-match watchdog with a misleading "matches no elements" hint. tab.select, tab.uploadFile, tab.press({ selector }), tab.screenshot({ selector }), and tab.drag now resolve refs too. Unknown/stale refs fail immediately with the "refresh refs" error.
tab.select no longer double-reports the previously selected option of a single <select>: the returned selection is read back after the full assignment pass instead of mid-loop.
Fixed transcript blocks being visibly duplicated during streaming (whole tool boxes and assistant paragraphs recommitted below their first copy on the terminal tape) by removing transcript committed-prefix compaction entirely. Dropping committed rows from the transcript's local frame shifted the frame under the engine's committed-prefix ledger, so the audit re-anchored and recommitted rows the tape already held. The transcript now always keeps its full local frame; committed finalized blocks still skip render() via the segment reuse bypass. Reverts the compaction half of #5930's fix (compose keeps the render bypass; the local frame is no longer truncated).
Fixed tmux pane growth during a live response blanking finalized chat history re-exposed from native scrollback by rebasing the in-place repaint's commit seam to the resized viewport tail (#6011).
Fixed classifier refusals (e.g. Anthropic stop_reason: "refusal") ending the turn with no visible error. Two independent regressions: (1) session events reached subscribers out of order when a turn's provider events landed in one tick — extension emits only await for event types with registered handlers, so the assistant message_end overtook its own message_start and the TUI skipped the error render entirely (no pinned banner, no inline Error: line); subscriber fan-out is now serialized in emission order. (2) Refusal turns are pruned from active context at settle (#3591), which also erased them from state.messages before prompt() resolved — print mode printed nothing and exited 0, and the task executor's getLastAssistantMessage() saw the previous turn. The pruned refusal is now retained until the next run starts, getLastAssistantMessage() reports it, and print mode reads the settled assistant via that accessor (exit 1 + refusal message on stderr). Additionally, #lastAssistantMessage is now set synchronously on message_end to prevent agent_end maintenance from reading a stale assistant turn when tool results and stops land in the same tick.
Fixed before_provider_request extension contexts exposing the primary session model for cross-provider Advisor requests instead of the request model (#6006).
Fixed isolated task subagents mutating the parent checkout and stacking parallel task branches. Copy isolation backends (reflink/apfs/btrfs/zfs/block-clone/rcopy) materialise the worktree by duplicating its .git verbatim; when the parent is a linked git worktree its .git is a pointer file, so the isolation shared the parent's HEAD/index/ref namespace and a task's git checkout/commit moved the parent's branch (and the rcopy git worktree add path leaked task branches into the shared namespace so a second task committed on top of the first). ensureIsolation now runs a new git.detachGitDir after isoStart: each isolation becomes a standalone repo with a frozen HEAD/refs/index snapshot that borrows the source object database through objects/info/alternates, so isolated git operations stay private, every task branch is parented on the requested base, and patch/branch capture (git fetch <merged>) still resolves objects. (#6003)
Fixed browser.run leaving Puppeteer request handlers and interception state with divergent lifetimes by removing run-scoped handlers, disabling interception, and releasing held requests on every exit path (#6004).
Fixed Codex web search to honor configured openai-codex base URLs, API keys, and headers without leaking official OAuth credentials to custom endpoints; explicitly selected providers now fail closed instead of silently falling back (#6001).
Fixed launch start waiting for a finite PTY command to exit when the broker's PID-file handoff was unavailable; PTY startup now reports the spawned PID directly and returns an authoritative running or exited snapshot promptly (#5996).
Fixed queued user steering aborting side-effecting hub start calls after the broker request may already have been written; only passive hub waits and followed logs are now interruptible (#5995).
Fixed JavaScript/TypeScript debugging by launching vscode-js-debug over TCP, handling recursive startDebugging child sessions, synchronizing breakpoints across the session tree, and terminating every child connection (#5984).
Fixed rich ask options showing preview content only for the highlighted choice; every option now renders its preview inline, with pageable long content and accurate configured paging and cancel hints (#5988 by @metaphorics).
Fixed legacy pi extensions failing extension validation when importing getPackageDir or getProjectDir from @earendil-works/pi-coding-agent (aliased to the legacy shim). The shim only re-exported getAgentDir; the two missing path helpers now resolve — getProjectDir from @oh-my-pi/pi-utils, and getPackageDir as a string-valued wrapper over omp's canonical package-root helper that falls back to the executable's directory inside bun --compile binaries (where the canonical helper returns undefined), matching pi's string contract. Extensions like @gotgenes/pi-permission-system install and load, and path.join(getPackageDir(), …) no longer crashes in the shipped binary (#5968).
Fixed headless print mode disposing the session before a final advisor review completed, which could drop the advisor transcript and usage (#5942).
Fixed capped zero-block assistant stops remaining in active/session history with the full failed-request usage, causing the next post-snapcompact continue to re-enter context maintenance at the same boundary; capped empty turns are now discarded and the failure names model switching or /shake images as recovery options (#5959).
Long sessions no longer re-run convertToLlm over settled history every turn. Conversion is memoized per message identity (plus the assistant interruptedNext neighbor flag): an exact re-convert of the same array reuses the outer Message[], append-only growth reuses the converted prefix via slice-on-growth, and the prune/shake/strip-images/prewalk-scrub rewrite seams invalidate the affected message before the next pass. On the llm-assembly bench (N=5000) steady/append convert and repeat estimate are all >10x faster with robust MAD-noise well under 20% (#5934).
Fixed /exit hanging on post-prompt work and stacking independent subsystem teardown delays by bounding the aborted-work drain, disposing independent session resources concurrently, and keeping long shutdown waits visible (#5932).
Fixed queued-message display updates being skipped by focused-editor keystroke frames by explicitly repainting the pending-message container (#5928).
Fixed RPC and RPC-UI startup crashes when an in-process extension claimed Bun's singleton stdin stream before the protocol reader (#5898).
Bash command timeouts now render with a warning (yellow) border instead of an error (red) border, reflecting that the timeout ran its course rather than the command failed. isError remains true on the result so the model still knows the command did not complete normally. The timedOut flag is now propagated from the bash executor to distinguish timeouts from user aborts.
Fixed Cursor responses streams stalling after an exec-channel tool completed without automatically recovering. The session now continues from the already-buffered tool result instead of replaying the side-effecting request. (#5790)
Fixed linked legacy pi extensions failing to load when they import DefaultPackageManager or linkedom: the coding-agent compatibility shim now enumerates OMP extension paths with plugin metadata, and extension-graph CommonJS modules load through synchronous default-export bridges with linkedom's bundled canvas fallback. (#5658)
Fixed the advisor retrying terminal, non-retriable provider failures (e.g. blocked prompts) three times before giving up; such failures now drop the bounded batch after a single attempt while transient failures keep the 3-attempt retry path (#5468).
Fixed reassigning the plan role model mid-planning not taking effect on the active planning turn; the change now applies at the next turn boundary instead of only the next plan-mode entry (#5657).
Added managed ctx.setInterval / ctx.setTimeout / ctx.clearTimer helpers on the extension context. Callbacks scheduled through them run with the same isolation as handler dispatch — a throw or rejected promise is logged and reported through the extension error channel instead of escaping as a process-fatal uncaughtException — and every outstanding timer is unref'd and cleared automatically on session_shutdown (#5664).
Fixed an extension's self-scheduled setInterval/setTimeout callback throwing being able to tear down the whole session. Such callbacks ran outside the handler-dispatch try/catch, surfaced as a process-level uncaughtException, and the global postmortem handler treated them as fatal; extension authors now have sanctioned managed timers (see Added), and the constraint is documented in docs/extensions.md / docs/skills/authoring-extensions.md (#5664).
Fixed /quit and /exit leaving failed or stalled automatic title-generation requests alive during session teardown; disposal now aborts both online provider and local tiny-model title requests (#5666).
Fixed startup.quiet still rendering the xdev: xd://: mounted … status line when MCP tools connect; quiet startup now suppresses only the user-visible mount notice while retaining the hidden model-facing device update (#5670).
Fixed command error in hub tool with a non-POSIX shell (#5682)
Fixed xdev-routed checkpoint and rewind writes not tracking checkpoint state and leaving rewinding results in rebuilt provider and session context.
Fixed the built-in advisor silently doing nothing when its model routes through the cursor provider: the advisor runs in its own Agent that was constructed without cursorExecHandlers, so on Cursor — where every tool executes server-side and is dispatched back through the client's exec handlers — each advisor tool call (including the MCP advise tool) came back toolNotFound/"tool not available" and no advice was ever routed. The advisor Agent now gets a Cursor exec bridge scoped to its own granted tool set, mirroring the primary agent. The bridge's native delete frame is gated so a read-only advisor cannot delete workspace files it was never granted a mutating tool for (#5680).
Fixed the fullscreen plan-review overlay staying visible until the approved execution turn finished, so after picking "Approve and keep context" (or any approve option) work proceeded underneath while the operator was stuck on the plan-review screen. The overlay is now hidden once execution begins — after the async transcript rebuild, before the blocking synthetic prompt is dispatched — instead of only after the whole turn returns (#5688).
Fixed MCP tools repeatedly unmounting and remounting mid-session when server names have overlapping sanitized prefixes (e.g. atlassian alongside an imported atlassian:atlassian), and stale tools remaining registered after disconnecting a server with special characters in its name.
Fixed the /usage show in use by this session: marker showing only the login email, so two same-email Anthropic credentials in different orgs (a Team seat and a personal Max plan) were indistinguishable. The marker now suffixes the active organization (email (OrgName)) via a shared formatActiveAccountLabel, matching the account list and login-success surfaces (#5691).
Fixed Windows stdio MCP servers launched through .cmd/.bat shims failing with Transport closed; the launch now builds a cmd.exe /d /e:ON /v:OFF /c command line escaped for cmd.exe's parser and spawned with windowsVerbatimArguments, so the resolved command path and arguments (including %VAR%, quotes, and shell metacharacters) reach the server intact and cannot inject commands (BatBadBut / CVE-2024-24576) (#5696).
Fixed the TUI usage panel truncating organization suffixes from same-email account labels even when the terminal has enough width (#5701).
Fixed a startup crash on Windows when running from a drive root (e.g. R:\): fs.realpath throws EISDIR there, but canonicalProjectDir in launch/presence.ts and launch/client.ts only recovered ENOENT. It now also falls back to path.resolve() on EISDIR (#5708 by @ve3xone).
Fixed unknown __omp_worker_* CLI selectors exiting 0 with empty output instead of erroring; an unrecognized worker-host selector now writes Error: unknown worker selector: … to stderr and exits nonzero, so a stale or mistyped selector can no longer look healthy to a parent process or install smoke path (#5712).
Fixed Plan Review capturing mouse drags as pointer events, preventing native terminal text selection (#5711).
Fixed orphaned TUI processes with revoked terminal descriptors remaining alive after a fatal error and amplifying shared log-rotation races into runaway memory, file-descriptor, swap, and disk consumption (#5716).
Fixed approved-plan execution looping through filesystem searches when a model rewrites the required local://<slug>-plan.md read as a same-basename working-directory path; a missing cwd-root alias now recovers the active session-local plan while preserving any real working-tree file (#5704).
Fixed Ask dialogs immediately accepting their highlighted single-select answer when they appear while the user is typing a space in the prompt editor (#5717).
Stopped post-compaction auto-continue from opening another primary turn after a terminal text answer with no queued work, and moved automatic auto-learn capture into an abortable private agent with only manage_skill and learn tools (#5715).
Fixed the write approval gate misclassifying xd:// device writes as exec when the mounted tool declared a function-valued (argument-dependent) approval: the gate discarded the function and never decoded the device JSON payload, so read/write device operations prompted in non-yolo modes their approval mode permits. It now parses valid object payloads and evaluates the mounted tool's normal approval decision, while malformed JSON, non-object payloads, and unknown devices still fall back to exec and prompt (#5727).
Fixed custom LSP servers such as roslyn-language-server crashing after initialization when they request unconfigured workspace/configuration sections; missing settings now receive the spec-required null instead of {} (#5745).
Fixed late user-initiated bash results and minimized-output artifacts being recorded in whichever session or branch was active when execution finished; bash now retains its originating transcript across new_session/switch_session/branch/tree navigation, and an intentionally dropped session stays deleted instead of being recreated by a straggling result (#5743).
Fixed Claude Code marketplace plugins with scope: "local" leaking skills, hooks, tools, commands, and MCP servers into unrelated projects (#5750).
Fixed headless omp -p waiting indefinitely after a completed turn when final mnemopi consolidation stalls; print mode now applies the same bounded consolidation shutdown budget as interactive exit and reaps the embed worker (#5753).
Fixed explicit-tool sessions bypassing xd:// presentation for ambient discoverable custom and MCP tools, which sent their schemas top-level and could exceed provider tool limits or trigger schema-compatibility errors.
Fixed providers.webSearch: kimi sending a Moonshot Open Platform credential (MOONSHOT_API_KEY / stored moonshot auth) to the Kimi Code search endpoint (api.kimi.com/coding/v1/search), which rejects it with 401 and silently falls back to another provider. Kimi web search now resolves and advertises Kimi Code credentials only — a Kimi Code Console key via KIMI_SEARCH_API_KEY / MOONSHOT_SEARCH_API_KEY or omp /login kimi-code (#5762).
Fixed extension/SDK/RPC registerTool demoting essential built-ins (read/write/bash/edit/glob/…) to discoverable when a re-registration omitted loadMode, which — with tools.xdev on — unmounted them from the top-level schema and broke the xd:// transport (read xd:///write xd://), leaving the model with no callable coding essentials. Omitted loadMode now defaults to "essential" for known essential built-in names at every adapter boundary, and read/write (the transport itself) are never mounted under xdev regardless of loadMode (#5764).
Fixed the advisor skipping the next real user instruction after auto-learn accepted and pruned a terminal empty assistant stop; advisor transcript cursors now detect rewritten prefixes and re-prime before slicing the next update (#5731).
Fixed built-in advisors retrying a quota- or rate-limited provider until becoming unavailable instead of applying the matching retry.fallbackChains model chain; advisor fallbacks now emit the same applied and succeeded lifecycle events as primary-agent fallbacks (#5740).
Made the model selector status messages use the role tag (SMOL, SLOW) instead of the display name (Fast, Thinking), matching the rest of the TUI and CLI/env role terminology (#5585).
Fixed Cursor models receiving only top-level tools by forwarding mounted xd:// devices, including user-configured MCP servers, through Cursor's request-context MCP catalog and execution bridge (#5650).
Fixed Windows bash crashes when a piped command times out while flushing output; explicit-timeout watchdogs now wait for bounded native teardown instead of returning mid-drain. (#5316)
Fixed a race where hub/IRC send and ensureLive could hand out or inject into a subagent session mid-park dispose: park now detaches and flips status to parked before session.dispose(), concurrent ensureLive cancels a pre-detach park or waits then revives, and IRC delivery always gates through ensureLive so receipts/unread counts stay truthful (#5633).
Migrated legacy dev.autoqa.consent → dev.autoqaConsent and todo.reminders.max → todo.remindersMax on settings load so pre-v17 nested or quoted-dotted config no longer leaves the parent path as an object (which made dev.autoqa truthy and enabled Auto QA, and discarded the reminder limit). Explicit new keys win, a separately configured parent boolean is preserved, an irrecoverable object parent falls back to the schema default, and only the new keys persist on save (#5632).
Fixed all keyboard input dying after the first keypress when a ~/.claude/tools (or .omp/tools) module attaches a stdin consumer at import time — e.g. an MCP StdioServerTransport constructed at module top level, or a bare process.stdin.resume(). The custom-tool/extension/hook/plugin loader guard now snapshots and restores process.stdin (listeners, paused state, raw mode) around third-party module evaluation, so a hijacked stdin reader can no longer starve the TUI's own listener (#5618).
Fixed the ask tool's "Other" custom-input dialog rendering the title, options, and hint one column to the right of the > input gutter; the prompt-style editor chrome now aligns to column 0 (#5313)
Fixed advisor context maintenance undercounting the provider context: the compaction decision now anchors on the advisor's provider-reported context usage (cached input + generated output) floored by a full local estimate that includes the advisor system prompt and tool schemas, rejects stale provider usage retained across advisor compaction, and recovers a provider overflow by clearing only the advisor's own context at the current primary cursor — retrying the bounded failing batch once against a fresh context without replaying old primary history and keeping later updates eligible (#5282)
Fixed RPC mode (--mode rpc) crashing the whole process with an uncaught SyntaxError: Failed to parse JSONL on any non-JSON stdin line. Malformed lines are now reported via a Failed to parse command error frame and the frame loop keeps running. (#5194)
Fixed the status line loop indicator to distinguish waiting, running, and paused states and show the remaining loop budget (#5832 by @wolfiesch).
Fixed single-model task agents ignoring an explicitly configured default retry fallback chain, which left subagents failed after their selected provider became unreachable instead of advancing to the configured fallback model.
Fixed the /extensions dashboard tab labeled "Agents (standard)" being confused with the /agents subagents feature — the .agent/.agents config-standard provider now presents as "Agent Dirs (.agent/.agents)" since it lists skills, rules, prompts, commands, and context/system files, never subagents (#5821).
Fixed non-raw read line selectors returning context outside the requested inclusive range (#5802).
Fixed LSP requests silently clamping explicit timeouts above 60 seconds by supporting documented budgets up to 300 seconds (#5804).
Clarified async task and hub guidance: inspecting a settled job consumes its automatic delivery, job IDs expire from process memory after roughly five minutes, and completion does not verify claimed artifacts (#5869).
Fixed vibe_wait TV-wall panels stacking duplicate frozen frames in native scrollback while two or more workers were live (#5777).
Fixed async task job rows omitting resolved subagent model and reasoning badges when task.showResolvedModelBadge is enabled. (#5060)
Fixed raw Puppeteer page/browser promises from crashing inline browser workers or killing dedicated workers when a target closed before the caller awaited the promise.
Fixed auto-compaction dead-ending in a warning loop ("Compaction freed too little context to make progress") when the single most-recent turn is itself over budget so prepareCompaction has nothing to summarize (findCutPoint never cuts inside a tool result). This !preparation short-circuit never ran the artifact-backed shake elide rescue that #3786 added to the post-maintenance guard, so snapcompact/context-full maintenance paused with no attempt to shrink the oversized tail. The dead-end now runs the same elide pass, re-prepares on the shrunken branch, and falls through to a normal compaction when the tail became summarizable — only pausing (single warning) when nothing is elide-eligible. (#4786)
Fixed GitHub-hosted repository file reads falling back to curl by adding a dedicated github file-read operation and explicit tool-routing guidance (#4805).
Bounded RPC JSONL frames to 1 MiB, compacted oversized agent_end frames without losing complete Python prompt results, and guaranteed worker reaping plus pending-request rejection after output-reader failures or explicit stops (#5405).
Fixed web_search being unreachable under default config: with tools.xdev: true, the discoverable web_search tool was mounted under xd:// and dropped from the top-level toolset, so models calling it directly got "Tool web_search not found". It is now pinned top-level via XDEV_KEEP_TOP_LEVEL while other discoverable tools keep mounting under xd:// (#5973).
Fixed onboarding omitting model selection by adding a persisted default-model step (fresh installs only — the setup version is not bumped, so existing installs are not re-prompted), and documented custom models.yml provider configuration and default-role selection (#5979).
Fixed autoResume crossing an explicit /new boundary: after /new a new session's JSONL is created lazily (only once assistant output exists), so exiting before any assistant message left the per-terminal breadcrumb pointing at a not-yet-materialized file. readTerminalBreadcrumbEntry rejected the missing target and continueRecent() fell back to the most-recent session — the pre-/new transcript — processing the next prompt with stale context. /new now records a durable fresh breadcrumb boundary that continueRecent() honors (starting fresh) even when the target is absent, while a genuinely stale/deleted breadcrumb still falls back to the most-recent session (#5730).
Fixed subagents that repeatedly submit malformed yield results from leaving the parent waiting forever; malformed submissions now repeat the required response format, and repeated invalid submissions fail the child with a clear error. (#4957)
Fixed configured or -e extensions in compiled binaries failing to resolve bundled @oh-my-pi/* value imports through the omp-legacy-pi-bundled: registry, and surfaced extension load failures during interactive and -p session startup. (#4954)

Removed

Fixed the Cursor-backed advisor losing entire turns when it selected server-native tools (bash, grep, etc.) outside its grant: exec-resolved native blocks are already rejected in-band by the advisor-scoped bridge, so they no longer trip the unavailable-tool quarantine and discard the advise emitted in the same turn (#5900).
Fixed custom anthropic-messages OAuth providers being unable to opt into configured Claude Code fingerprint header overrides. (#5888)
Fixed authoritative providers (e.g. openai-codex) keeping unsupported bundled models selectable when a fresh model cache and an expired OAuth token coincided: built-in discovery now forces the OAuth refresh so the provider's model manager is constructed and prunes stale bundled entries (e.g. gpt-5.4-nano) instead of waiting out the cache TTL. (#5364)

[17.0.5] - 2026-07-18

Added

Added support for Codex (ChatGPT subscription) in generate_image via the providers.image: "openai-codex" option, including automatic subscription detection and fallback logic.
Added an optional provider parameter to generate_image to override the global image provider setting for a single request.
Added OpenTelemetry log and metric export capabilities alongside existing trace exports, supporting standard OTLP environment variables.
Added support for id-prefixed targets and keys in retry.fallbackChains wildcards (e.g., "openrouter/google/*").
Added support for Shift+Enter in the session tree selector (/tree, /branch) to summarize and switch branches in a single step.
Added the PI_CONFIG_FILES environment variable to load settings overlays before --config overlays.

Changed

Changed bundled TTSR rules to warn instead of interrupting generation.
Renamed the system prompt's project-context section wrapper from <context> to <repo-rules> to prevent XML tag collisions with in-band tool dialects.
Renamed the /extensions dashboard tab "Agents (standard)" to "Agent Dirs (.agent/.agents)" to clarify its purpose.
Optimized performance by reducing concurrent subagent update CPU usage, skipping unnecessary title generation in non-interactive hosts, and memoizing convertToLlm conversions over settled history.
Improved the display of read xd:// calls by rendering them in a compact grouped view instead of full tool-execution cards.
Made the hashline seen-line guard opt-in and off by default via edit.enforceSeenLines.

Fixed

Fixed isolated task subagents mutating the parent git checkout and stacking parallel task branches by detaching the git directory.
Fixed Windows compatibility issues, including launch start daemons opening visible console windows, startup crashes when running from a drive root, and command errors in the hub tool with non-POSIX shells.
Fixed Windows stdio MCP servers launched through .cmd/.bat shims failing with Transport closed by escaping arguments properly.
Fixed browser tool selectors (tab.click, tab.select, etc.) to accept bare snapshot refs (e.g., tab.click("e501"), @e501) and resolved crashes when handed ElementHandle objects.
Fixed classifier refusals (e.g., Anthropic stop_reason: "refusal") ending turns with no visible error or exiting 0 in print mode.
Fixed transcript blocks being duplicated during streaming and tmux pane growth blanking finalized chat history.
Fixed JS/TS debugging by launching vscode-js-debug over TCP and synchronizing breakpoints across the session tree.
Fixed legacy pi extensions failing validation or loading when importing path helpers or package managers.
Fixed /login and /logout failing to refresh model discovery with fresh credentials due to stale cached data.
Fixed Cursor models and advisors failing to receive or execute mounted xd:// devices and MCP tools.
Fixed MCP tools repeatedly unmounting/remounting with overlapping sanitized prefixes, and stale tools remaining after disconnecting.
Fixed custom LSP servers crashing when requesting unconfigured workspace/configuration sections.
Fixed keyboard input dying after the first keypress when custom tools or modules hijack stdin at import time.
Fixed auto-compaction dead-ending in a warning loop when the most recent turn is over budget.
Fixed GitHub-hosted repository file reads falling back to curl by adding a dedicated github file-read operation.
Fixed active session markers and TUI usage panels truncating organization suffixes from same-email account labels.
Fixed write approval gates misclassifying xd:// device writes as exec.
Fixed bash command timeouts rendering with an incorrect error border, and resolved Windows bash crashes when piped commands time out.
Migrated legacy nested/quoted-dotted config keys (e.g., dev.autoqa.consent -> dev.autoqaConsent) on settings load.
Added managed ctx.setInterval / ctx.setTimeout / ctx.clearTimer helpers on extension contexts to prevent uncaught exceptions from crashing sessions.

[17.0.4] - 2026-07-18

Fixed

Fixed bundled Linux ffmpeg recording by selecting its available ALSA input when PulseAudio support is absent, and surfaced recorder stderr when capture fails (#5907).
Session load now skips the recursive async blob-ref resolver for entries with no blob:sha256: references. A cheap synchronous precheck gates the walk per entry (preserving the previous per-entry initiation order under synchronous store mutation), so text-heavy histories no longer pay the Promise.all tree descent for every non-session entry (#5922).
Fixed task tool schemas emitting boolean subschemas that llama.cpp grammar generation cannot parse (#5957).
Fixed the transcript keeping finalized assistant blocks in the live compose walk after their rows entered native terminal scrollback, making each stream tick's TranscriptContainer.render depth-linear in session length. Fully committed finalized blocks are now compacted out of the local frame regardless of post-finalize version tracking; a later mutation no longer recommits on ordinary frames (no duplication) and rehydrates on the next destructive full replay (no loss). Compose cost for a live tail tick is now flat as depth grows (bench/transcript-compose.bench.ts: ratio(N5000/N500) 2.30 → 0.90) (#5930).
Fixed /quit and /exit hanging during interactive shutdown by making the mnemopi dispose path retain the current session and flush in-flight extractions without sleeping the bank; the /memory enqueue path and end-of-session backend enqueue still perform full cross-session consolidation. (#3641)
Fixed interactive bash shortcut cd commands leaving the OMP session and status-line working directory unchanged.

[17.0.3] - 2026-07-17

Changed

omp usage and the in-session /usage view now show the Anthropic organization next to the account for org-scoped credentials (with --redact masking applied per part in the CLI, falling back to the org id when no display name is available), attribute "no usage data" rows per organization, and match the "in use by this session" marker by organization so only the active subscription is flagged. The OAuth login success message names the account and organization that was stored — a login landing on an unintended subscription is visible immediately.
/logout labels Anthropic accounts with their organization and marks only the credential of the active organization as active; omp token --list shows the organization next to each account. Two subscriptions sharing one email are distinguishable when selecting which to remove or mint a token for.
omp auth-broker migrate --from-local dedupes Anthropic OAuth identities per organization, so a Team seat already on the broker no longer blocks uploading the personal plan under the same email.
The status line invalidates its cached usage when the session rotates to a different Anthropic organization (previously the old subscription's quota could linger for the cache TTL), and omp auth-gateway check labels each credential with its organization so a failing row says which subscription needs re-login.
omp usage "no usage data" attribution is org-decisive whenever either the stored account or a report carries an organization: an org-less legacy credential whose own fetch failed is no longer hidden by an org-attributed sibling report sharing the same email.
Active-account matching for /usage, /logout, and omp token --list now treats a shared organization as a qualifier rather than a match: two Anthropic Team seats in one org (same org id, per-user pools) no longer flag each other's rows or reports as "in use by this session" — the base identity (account/email/project) is still required, with org-only sessions matching on the org alone.
omp usage "no usage data" coverage now requires the member's own identity within a shared organization: a sibling Team member's same-org report no longer counts as coverage for an account whose own report is missing, while an org-only account remains covered by any same-org report.
omp auth-broker migrate --from-local reruns now recognize an already-migrated org-only Anthropic row (login recovered neither email nor account) by its organization id instead of re-uploading it, which could overwrite the broker's newer refresh token with the stale local one.
Updated tangential agent forks to ignore parent session history and focus exclusively on the new request
Hardened /tan fork isolation: the clone's inherited todo list is cleared at fork (parent todo reminders no longer drag the tan back onto the parent's task), the fork notice warns that the parent is concurrently editing the same working directory, and the notice is re-injected after each compaction so the fork boundary survives summarization
Added visual markers in the transcript for elided tool calls that have no corresponding result
Updated status event log to prioritize the most recent entries in the display window
Updated the snapcompact shape preview transcript to use the compact scope format shown to models during compaction.

Fixed

Fixed xd:// mount notices triggering unsolicited model turns by deferring hidden notices until the next user prompt.
Fixed xd:// device tools appearing in the direct tool inventory and prompting invalid function calls (#5797).
Fixed history:// read selectors being treated as part of the agent id instead of paging the transcript (#5806).
Narrowed the history:// contract in the system prompt to match the implementation: it serves registered agents process-wide plus persisted subagents discoverable from their artifact trees, but does not discover unregistered top-level sessions solely from persisted session files (#5839).
Fixed expanded ! bash and eval output keeping a stale … N more lines (ctrl+o to expand) footer after Ctrl+O revealed every line (#5842).
Fixed MCP reauthentication continuing to an authorization URL without client_id after dynamic client registration fails; the registration error now blocks the flow with the provider response details (#5852).
Fixed collapsed todo views hiding the in-progress task in large phases. Both the transient Todo tool result and the sticky Todos HUD now share one walking-viewport policy: completed/abandoned tasks are omitted, every active task (the in-progress one, or a pending task a live subagent is executing) is pulled to the head in todo order, remaining rows fill with the following pending tasks, and an explicit … N more active todos summary is shown when active work alone exceeds the preview cap (#5873).
Fixed legacy provider extensions failing to load when they use the historical synchronous auth-storage surface (#5879).
Fixed orphaned detached MCP stdio server process trees surviving session dispose by escalating stdin-EOF → group SIGTERM → group SIGKILL on close() (#5578)
Fixed /new starting an unsolicited old-context provider turn when a hidden steer (e.g. an xd:// mount notice) was queued: the session transition is now an atomic boundary, so a queued steer/follow-up can no longer auto-resume against the pre-/new context while the session is disconnected mid-transition. /compact still resumes a steer/follow-up that arrives while it runs, draining the queue once it reconnects (#5800).
Fixed signed thinking-only Claude stops being discarded and retried as empty responses (#5881).
Fixed repeated URL reads and URL-backed searches returning stale same-session responses instead of refetching the resource (#5803).
Fixed read/write not recognizing ZIP-based .jar/.war/.ear/.apk files as archives, so read lib.jar:META-INF/MANIFEST.MF failed with path-not-found (#5808).
Fixed legacy binary .doc/.ppt/.xls/.rtf being advertised as convertible in read, fetch, and CLI @file handling despite having no markit converter, which surfaced an Unsupported format error instead of falling through to normal file handling (#5808).
Fixed Kimi Code transport selection to follow live per-model protocol metadata by default while preserving explicit OpenAI and Anthropic overrides (#5893).
Fixed repeated edit-tool rejections from local models by recovering comma-separated ranges and malformed trailers, while clarifying canonical string input and .= syntax (#5805).
Fixed LSP diagnostics and edit-time diagnostics writethrough for pull-only servers that advertise textDocument/diagnostic statically or through dynamic registration (#5825).
Fixed local ! command output concatenating carriage-return progress updates by preserving them as readable line boundaries (#5845).
Fixed hub/irc peer discovery after process-crash resume by restoring persisted subagents as parked peers before listing the roster (#5864).

Removed

Removed the unreliable Bing and Yahoo HTML-scraping web search providers

[17.0.2] - 2026-07-17

Added

Added native Warp CLI-agent events for rich session status, tool approvals, and completion notifications.
Added support for ChatGPT/Codex subscriptions in the generate_image tool, allowing image generation without a metered OPENAI_API_KEY even when using other active models.
Added an optional provider parameter to generate_image to override the global providers.image setting for a single request.
Added OpenTelemetry log and metric export support alongside existing trace exports, enabling forwarding of centralized-logger events and GenAI-semconv metrics when configured.
Enhanced retry.fallbackChains wildcards to support id-prefixed targets and keys, allowing more flexible model fallback routing across different providers.
Added an opt-in per-project model role storage mode with global fallback from the model selector.
Added per-advisor on/off toggle (enabled: false in WATCHDOG.yml): advisors stay in the roster but their runtime is never built — they show ○ in /advisor status rather than disappearing. Existing configs are backward-compatible (defaults to true when absent).
Colored the status line's advisor ++ badge by roster health (green all running, yellow quota-exhausted, red failed, dim paused); per-advisor glyphs (●/○/✕) show in /advisor status.
Added real provider quota display (usage percent, window, reset timer) to /advisor status and the /advisor configure preview.

Changed

Changed Bash command timeouts to render with a warning (yellow) border instead of an error (red) border, while still indicating to the model that the command did not complete normally.
Made the hashline seen-line guard opt-in and off by default via edit.enforceSeenLines, and improved handling of column-clipped lines so single-line edits on long lines apply without a full-width re-read.
Changed bundled TTSR rules to warn without interrupting generation.
Renamed the system prompt's project-context section wrapper from <context> to <repo-rules> to prevent collisions with the task tool's context parameter.
Enriched /advisor status to show per-advisor status glyphs, model, spend breakdown, and quota window for every configured advisor (including disabled ones), replacing the previous single-advisor-only summary.

Fixed

Fixed loading issues for linked legacy extensions importing DefaultPackageManager or linkedom.
Fixed the advisor retrying terminal, non-retriable provider failures (e.g., blocked prompts), ensuring they fail immediately while transient failures still retry.
Fixed an issue where reassigning the plan role model mid-planning did not take effect until the next plan-mode entry; it now applies at the next turn boundary.
Added managed timer helpers (ctx.setInterval, ctx.setTimeout, ctx.clearTimer) to the extension context to prevent self-scheduled callbacks from throwing uncaught exceptions and crashing the session.
Fixed /quit and /exit leaving stalled automatic title-generation requests alive during session teardown.
Fixed startup.quiet still rendering the xdev: xd://: mounted status line when MCP tools connect.
Fixed command errors in the hub tool when using a non-POSIX shell.
Fixed xdev-routed checkpoint and rewind writes not tracking checkpoint state.
Fixed the built-in advisor silently failing when its model routes through the Cursor provider by adding a Cursor execution bridge scoped to its granted tool set.
Fixed the fullscreen plan-review overlay staying visible until the approved execution turn finished; it is now hidden as soon as execution begins.
Fixed MCP tools repeatedly unmounting and remounting mid-session due to overlapping sanitized prefixes, and resolved stale tools remaining registered after disconnecting a server with special characters.
Fixed the /usage show marker and TUI usage panel to display and preserve the active organization suffix (email (OrgName)) to distinguish between multiple credentials with the same email.
Fixed Windows stdio MCP servers launched through .cmd or .bat shims failing with Transport closed by properly escaping arguments and spawning with windowsVerbatimArguments.
Fixed a startup crash on Windows when running from a drive root (e.g., R:\).
Fixed unknown __omp_worker_* CLI selectors exiting with code 0 instead of throwing an error.
Fixed Plan Review capturing mouse drags as pointer events, which prevented native terminal text selection.
Fixed orphaned TUI processes remaining alive after a fatal error and causing high resource consumption.
Fixed approved-plan execution looping through filesystem searches when a model rewrites the required plan read path.
Fixed Ask dialogs immediately accepting highlighted answers when they appear while the user is typing a space.
Stopped post-compaction auto-continue from opening another primary turn after a terminal text answer with no queued work.
Fixed the write approval gate misclassifying xd:// device writes as exec when the mounted tool declared a function-valued approval.
Fixed custom LSP servers (such as roslyn-language-server) crashing when requesting unconfigured workspace configuration sections.
Fixed late user-initiated bash results being recorded in whichever session or branch was active when execution finished; they now retain their originating transcript.
Fixed Claude Code marketplace plugins with scope: "local" leaking skills, hooks, tools, commands, and MCP servers into unrelated projects.
Fixed headless omp -p waiting indefinitely after a completed turn when final consolidation stalls.
Fixed explicit-tool sessions bypassing xd:// presentation for ambient discoverable custom and MCP tools.
Fixed providers.webSearch: kimi incorrectly sending Moonshot credentials instead of Kimi Code credentials.
Fixed registerTool demoting essential built-in tools to discoverable when a re-registration omitted loadMode.
Fixed the advisor skipping the next user instruction after auto-learn accepted and pruned a terminal empty assistant stop.
Fixed built-in advisors retrying quota- or rate-limited providers instead of applying the matching retry.fallbackChains model chain.
Updated model selector status messages to use role tags (SMOL, SLOW) instead of display names (Fast, Thinking) for consistency.
Fixed Cursor models receiving only top-level tools by forwarding mounted xd:// devices through Cursor's request-context MCP catalog.
Fixed Windows bash crashes when a piped command times out while flushing output.
Fixed a race condition where hub/IRC send and ensureLive could inject into a subagent session during disposal.
Migrated legacy nested/dotted configuration keys (dev.autoqa.consent and todo.reminders.max) to flat keys (dev.autoqaConsent and todo.remindersMax) on settings load.
Fixed keyboard input dying after the first keypress when a custom tool module attaches a stdin consumer at import time.
Fixed the alignment of the ask tool's custom-input dialog to start at column 0.
Fixed advisor context maintenance undercounting the provider context by anchoring compaction decisions on provider-reported context usage.
Fixed RPC mode (--mode rpc) crashing on non-JSON stdin lines; malformed lines are now reported as errors while the frame loop continues.
Fixed local llama.cpp Qwen-family models not honoring the --thinking off flag.
Documented the ultrathink, orchestrate, and workflowz magic keywords, including their effects, matching rules, and settings.
Fixed /clear autocomplete selecting /autoresearch and updated /clear to start a new session as an alias for /new.
Fixed /review aborting entirely when GitHub rejects a pull request's aggregate diff for exceeding the line limit by falling back to the paginated per-file endpoint.
Fixed /q + Enter running /queue instead of /quit by adding an explicit q alias to /quit.
Fixed Ctrl+L (app.display.reset) not refreshing the dark/light theme on certain terminals by issuing a background re-query before repainting.
Fixed a failing advisor stalling the primary agent: the per-turn catch-up gate parked the primary for up to its full 30s budget while a broken advisor (unsupported model, dead endpoint, render bug) retried — and an advisor exception could abort the primary's turn-end outright. A failing advisor now releases parked waiters the moment its turn fails (before any async hook), refuses new parks until a turn succeeds, and the turn-end boundary isolates advisor exceptions completely; a failed render restores the delta cursor so nothing is lost when the advisor recovers.
Fixed advisors retrying a permanently rejected request forever (e.g. invalid_request_error: model not supported with this account): unlike quota exhaustion — which pauses with a notice until an explicit reset — this class notified once and silently kept re-attempting every turn, re-building heavy context in a shared daemon. The runtime now hard-stops after a permanent rejection or three consecutive backlog-drop cycles, with a visible notice; an explicit reset (/new, config rebuild, restart) re-enables it. waitForCatchup resolves immediately while halted so the primary agent is never parked on a runtime that cannot drain.

[17.0.1] - 2026-07-16

Changed

omp usage and the in-session /usage view now show the Anthropic organization next to the account for org-scoped credentials (with --redact masking applied per part in the CLI, falling back to the org id when no display name is available), attribute "no usage data" rows per organization, and match the "in use by this session" marker by organization so only the active subscription is flagged. The OAuth login success message names the account and organization that was stored — a login landing on an unintended subscription is visible immediately.
/logout labels Anthropic accounts with their organization and marks only the credential of the active organization as active; omp token --list shows the organization next to each account. Two subscriptions sharing one email are distinguishable when selecting which to remove or mint a token for.
omp auth-broker migrate --from-local dedupes Anthropic OAuth identities per organization, so a Team seat already on the broker no longer blocks uploading the personal plan under the same email.
The status line invalidates its cached usage when the session rotates to a different Anthropic organization (previously the old subscription's quota could linger for the cache TTL), and omp auth-gateway check labels each credential with its organization so a failing row says which subscription needs re-login.
omp usage "no usage data" attribution is org-decisive whenever either the stored account or a report carries an organization: an org-less legacy credential whose own fetch failed is no longer hidden by an org-attributed sibling report sharing the same email.
Active-account matching for /usage, /logout, and omp token --list now treats a shared organization as a qualifier rather than a match: two Anthropic Team seats in one org (same org id, per-user pools) no longer flag each other's rows or reports as "in use by this session" — the base identity (account/email/project) is still required, with org-only sessions matching on the org alone.
omp usage "no usage data" coverage now requires the member's own identity within a shared organization: a sibling Team member's same-org report no longer counts as coverage for an account whose own report is missing, while an org-only account remains covered by any same-org report.
omp auth-broker migrate --from-local reruns now recognize an already-migrated org-only Anthropic row (login recovered neither email nor account) by its organization id instead of re-uploading it, which could overwrite the broker's newer refresh token with the stale local one.
Updated tangential agent forks to ignore parent session history and focus exclusively on the new request
Hardened /tan fork isolation: the clone's inherited todo list is cleared at fork (parent todo reminders no longer drag the tan back onto the parent's task), the fork notice warns that the parent is concurrently editing the same working directory, and the notice is re-injected after each compaction so the fork boundary survives summarization
Added visual markers in the transcript for elided tool calls that have no corresponding result
Updated status event log to prioritize the most recent entries in the display window
Updated the snapcompact shape preview transcript to use the compact scope format shown to models during compaction.

Fixed

Fixed xd:// mount notices triggering unsolicited model turns by deferring hidden notices until the next user prompt.
Fixed xd:// device tools appearing in the direct tool inventory and prompting invalid function calls (#5797).
Fixed history:// read selectors being treated as part of the agent id instead of paging the transcript (#5806).
Narrowed the history:// contract in the system prompt to match the implementation: it serves registered agents process-wide plus persisted subagents discoverable from their artifact trees, but does not discover unregistered top-level sessions solely from persisted session files (#5839).
Fixed expanded ! bash and eval output keeping a stale … N more lines (ctrl+o to expand) footer after Ctrl+O revealed every line (#5842).
Fixed MCP reauthentication continuing to an authorization URL without client_id after dynamic client registration fails; the registration error now blocks the flow with the provider response details (#5852).
Fixed collapsed todo views hiding the in-progress task in large phases. Both the transient Todo tool result and the sticky Todos HUD now share one walking-viewport policy: completed/abandoned tasks are omitted, every active task (the in-progress one, or a pending task a live subagent is executing) is pulled to the head in todo order, remaining rows fill with the following pending tasks, and an explicit … N more active todos summary is shown when active work alone exceeds the preview cap (#5873).
Fixed legacy provider extensions failing to load when they use the historical synchronous auth-storage surface (#5879).
Fixed orphaned detached MCP stdio server process trees surviving session dispose by escalating stdin-EOF → group SIGTERM → group SIGKILL on close() (#5578)
Fixed /new starting an unsolicited old-context provider turn when a hidden steer (e.g. an xd:// mount notice) was queued: the session transition is now an atomic boundary, so a queued steer/follow-up can no longer auto-resume against the pre-/new context while the session is disconnected mid-transition. /compact still resumes a steer/follow-up that arrives while it runs, draining the queue once it reconnects (#5800).
Fixed signed thinking-only Claude stops being discarded and retried as empty responses (#5881).
Fixed repeated URL reads and URL-backed searches returning stale same-session responses instead of refetching the resource (#5803).
Fixed read/write not recognizing ZIP-based .jar/.war/.ear/.apk files as archives, so read lib.jar:META-INF/MANIFEST.MF failed with path-not-found (#5808).
Fixed legacy binary .doc/.ppt/.xls/.rtf being advertised as convertible in read, fetch, and CLI @file handling despite having no markit converter, which surfaced an Unsupported format error instead of falling through to normal file handling (#5808).
Fixed Kimi Code transport selection to follow live per-model protocol metadata by default while preserving explicit OpenAI and Anthropic overrides (#5893).
Fixed repeated edit-tool rejections from local models by recovering comma-separated ranges and malformed trailers, while clarifying canonical string input and .= syntax (#5805).
Fixed LSP diagnostics and edit-time diagnostics writethrough for pull-only servers that advertise textDocument/diagnostic statically or through dynamic registration (#5825).
Fixed local ! command output concatenating carriage-return progress updates by preserving them as readable line boundaries (#5845).
Fixed hub/irc peer discovery after process-crash resume by restoring persisted subagents as parked peers before listing the roster (#5864).

Removed

Removed the unreliable Bing and Yahoo HTML-scraping web search providers

[17.0.1] - 2026-07-16

Removed

Fixed a crash when a plugin/custom tool renderer returns a component that throws during its later render() pass (e.g. TypeError: th.bold is not a function from a plugin that styles its header off an object without a bold method). ToolExecutionComponent now wraps every renderer-returned call/result component so a throwing render() degrades to the safe fallback (tool label or raw result text) instead of taking down the transcript (#4978).
Fixed /login for paste-code providers (Codex, Anthropic, Gemini CLI, GitLab Duo, Antigravity, Devin) dropping the pasted fallback redirect URL: the login dialog captured focus but never mounted an input, and the "complete pairing with /login <url>" tip pointed at the hidden, unfocused editor. The dialog now mounts a focused input for the manual code/URL paste (#5339).
Fixed /tan and /fork clones cold-missing the provider prompt cache: the per-turn supersede/useless-result prune rewrote the live context without persisting it, so file-based forks and resume rebuilt a divergent (un-pruned) prefix and re-wrote the entire cache
Fixed /tan pinning the clone's prompt-cache key to the parent's session id instead of the parent's effective cache key, dropping shard affinity when the parent was itself a fork or tan
Fixed /resume and plan approval exposing the previous session while their asynchronous session replacement was still loading by keeping fullscreen overlays mounted until the rebuilt transcript is ready (#5319).
Fixed inconsistent history rendering when toggling the display setting for compacted items
Fixed configured retry.fallbackChains never engaging on non-retryable provider errors (e.g. "Cloud Code Assist API returned an empty response"): a hard error on a model covered by a fallback chain now switches to the next candidate instead of failing the turn, while still never backoff-retrying the failing model itself
Fixed transcript rebuilds (compaction, /compact, and toggling history display) repainting content below stale scrollback when collapsing history; rebuilds now correctly clear the scrollback buffer when history is collapsed
Improved auto-compaction to automatically drop images and elide content when context is tight, and added persistent warning badges to the compaction divider when manual intervention is required
Fixed backgrounded Bash blocks continuing to repaint with live and final job output; they now freeze with a compact job notice while completion is delivered separately
Fixed the downshift plan nudge silently ending the run with no code written when the model answered with a text-only reply (no tool call): the agent loop treats a tool-call-free turn as a natural stop and never prompts again, which the nudge's own "write the plan in your next reply" instruction makes common. The nudge now explicitly tells the model this is a checkpoint, not a final answer, and the session forces one more turn whenever a post-nudge reply lands with zero tool calls
Fixed launch tool rendering stacking a stale pending header over a bare ✓ Launch line and raw text: the tool now uses a merged registry renderer with one per-op status header (op, target, state · pid · uptime meta), stripped log cursor suffixes, capped collapsed log/list previews, and a launch tool glyph
Fixed confusing launch start/wait results when readiness timed out with the log pattern already matched (readiness needs log AND port): the result printed a contradictory Ready: <match> next to Readiness timed out without naming the failing condition. Daemon snapshots now carry the unmet conditions (readyPending), and start/wait results state exactly what never happened (e.g. port 3100 on 127.0.0.1 never accepted connections); the TUI shows a waiting on port badge on starting daemons
Fixed the in-process stat builtin mangling BSD-style invocations like stat -f "%Sm %N" file (macOS muscle memory): GNU -f means --file-system, so the format string was treated as a file operand — printing filesystem info for the real operands and erroring with cannot read file system information for '%Sm %N'. A -f whose format value contains % is now detected as BSD syntax and translated to the GNU equivalent (%Sm→%y, %N→%n, %z→%s, epoch/S-form times, owner/group/permission and H/L sub-field directives, -L/-n/-q/-F flag clusters, with %n/%t as literal newline/tab); directives with no GNU counterpart fail with a clear unsupported BSD format directive error
Fixed the remaining GNU-flavored shell builtins that broke under macOS/BSD muscle memory, using the same unambiguous-detection approach as the stat fix (only invocations that are invalid or nonsensical under GNU semantics are reinterpreted; unsupported BSD forms fail loudly instead of producing wrong output): date -r <epoch> formats the epoch when no such file exists (GNU -r FILE mtime preserved), signed date -v±N<unit> adjustments translate to -d relative dates and -j is accepted (-j -f strptime parse mode and field-set -v error clearly); sed -i '' 's/…/…/' file drops the BSD empty backup-suffix token instead of treating it as the script; mktemp -t prefix without X's creates $TMPDIR/prefix.XXXXXXXXXX (the GNU too few X's error path); tail -r reverses input by delegating to tac (with -n/-c/-f combinations erroring clearly); find -E maps to -regextype posix-extended ahead of the expression; base64 -D decodes as an alias of -d; and ln -sfh works via a -h alias of --no-dereference (clap's -h help short is dropped to match real GNU/BSD ln; --help unchanged)
Fixed the browser tool crashing the whole process (parent session and every subagent) when a CDP world re-acquire failed mid-navigation: the stealth puppeteer-core patch called the bare debugError logger, which is undefined while the puppeteer:error debug channel is disabled (the default), turning a transient acquire failure into a fatal TypeError unhandled rejection. The patched FrameManager/WebWorker acquire paths now use debugCatchError (#5296)
Fixed a role with a :high thinking suffix resolving to a longer sibling model whose id embeds the tier name (e.g. kimi-for-coding:high → kimi-for-coding-highspeed). The thinking suffix is now stripped before any fuzzy match, so provider/model:high keeps the exact model at high effort (#5151).

Fixed

Fixed the omp grep CLI subcommand failing on paths with a stray leading colon (e.g. :/abs/path); it now routes the path argument through expandPath like read/edit/in-agent grep. Broadened expandPath's leading-colon strip to also recover Windows-style shapes (:C:\repo\file, :.\src, :..\rel, :\\server\share) (#5624).
Fixed the tail builtin exiting the entire omp process with code 13 on Windows when its output pipe broke (e.g. seq ... | tail -n 3 | head -n 0); a broken pipe now surfaces as a normal error instead of calling std::process::exit (#5609).
Fixed a late advisor blocker after a terminal primary answer being deferred to the next user turn instead of continuing the current turn: resolveAdvisorDeliveryChannel preserved every interrupting severity as a passive card once the primary ended with a terminal text answer and no queued work remained, so a blocker flagging a mistake in the final output sat idle until the next prompt. A blocker now steers a triggered turn so the primary acknowledges and continues before the turn is considered done; a late concern still preserves as a visible card (#5628).
Fixed independently keyed extension hook statuses sharing one truncated line; each status now renders on its own deterministic line (#5617).
Fixed xAI web search bypassing configured xai / xai-oauth proxy endpoints and headers, while preventing official OAuth tokens from being sent to custom endpoints (#5599).
Fixed models.yml rejecting the Anthropic compat.supportsEagerToolInputStreaming override for custom endpoints (#5572).
Fixed long streamed table responses duplicating in terminal scrollback when later rows widened an earlier column.
Fixed Bash internal URLs remaining unresolved when used as unquoted arguments inside command substitutions (#5535).
Fixed the built-in fd printing fd: Broken pipe (os error 32) when a downstream pipeline reader exited early (e.g. fd … | head); it now exits silently with 141 (128+SIGPIPE), matching real fd.
Fixed prewalk repeatedly continuing after a bash-only task such as commit had already completed (#5551).

Added

Fixed the Codex config.toml MCP importer dropping cwd and leaving relative command values unrooted, which broke the bundled Codex Computer Use server (ENOENT on spawn); relative command/cwd now resolve against the Codex config directory like the claude-plugins/omp-plugins providers (#5561).
Fixed streamed replace-mode edits with ssh:// paths terminating the active prompt before normal tool dispatch (#5552).
Fixed concurrent provider OAuth refreshes from invalidating Anthropic's rotating refresh token, and prevented background usage probes from permanently disabling credentials after refresh failures (#5396).
Restored the regression test guarding compiled-binary web-search header generation: browser-headers.ts lazily constructs header-generator and falls back to a static Chrome profile when its data_files are absent, but the test defending that contract had been removed. Without the guard, a compiled binary threw ENOENT at import time, breaking the Bing web_search provider (undefined is not a constructor) and extension loading / omp plugin install. (#5256)
Fixed browser tabs hanging indefinitely at Closing <tab> when a worker, CDP target, browser process, or cmux surface stalls during teardown; close deadlines now release the operation with backend, tab, and pending-resource diagnostics. (#5259)
Fixed the agent://<parent>/<child> slash form failing to resolve a nested subagent's output: the path segment was always treated as a jq JSON-extraction key against <parent>.md, so a precise-planner reading its own scout child (agent://Plan/Scout) got Not found. The slash is now a hierarchy separator first (agent://Parent/Child → Parent.Child.md), falling back to JSON extraction only when no nested output matches the path. (#5238)
Fixed fullscreen Plan Review jumping to the top while scrolling when a transient terminal resize or Markdown reflow made the body temporarily non-scrollable. (#5232)
Fixed generate_image preferring Antigravity over the active session provider and stopping instead of trying the next credentialed provider after an image HTTP failure. (#5218)
Fixed /reload-plugins, plugin setting changes, and manage_skill writes leaving runtime skills and skill:// resolution stale until restart; sessions now rediscover enabled skills and rebuild /skill:<name> commands before the next prompt (#4996).
Fixed documented omp marketplace/discover/upgrade/uninstall/enable/disable CLI verbs silently leaking to the model as a launch prompt instead of managing plugins. omp marketplace add xyz (and similar multi-word invocations following the documented omp plugin <action> grammar) now surface a hint pointing at the real omp plugin <action> command, while genuine prose prompts beginning with these words still route to launch (#4845).
Fixed /mcp, /mcp list, and /tools output duplicating in terminal scrollback when invoked during agent streaming by deferring command panels until the active turn ends (#4806).
Fixed bash/eval/ssh output that was only per-line column-capped being misreported as byte-window truncation, which appended a bogus Showing lines X-Y of Z (…B limit). Read artifact://N for full output footer even though every line was shown. Column-cap trimming now surfaces solely as the Some lines truncated to N chars notice (#4735).
Fixed the nerd status-line preset's session icon using a removed Nerd Fonts v2 codepoint instead of the current Nerd Fonts v3 mapping (#4795).
Fixed omp crashing at startup (TypeError: undefined is not an object (evaluating 'this.#theme.symbols.boxRound')) after installing a plugin whose custom editor subclasses CustomEditor/Editor and forwards the upstream-pi super(tui, theme, keybindings) constructor — the arg order that setEditorComponent's factory contract advertises. CustomEditor now resolves the real EditorTheme by shape rather than position and captures a leading TUI for plugin overrides (#4766).
Fixed Other response editors leaving Windows Terminal IME candidate windows at the terminal edge by forwarding dialog focus to the nested editor (#4760).
Rendered and persisted native OpenAI Responses image_generation_call results as session images (#4768).
Fixed ACP stdio EOF/EPIPE disconnects bypassing awaited session teardown and leaving in-flight tool calls pending in persisted rollouts (#4788).
Routed the print-mode assistant-error/aborted exit, RPC pi.shutdown() and stdin-EOF shutdowns, and the extension command-context shutdown() through the awaited, idempotent session.dispose() before process.exit(), so the bounded browser reaper (releaseTabsForOwner) always runs and OMP-owned Chromium no longer outlives the process (#5643).

[17.0.0] - 2026-07-15

Breaking Changes

Merged the irc, job, and launch tools into a single unified hub tool (loadMode: "essential"). Messaging, job control, and process supervision operations are now routed through this single tool. SDK: IrcTool, JobTool, LaunchTool, IrcDetails, and JobToolDetails have been removed in favor of HubTool, CoordinationDetails, LaunchToolDetails, and hubToolRenderer from tools/hub.
Removed the hidden resolve tool. Staged actions are now finalized through three plain-text resolution devices: xd://resolve (apply preview), xd://reject (discard preview), and xd://propose (submit a plan slug/title for approval). The write tool is now auto-included whenever a deferrable tool is present or plan mode is enabled. SDK: ResolveTool and HIDDEN_TOOLS.resolve have been removed in favor of dispatchResolutionDevice() and queueResolveHandler().
Unified tool presentation on loadMode (essential | discoverable), replacing the custom-tool xdev?: boolean opt-out. Custom, extension, MCP, RPC host, image-generation, and TTS tools now default to discoverable and mount under xd:// when enabled. generate_image, tts, and MCP tools are now exposed as xd:// devices in default sessions instead of shipping their schemas top-level.
Removed the BM25 tool-discovery system, including the search_tool_bm25 tool, the tools.discoveryMode, mcp.discoveryMode, and mcp.discoveryDefaultServers settings, and per-tool MCP selection. All connected MCP tools are now enabled and mounted under the xd:// transport. SDK: search_tool_bm25, the tool-discovery module, and associated AgentSession discovery/MCP-selection methods have been removed.
Removed the legacy report_finding tool. Reviewer agents now record findings through incremental yield sections. SDK: reportFindingTool and HIDDEN_TOOLS.report_finding have been removed.
Removed the ssh agent tool (remote command execution). The ssh:// read/write/search protocol, the omp ssh host-management CLI, and SSH host discovery are retained. SDK: SshTool, loadSshTool, the ssh/ssh-executor module, and AgentSession.refreshSshTool have been removed.
Removed the tools.essentialOverride setting, the mcp_tool_selection session message type, and the xdev --tools token.

Added

Added xd:// virtual tool devices (controlled by the tools.xdev setting, default on), allowing mounted tools to be discovered via read xd://, documented via read xd://<tool>, and executed via write xd://<tool>.
Added the edit.enforceSeenLines setting (default off) to gate the hashline seen-line guard. When enabled, edits anchored on lines that a prior read or grep never displayed are rejected.
Added per-agent prewalk for subagents, featuring a prewalk frontmatter field, a task.agentPrewalk settings override toggled from the /agents dashboard, and a task.prewalk boolean (default off) to arm the bundled generic task agent.

Changed

Renamed "dev.autoqa.consent" to "dev.autoqaConsent" and "todo.reminders.max" to "todo.remindersMax" to eliminate nested configuration prefix collisions in standard JSON/YAML.
Made the hashline seen-line guard opt-in and off by default, and stopped excluding column-clipped (>512-char) lines from a snapshot's seen set, allowing single-line edits on long lines to apply without a full-width re-read.
Changed the default astGrep.enabled setting to false.
Batched todo operations with real tool calls to prevent solo todo turns and extra round trips.
Changed all bundled TTSR rules to warn instead of interrupting generation.
Renamed the system prompt's project-context section wrapper from <context> to <repo-rules> to prevent collisions with the task tool's context parameter under in-band XML tool dialects.
Rendered read xd:// calls in a compact grouped read view instead of a full tool-execution card.
Updated --tools to reject unknown tool names with a usage error instead of silently narrowing the toolset.
Capped the xd:// device docs inlined into the system prompt to a 48k-char budget (10k per device) to prevent large MCP catalogs from bloating requests; devices past the cap are listed by name and summary and fetched on demand. External (dynamic-mount) tool descriptions embed at most 200 chars — schemas stay intact, full text via read xd://<tool>.
Dead BM25-discovery settings keys (tools.discoveryMode, tools.essentialOverride, mcp.discoveryMode, mcp.discoveryDefaultServers) are cleaned from configs on load; tools.xdev keeps its default (true).
Mid-session xd:// mount changes (e.g. MCP connect/disconnect) no longer rewrite the system prompt: the delta is announced to the model as a steered system notice ("these tools became available" / "no longer mounted"), so the provider prompt cache stays intact; device docs join the prompt on the next unrelated rebuild.

Fixed

Fixed a bug where a nested configuration value (like dev.autoqa.consent / dev.autoqaConsent) would incorrectly satisfy a parent key lookup (like dev.autoqa), causing Auto QA to be enabled and prompt for consent by default when it should have been disabled.
Fixed compiled appserver startup deadlocking before socket creation when user extensions were present.
Fixed Bash internal URLs remaining unresolved when used as unquoted arguments inside command substitutions.
Fixed --tools silently dropping hidden tool names like xdev and yield.
Fixed the built-in fd printing broken pipe errors when a downstream pipeline reader exited early; it now exits silently with status 141.
Fixed prewalk repeatedly continuing after a bash-only task (such as commit) had already completed.
Fixed the Bash tool hanging when in-process commands read process substitution operands.
Fixed /share and /export web views rendering inline Markdown inside list items as literal text.
Fixed plan-mode re-entry dropping a new plan request when a prior plan artifact existed; re-entry now anchors on the new request and folds old-plan corrections into it.
Fixed ACP clients rendering xd:// device dispatches as file edits; they now map to an execute-kind tool call titled with the device URL.
Fixed non-yolo approval modes double-prompting for xd:// device dispatches.
Fixed TTSR rules with leading inline regex flags failing to compile and being silently dropped in Bun/JS environments, and recovered scope tokens and sibling values from malformed frontmatter.

[16.5.2] - 2026-07-14

Breaking Changes

Removed the separate selector parameters from read and grep; line ranges and read modes must now be appended to path.

Added

Added a generate_image.enabled setting (Settings › Tools › Generate Image) to allow toggling the image generation tool.

Changed

Expanded provider rate-limit response header ingestion to all supported providers with header parsers (previously Anthropic-only), enabling proactive account rotation for multi-account sessions before hitting 429 errors.
Restored CPU model metadata in workstation prompts on non-Linux hosts.

Fixed

Fixed omp config list --json output truncation at 64 KiB when stdout is piped.
Fixed vim-style navigation (h/j/k/l) under the Kitty keyboard protocol.
Fixed /guided-goal throwing Model not found errors on websocket-only Codex models by routing the interview through the session's provider transport and reusing a single isolated side session.
Fixed tool_result extension handlers being unable to rewrite the model-visible content of a thrown tool failure.
Fixed intermittent Perplexity OAuth web search failures (401 errors) caused by transient transport drops on the ask endpoint.
Fixed the generate_image tool ignoring --no-tools and explicit tool whitelists.
Fixed markerless prose thinking preambles incorrectly becoming session titles when title models omit the <title> marker.
Fixed the agent recreating a todo list immediately after a user clears it with /todo rm.
Fixed internal-URL autocomplete (agent://, skill://, omp://, etc.) not triggering inside slash command arguments.
Fixed the browser tool hanging indefinitely during tab closure when the headless Chromium process is wedged on Windows.
Fixed history:// URLs failing to resolve for unregistered, released, or resumed subagents by falling back to scanning artifacts directories on disk.
Fixed eval cells treating timeout: 0 as a one-second deadline and reporting session-deadline cancellations as user aborts.
Fixed MCP OAuth dynamic client registration for pathful authorization-server issuers by preserving the discovered registration endpoint.
Fixed the launch tool failing to start Windows executables due to double-escaped PTY commands and arguments.
Fixed the built-in advisor warning Advisor unavailable for silent reviews: a content-less stop is a valid "nothing to add" outcome and is never retried or warned about, regardless of reported token usage (#5212 follow-up).
Fixed read, edit, and grep tools failing on paths with a stray leading colon emitted by some models.
Fixed git plugin re-installs retaining stale commits by fetching Bun's cached clone before updating the lockfile pin.
Fixed overlapping Bash timeout and interrupt cleanup to explicitly abort isolated shells instead of leaving child processes running.
Fixed a bug where generic provider aborts arriving as stopReason: "error" were not auto-retried.
Fixed switching from a vision model to a text-only model mid-session sending historical image blocks to the new provider.
Fixed inline images in Agent Hub transcripts by routing replayed images through the shared image budget and Kitty placeholder renderer.
Fixed OSC 5522 paste in direct API-key login prompts being routed to the hidden main chat editor instead of the focused credential field.
Fixed plugin installation failures when an ES module extension synchronously requires CommonJS helpers.
Fixed GitHub code search rejecting empty optional date placeholders.
Fixed /tree navigation onto a /skill: injection node landing on the incorrect entry.
Fixed interactive TUI sessions crashing with concurrent JS runtime errors when the JS eval worker falls back to the in-process inline path.
Fixed compaction aborting instead of trying an authenticated fallback model when Amazon Bedrock credential resolution fails.
Fixed full-context forks and /tan clones cold-missing OpenAI prompt caches by properly persisting and inheriting provider prompt-cache keys.
Fixed Codex advisor requests using local session labels as provider session IDs.
Fixed macOS stdio MCP servers launching in a detached session, allowing the TCC Apple Events permission prompt to trigger.
Fixed the ask tool timeout to auto-select the recommended option when the UI selector does not settle.
Fixed LSP workspace diagnostics for Go workspaces to correctly recognize go.work roots and include all used modules.
Fixed interactive OAuth login success messages waiting on background model discovery.
Fixed Windows bash tool crashes when an explicit timeout fires while a piped command is still streaming.
Fixed subagent yield tool calls being discarded when the soft request budget hard-aborted the same assistant turn.
Fixed --tools filtering in interactive sessions disabling deferred MCP tools.
Fixed kept-alive task subagents entering repeated provider-call loops after an IRC wake and terminal yield.
Fixed manual /compact with the snapcompact strategy hard-failing on text-only active models.
Fixed the empty-editor ←← gesture trapping input when opening the Agent Hub from persisted/parked subagents.
Fixed eval read() URI handling in Python and JS runtimes to correctly delegate URI reads and pass pagination arguments.
Fixed discovered plugin .mcp.json stdio servers launching relative command or cwd values against the session cwd instead of the plugin's config directory.
Fixed Model Hub DEFAULT role assignments with auto retaining a stale concrete reasoning suffix after restart.
Fixed configured retry.fallbackChains failing to engage on non-retryable provider errors.
Fixed backgrounded Bash blocks continuing to repaint with live output after completion.
Fixed --reasoning-slide-plan silently ending the run with no code written when the model answered with a text-only reply.
Fixed launch tool rendering issues, including stacked pending headers and confusing start/wait results when readiness timed out.
Fixed the in-process stat and other GNU-flavored shell builtins (such as date, sed, mktemp, tail, find, base64, and ln) mangling or failing on macOS/BSD-style invocations.

[16.5.1] - 2026-07-14

Changed

Enhanced Anthropic credential and usage management to support organization-scoped accounts, including displaying organization names in /usage, /logout, omp token --list, and OAuth login success messages, resolving active-account matching for shared organizations, and deduplicating identities during migration.

[16.5.0] - 2026-07-13

Breaking Changes

Replaced the --reasoning-slide-* flag family with a unified --prewalk mechanism (--prewalk, --prewalk-into <model>, and --no-prewalk) to manage model handoffs during execution.

Added

Added a new --prewalk execution flow (with --prewalk-into <model> and --no-prewalk overrides) that starts tasks on a strong model for planning and todo initialization before handing off to a faster, cheaper model for implementation.
Added a status line annotation for the active prewalk phase (armed or active).
Added the tui.scrollbackRebuild setting to gate the erase-and-replay native scrollback rebuild mechanism (defaults to off).
Added a display setting to toggle between collapsing or keeping compacted history inline in live session displays.
Added a compact session-only model picker (Alt+P) for quick model switching, featuring @ search to quickly list and apply configured quick roles.
Redesigned Agent Hub entries into a cleaner two-line card layout showing identity, active model, reasoning level, age, and task description.
Added a project-scoped launch tool (gated by launch.enabled) for managing shared long-running services and debuggers, featuring readiness probes, bounded logs, PTY input, restart policies, and automatic teardown.
Added support for detached launches, allowing standalone services to survive broker shutdowns and reconnect to subsequent sessions.

Changed

Updated JSON logs (--mode json) to include provider payloads in auto-compaction events.
Updated tangential agent forks (/tan) to ignore parent session history and focus exclusively on the new request, hardening isolation with cleared todo lists and concurrent editing warnings.
Added visual markers in the transcript for elided tool calls that have no corresponding result.
Updated the status event log to prioritize the most recent entries in the display window.
Upgraded @agentclientprotocol/sdk to version 1.2.1.

Fixed

Fixed terminal scrollback duplication issues with expanded streaming edit previews (Ctrl+O) by using a viewport-sized tail window.
Fixed custom model role resolution and alias parsing, ensuring canonical role selectors (@role) and thinking suffixes resolve correctly across all configuration surfaces.
Fixed quadratic growth in JSON logs by eliding redundant message snapshots and payloads.
Fixed prompt cache misses and incorrect cache key pinning for /tan and /fork clones.
Fixed inconsistent history rendering and scrollback repainting when toggling the display setting for compacted items.
Fixed retry.fallbackChains failing to engage on non-retryable provider errors, ensuring the agent correctly falls back to the next candidate model.
Improved auto-compaction to automatically drop images and elide content when context is tight, and added persistent warning badges when manual intervention is required.
Fixed backgrounded Bash blocks continuing to repaint with live output; they now freeze with a compact job notice while completion is delivered separately.
Fixed rendering, status display, and PTY control sequence formatting issues in the launch tool.
Fixed in-process shell builtins (including stat, date, sed, mktemp, tail, find, base64, and ln) to correctly detect and translate macOS/BSD-style arguments and flags, preventing failures caused by GNU-only assumptions.

Removed

Removed the --prewalk-boomerang feature and its associated configuration setting.
Removed the unreliable Bing and Yahoo HTML-scraping web search providers.

[16.4.8] - 2026-07-12

Added

Added invocation-specific schemas to task subagents and unified task/eval agent execution, including host-enforced read-only plan-mode agents (#5279)

Fixed

Fixed compatibility of GNU-flavored shell builtins (such as stat, date, sed, mktemp, tail, find, base64, and ln) when invoked with macOS/BSD-style arguments.
Fixed subagent model and thinking level resolution to correctly respect the configured modelRoles.task selector instead of intermittently falling back to the parent session's model.
Fixed TUI rendering issues, including preventing macOS runtime diagnostics from painting into the viewport, bounding transcript retention in long sessions, and fixing scrollback repainting when collapsing history.
Fixed /tan and /fork clones failing to inherit or persist the parent session's prompt cache keys.
Fixed Python and JavaScript evaluation kernels suspending the CLI on subprocess foregrounding, deadlocking on non-serializable values, or losing in-flight subagent work during external aborts.
Fixed configured retry.fallbackChains failing to engage when encountering non-retryable provider errors.
Improved auto-compaction to automatically drop images and elide content when context is tight, and added persistent warning badges to the compaction divider when manual intervention is required.
Fixed the downshift plan nudge silently ending runs with no code written when the model answered with a text-only reply.
Fixed launch tool rendering and status reporting, including resolving contradictory readiness timeout messages and preventing backgrounded Bash blocks from continuing to repaint.
Fixed Advisor containment and timing issues, preventing hallucinated tool calls from contaminating later advice and ensuring late-arriving transcript deltas are coalesced before advisor calls.
Fixed omp update on npm-managed Windows installations to prevent downloaded release binaries from overwriting npm launchers.
Fixed --max-time duration values (e.g., 5s, 10m, 1h) being ignored instead of setting a session deadline.
Fixed omp plugin install --force failing with a dependency loop when replacing an existing pinned Git plugin source.
Fixed MCP tools receiving session image attachments as raw local:// URIs instead of resolving them to local filesystem paths.
Fixed Pyright LSP semantic requests hanging during startup.
Fixed Codex web search requests for GPT-5.6 Responses-Lite models.
Fixed custom model/provider configuration discovery to correctly load ~/.omp/agent/models.yaml when models.yml is absent.

[16.4.8] - 2026-07-12

Added

Added a predicate form to the browser run's wait() helper: wait(fn, { timeout?, interval? }) polls the function (sync or async) until truthy and resolves with that value, failing with a named timeout error (deadline clamped under the cell budget so it always beats the opaque whole-cell timeout) instead of Bun's sleep expects a number or a whole-cell stall from in-page polling Promises; both wait forms now register in the stall diagnosis of cell timeouts
Added --reasoning-slide-model and --reasoning-slide-turns to switch a running agent from its initial model after a fixed number of completed assistant turns
Added --reasoning-slide-plan (with --reasoning-slide-plan-at) to steer a hidden deep-planning nudge into the run before the reasoning slide; the switch is held until a substantial plan turn actually lands (bounded by a grace window) and the nudge is scrubbed from the LLM context at the switch so the fast model inherits only the produced plan
Added --reasoning-slide-on-action to trigger the reasoning slide at the first completed turn that ran an edit/write tool instead of a fixed turn count (bash is excluded — it doubles as exploration)

Changed

Replaced the Alt+P / /switch temporary model selector's fullscreen /models hub with a compact full-width floating overlay anchored above the editor (~40% of the terminal height): just the searchable model list — no provider sidebar or role management — with the session's active model highlighted and preselected
Improved tab recovery after timeouts by automatically clearing pending navigation and JS dialogs
Made tab.goto navigation failures catchable with a named error instead of triggering a whole-cell timeout
Made tab.evaluate run in the page's main JavaScript world so page-defined globals are available without a directive
Enhanced cell timeout messages to include identification of stalled operations and blocking JS dialogs
Browser run on a tab the supervisor force-killed now reports the kill reason instead of a bare "not alive"
Refined agent workflow to prioritize smoke testing and reduce mandatory upfront test generation

Fixed

Fixed the eval tool's status-event tree truncating from the bottom: the newest log() progress lines were hidden behind an … N more marker while the oldest stayed visible; the tree now shows a tail window behind an … N earlier marker, and the expanded view widens to the viewport instead of a fixed 10 events
Fixed the //!world=main directive being silently ignored for string expressions passed to raw Puppeteer evaluation APIs
Fixed tab reuse issues where hung navigation or unhandled modals would cause initialization to stall and trigger a force-kill
Improved search reliability for Perplexity provider by forcing retrieval for all queries
Fixed JS eval cells losing top-level function and var declarations across cells when the defining cell contained top-level await — the async wrapper scoped them to the cell's IIFE instead of publishing them to the worker global

[16.4.7] - 2026-07-12

Added

Enabled Home and End keyboard navigation in the model browser
Added a c hotkey in the plan-review overlay that copies the current reviewed plan markdown to the system clipboard, including in-overlay edits.

Changed

Streamlined list view styling by removing inline model role chips from row entries
Reworked /models hub selection visuals: the background highlight band is reserved for mouse hover, the keyboard position is a cursor glyph drawn only in the pane that owns the arrow keys, and the sidebar's active scope renders as a bold accent label
Removed the redundant "login" label from inactive (locked) provider entries in the Model Hub sidebar

Fixed

Fixed PageUp/PageDown in the model browser wrapping past the list edges instead of clamping
Fixed the hover highlight sticking to the last hovered model row when the pointer moved into the provider sidebar

[16.4.6] - 2026-07-12

Added

Added invalidate action to the usage command to clear cached usage reports
Added model-oriented keys and wildcard entries to retry.fallbackChains: a provider/model-id key attaches a fallback chain to that exact model, a provider/* key covers every current or future model of a provider, and a provider/* chain entry keeps the failing model's id while swapping the provider (google-antigravity/x → google/x) — so fallbacks survive role and model reassignments without config edits. Keys resolve by specificity: exact model, then provider wildcard, then role, then default.
Added fallback-chain editing to the /models Roles view: each role's retry.fallbackChains entries render as indented rows beneath it, f picks a fallback model to append, Enter on an entry replaces it, x/backspace removes it, and [/] (or shift+↑/↓) reorder the chain.
Added model-keyed fallback management to the /models Roles view: model and provider/* chains render as a separate section below the roles (divider + "+ New fallback…" row for creating one by picking the protected model, then keying it by model or provider), with the same replace/remove/reorder editing as role chains; the model strip gains fallbacks:<model> and fallbacks:<provider>/* chips as shortcuts.
Added /queue <message> plus -> / => composer shorthand for follow-up messages that wait until the agent yields. The shorthand opens a dim Queueing header and splits sequential numeric, Roman-numeral, or alphabetic lists into separately highlighted queue entries.
Added per-model TPS/TTFT tracking: every completed assistant turn folds its timing into recency-weighted aggregates in ~/.omp/agent.db, and the /models browser shows measured speed — a right-aligned 118t/s column on wide terminals (plus TTFT, e.g. 0.9s 118t/s, when wider) and ~118t/s · 0.9s ttft facts in the selection detail line — with no dependency on the omp stats session scan.

Changed

Retain completed and abandoned tasks in session history for improved context on resume
Changed the Model Hub retry-fallback strip chip to append the model to the default fallback chain instead of prepending it, matching the chain-building order of the Roles view (already-registered models are a no-op).
Changed per-model perf recording (recordModelPerf) to be deferred like prompt history: samples are batched and written to agent.db in one transaction ~100ms later, keeping SQLite writes off the turn-completion hot path.

Fixed

Fixed failure to trigger model fallback when the retry budget is exhausted by credential rotation
Fixed uncontrollable mouse-wheel scrolling in the /models hub: the wheel moved the selection (one step per wheel event, so a single trackpad flick skipped many rows) and wrapped from the bottom back to the top. Wheel scrolling now pans the list viewport only, clamps at the ends, and leaves the selection where it is; keyboard navigation still scrolls the selection into view. Likewise, the wheel over the provider sidebar no longer switches the active scope (or triggers provider refreshes) — it just scrolls the sidebar.
Fixed TPS being inflated several-fold when a provider hides reasoning tokens until late in the stream (e.g. google/gemini-3.5 vs google-vertex/gemini-3.5 reporting 648 vs 186 TPS for identical durations): omp bench, the per-turn usage row, and the /models perf aggregates now measure tokens/sec over the total request duration instead of the post-TTFT decode window, matching omp stats. Stored perf aggregates are purged and re-backfilled from stats history with the corrected math on first launch.
Fixed the model-perf stats.db backfill freezing the TUI (~30s on multi-million-row stats databases) when /models triggered it: the import is now fire-and-forget, walks the newest rows in small chunks with event-loop yields between them, and is bounded to 90 days / 256 newest samples per model — beyond either bound the recency decay would erase the contribution anyway.
Fixed compiled release binaries bundling fastembed and baking the build-machine @anush008/tokenizers path; native runtime dependencies now stay external for every compiled build path so Mnemopi resolves its on-demand install instead. (#5195)
Fixed /btw side-channel turns on Codex models such as gpt-5.6-luna by preserving the session websocket preference instead of forcing SSE, and made Esc dismiss the active /btw panel before interrupting loop/maintenance work. (#5213)
Fixed the Model Hub role-assignment strip hiding the selected chip once the row overflowed; the strip now scrolls horizontally, truncating passed chips behind a leading ellipsis so the selection (plus one chip of lookahead) stays visible.
Fixed mouse hover and clicks in the /models Roles view landing one row above the pointer (the row mapping subtracted the status row twice).
Fixed model search keeping the most-recently-used model on top of the results: match quality now ranks first (an exact gpt-5.5 beats the active gpt-5.6-sol), with MRU order only breaking ties between equally good matches.

[16.4.5] - 2026-07-11

Breaking Changes

Reworked the task tool wire schema: moved the top-level agent field into individual task items, renamed assignment to task and id to name, and removed the role and description fields. UI labels are now automatically generated from the task text.

Added

Introduced a fullscreen, mouse-supported Model Hub (via /model) featuring a sidebar of scopes, metadata-aligned model tables, inline role/thinking assignment strips, custom role creation, quick-switch cycle editing, and manual provider refreshing.
Added a /pause command to freeze all active agents (main, subagents, and advisor) at their next safe step, allowing manual repository edits mid-run before resuming.
Added support for per-ID bulk conflict directives via write({ path: "conflict://*", content: "..." }) to resolve multiple conflicts in a single call.
Added auto as a valid thinking-level in agent frontmatter, which is now the default for the bundled task subagent.
Added rich, interactive, fixed-height ask dialogs featuring question tabs, option previews, notes, and multi-select toggles.

Changed

Redesigned OAuth logins to run inside a cancellable dialog (aborted via Esc) rather than an inescapable pairing prompt.
Reworked the subagent soft request budget to gracefully steer child agents to wrap up and yield partial findings instead of silently terminating them, and raised the default budget to 200 requests.
Updated task rendering to retain the agent type badge on live progress and finished result rows.

Fixed

Fixed issues where in-flight tool calls or task blocks would disappear from the chat during mid-turn transcript rebuilds or when background jobs settled.
Fixed mixed blocking and non-blocking task batches degrading all spawns to synchronous execution; execution mode is now correctly handled per item.
Fixed budget-stopped subagents becoming unreachable, allowing them to remain adopted and resumable via irc with full context.
Fixed code duplication in write conflict:// when models pasted lines adjacent to the marker block.
Fixed visible per-keystroke lag when searching in the /resume session picker by caching search targets and debouncing SQLite lookups.
Fixed compiled Linux binary extension loading failures related to bundled web-search header generation data paths.
Fixed job list and empty-poll snapshots returning empty output, ensuring running subagents without backing jobs are properly listed.
Fixed agents getting stuck waiting for messages from peers that have already stopped running.
Fixed compiled Linux binary extension loading when bundled web-search header generation cannot read header-generator data files from the build-time path. (#5178)
Fixed plugin custom tool loading to skip and report invalid feature entries instead of crashing startup when a plugin dependency tree leaves one feature unresolved. (#5189)

[16.4.4] - 2026-07-11

Changed

Optimized session title generation and auto-thinking classification for sub-billion-parameter tiny models (such as LFM2) by rewriting system prompts, improving input truncation to preserve message context, and unifying preprocessing to filter out noise like ANSI codes, XML tags, and long commit hashes.

Fixed

Fixed an issue where the Windows binary exited silently without running the CLI, which also caused omp update to roll back.
Fixed native Windows binary compatibility on older Windows 10 CPUs by building the omp-windows-x64.exe release asset with a baseline x64 runtime instead of AVX2. (#5172)
Fixed GenerateImage rejecting OpenAI Codex-compatible proxy bearer keys when the token does not expose a chatgpt-account-id. (#5174)
Fixed context promotion documentation to accurately reflect the contextPromotionTarget runtime behavior and contextPromotion.enabled default. (#5163)

[16.4.3] - 2026-07-11

Added

Added /vibe mode, allowing the model to act as a director driving persistent background worker sessions (fast and good tiers) with dedicated session tools (vibe_spawn, vibe_send, vibe_wait, vibe_kill, vibe_list) and a live TUI "TV wall" showing active worker activity, tool traces, and streamed output.
Added multiple credential-free web search providers (Google, Bing, Yahoo, Startpage, Ecosia, Mojeek) with stealth-browser escalation, bot challenge detection, and recency filters, alongside a parallel public ("Public Web") provider that aggregates and deduplicates results across all engines.
Added PCRE2 fallback support for grep lookaround and backreferences when the default Rust regex engine rejects a pattern.
Added a helpful stderr hint when launching omp acp from an interactive terminal to clarify that the command communicates via JSON-RPC over stdout.

Changed

Reduced browser action timeout from 15s to 8s to improve agent iteration speed.
Refined agent delegation logic to prioritize top-level planning by the primary agent, discourage single-agent delegation, and handle prerequisite work inline.
Optimized credential-free web search engine ordering and routing, prioritizing Startpage and Ecosia, and using randomized desktop Chrome profiles with stealth-browser escalation for blocked requests.

Fixed

Fixed advisor config preserving an explicit empty tool list so /advisor config can disable all advisor tools. (#5155)
Fixed npm bundle generation failing on Linux with E2BIG by building dist/cli.js in-process via Bun.build instead of passing the embedded docs payload as a CLI --define argument.
Fixed compiled-binary extensions failing to load native .node FFI dependencies by resolving platform-specific packages against the extension's own node_modules.
Fixed a hang in omp search and omp q CLI commands by ensuring the AuthStorage connection is properly closed upon completion.
Fixed write blocking for the full LSP diagnostics poll by deferring slow diagnostics to a late-diagnostics channel.
Fixed multiple browser and puppeteer tool issues, including locator action timeouts, missing console/print output buffering, missing fill() method on element handles, and incorrect viewport screenshots on multi-tab setups.
Improved browser selector error diagnostics to report match counts and abort early on empty matches instead of waiting for the full timeout.
Fixed silent failures, duplicate error messages, and unhandled provider errors in ACP mode when errors occurred before streaming assistant text.
Fixed glob reporting contradictory "no files found" messages on timeout by explicitly stating the scan was incomplete.
Fixed read adding unwanted context padding to raw range selectors (e.g., raw:31-31).
Fixed first-run interactive startup rendering the entire changelog when the last-seen marker is missing or unreadable (#5135).
Fixed empty local-model stop responses exhausting retries without displaying a user-visible error (#5128).
Fixed serialization of BigInt values in tool arguments during session compaction.
Fixed GPT-5.6 over-delegating work by centralizing task fan-out and concurrency policy in the system prompt.
Fixed session title generation including leaked thinking markup from OpenAI-compatible endpoints (#5122).
Fixed bare skill:// path-only resolution for bash, grep, and glob to resolve to the skill directory instead of SKILL.md (#5087).
Fixed macOS stdio MCP servers missing Apple Events TCC prompts by spawning MCP children via the correct Bun.spawn overload (#5085).
Fixed the /move directory picker drawing a narrow 68-column frame inside wider overlays (#5067).
Fixed snapcompact inline imaging for GitHub Copilot Business and Enterprise models (#4779).
Fixed omp commit agent sessions to ensure valid proposals are committed before teardown, handle missing host outputs with non-zero exits, and prevent forcing GPG_TTY on signing-enabled repositories (#4794).

Removed

Removed the bundled plan subagent from available task agents.

[16.4.2] - 2026-07-10

Fixed

Fixed an issue where BigInt values in tool arguments failed to serialize during session compaction.
Resolved an issue where GPT-5.6 over-delegated tasks by refining task fan-out and concurrency policies in the system prompt.
Fixed a race condition in concurrent MCP OAuth token refreshes across processes, ensuring rotating refresh tokens are only refreshed once and preventing stale token errors from clearing valid credentials.

[16.4.1] - 2026-07-10

Changed

Reduced agent bias against large diffs and refactors in advisor prompts
Updated advisor blocker criteria to prioritize explicit user instructions over plan size

Fixed

Fixed MCP OAuth dynamic client registration omitting discovered scopes on the RFC 7591 registration body. Providers such as Clerk bind DCR-created clients to only the scopes declared at registration, then reject the subsequent authorize request when it asks for openid (from scopes_supported). Registration now includes config.scopes when present, matching Claude Code and the scopes already sent on authorize.

[16.4.0] - 2026-07-10

Breaking Changes

Renamed the bundled agent explore to scout, including its configuration keys, prompt files, and task definitions. Any configurations, allowlists, or invocations referencing explore must now use scout.
Changed the public selectLaunchAdapter() result from DapResolvedAdapter | null to LaunchAdapterSelection; callers must handle adapter, unavailable, and none outcomes.

Added

Added a native, first-class max thinking tier for supported models, including a new thinkingBudgets.max configuration setting, support in CLI flags (--thinking, :max model suffixes), and terminal theme customization (thinkingMax border color and icons).

Fixed

Fixed Go debug launches falling back to native debuggers when Delve is unavailable; nested modules and go.work workspaces now resolve local Delve adapters before PATH, newly installed adapters are detected without restart, and missing adapter errors include install or configuration guidance. (#5037)
Fixed a memory leak (large retained JavaScriptCore heaps) in the TUI during session transcript rebuilds and refreshes by properly handling snapcompact archive image frames.
Fixed a crash in interactive TUI sessions (Cannot set cwd while another same-realm JS runtime is running) when the JS evaluation worker falls back to the in-process inline path.
Fixed compaction aborting when Amazon Bedrock credential resolution fails, ensuring it now falls back to trying an authenticated model.
Improved OpenAI prompt cache hit rates for full-context forks by persisting inherited provider prompt-cache keys separately from session IDs, and added a --prompt-cache-key flag for explicit cache affinity.
Fixed Codex advisor requests incorrectly using local session labels as provider session IDs, switching to stable UUIDv7 provider identities.
Fixed macOS stdio MCP servers launching in detached sessions, allowing tools like xcrun mcpbridge to successfully trigger TCC Apple Events permission prompts.
Fixed the ask tool timeout behavior to automatically select the recommended option if the UI selector does not settle.
Fixed LSP workspace diagnostics for Go workspaces to correctly recognize go.work roots and include all specified modules in go build package patterns.
Fixed interactive OAuth login (/login xai-oauth) delaying success messages; credentials are now reported immediately while model metadata refreshes in the background.
Fixed a crash in the Windows bash tool when a timeout occurs while a piped command is streaming output.
Fixed subagent yield tool calls being discarded when a soft request budget aborts the assistant turn before the yield event completes.
Fixed --tools filtering in interactive sessions incorrectly disabling deferred MCP tools from configured servers.
Fixed kept-alive task subagents entering infinite provider-call loops after an IRC wake and terminal yield.

[16.3.15] - 2026-07-09

Changed

Integrated testing guidance directly into the main system prompt for improved workflow cohesion
Moved testing guidance into the main system prompt and removed the bundled Tester subagent.

[16.3.14] - 2026-07-09

Fixed

Fixed issue where unfinalized tool blocks could incorrectly pin the live-region scroll seam
Improved rendering of raw thinking blocks by stripping empty HTML comment noise
Fixed display of thinking blocks consisting entirely of hidden comment noise
Fixed gpt-5.6 reasoning summaries rendering literal  sentinel lines in thinking blocks; empty HTML comments (and the unterminated <!-- tail while streaming) are now dropped from the thinking display, and blocks reduced to pure comment noise are hidden entirely.

[16.3.13] - 2026-07-09

Fixed

Fixed read and grep treating empty optional selector fields emitted by models as invalid selectors instead of behaving like omitted selectors. (#4879)
Fixed grep explicit line selectors on directory searches so they filter each matched file by line number instead of aborting with a single-file-only error (#4898).
Fixed Read tool previews dropping explicit selector arguments, so line ranges and raw modifiers render in terminal read call titles again (#4899).
Fixed named profiles dropping default user keybindings from ~/.omp/agent/keybindings.*; profile keybindings now inherit those defaults and override only the keys they define (#4867).
Fixed pi extensions calling ctx.ui.addAutocompleteProvider(...) crashing at load with TypeError: ... is not a function — a failure that, for extensions guarding init in one try/catch (e.g. @ff-labs/pi-fff), aborted the extension's entire initialization. ExtensionUIContext now implements pi's autocomplete-provider API: interactive mode stacks each registered factory on top of the built-in editor provider (re-applied on every slash-command refresh, with throwing or malformed factories skipped), while RPC/ACP/headless contexts accept the factory as a no-op. (#4919)
Fixed bundled reviewer and plan subagents to inherit their model roles' explicit thinking effort suffixes instead of pinning high (#4761).
Built-in provider model discovery now refreshes an expired stored OAuth credential before an online refresh needs it, instead of silently skipping the provider. The refresh is scoped to the providers actually being discovered (refreshProvider cannot rotate unrelated credentials), fires under online-if-uncached only when the model manager will actually fetch, and offline discovery stays peek-only (#4893).
Fixed Escape during an active TUI prompt requiring a second press before canceling; the first Escape now aborts the streaming turn immediately. (#4921)
Fixed the streamed write tool's collapsed pending tail preview leaving stale rows above the first partial-result frame in the TUI; the first result now replays the viewport like the SSH placeholder seam already did (#4477)
Fixed first-run setup ignoring a pre-seeded config.yaml: the settings loader now treats config.yml and config.yaml as equivalent existing main config files, writes back to the existing extension, and only creates canonical config.yml for fresh installs. (#4914)
Fixed extension sendUserMessage() without deliverAs surfacing AgentBusyError during active streams; omitted deliverAs now queues a steer through the normal prompt flow, and ACP/RPC skill-command prompts queue while streaming (RPC honors the prompt command's streamingBehavior, defaulting to steer) (#4923).

[16.3.12] - 2026-07-08

Added

Typing #<number> (e.g. #3164) in the prompt now offers PR and Issue autocomplete candidates that rewrite to the pr:///issue:// internal URL, resolved from the current repo's git remote via the existing read tool → InternalUrlRouter → gh pipeline. Naming the type (pr #3164 / issue #3164) constrains the candidates to that kind, and embedded hashes like owner/repo#N, foo#N, or URL fragments are left untouched (#3218)

Changed

Memoized non-message token totals (system prompt, tool schemas, skills) so the per-turn compaction and context-threshold paths recompute them at most once per input change instead of on every call. getContextBreakdown and #estimateStoredContextTokens previously re-tokenized the system prompt and every tool's wire schema (per-tool JSON.stringify) several times per turn over inputs that change at most once per turn.

Fixed

Improved handling of unawaited promises in JS eval cells to prevent process crashes
Added warning logs for unhandled rejections originating from finished eval cells
Improved advisor robustness by blocking exhausted accounts during consecutive turn failures
Fixed advisor turns hammering the same usage-limited account: a failed advisor turn now marks the exhausted credential blocked (with the provider's retry hint and usage-report reset time), so the next retry rotates to a sibling instead of re-picking the blocked account every few seconds. Previously the in-stream auth retry rotated within a request but never blocked the last failing credential, and the advisor loop — unlike the primary retry pipeline — never called markUsageLimitReached.
Added the account key to the codex-auto-reset: skipped debug log so skip reasons (e.g. weekly-not-exhausted) can be attributed to the evaluated account.
Fixed unawaited promise rejections in JS eval cells crashing the session: a floating rejection now fails the owning cell run (Unhandled rejection (missing await?): …) instead of escaping to the global unhandledRejection handler, which printed [Unhandled Rejection] and killed the process (inline fallback) or tore down the eval worker (dedicated worker). Rejections surfacing after a cell settled are downgraded to a warn log attributed to the finished cell.
Fixed project .omp/RULES.md sticky rules being shadowed by user ~/.omp/agent/RULES.md rules with the same synthesized RULES name, so both user and project sticky rules now inject (#4739).
Fixed bash internal-URL expansion so unresolved literal memory:// / skill:// text stays verbatim instead of aborting command execution (#4737).
Fixed agent://<id> (and the output() eval helper) failing with Not found for a subagent spawned by another subagent (any spawn chain 2+ levels deep). artifactsDirsFromRegistry scanned only each ref's adopted (root-wide) ArtifactManager dir, but a subagent's own children are written one level deeper under its sessionFile-derived dir — so a live, addressable nested peer's output was unresolvable. The resolver now collects both candidate dirs per registered agent. (#4650)
Fixed plan mode to document local:// artifacts as writable session-local planning files and to carry every pre-approval local:// artifact into the fresh session created by Approve and Execute.
Fixed the browser tool failing to launch Microsoft Edge-only Windows installs with Puppeteer's empty Code: 0 launch error by keeping Edge's required --enable-automation default while preserving Chrome/Chromium stealth launch defaults.
Fixed bash/tool command environments inheriting Bun-autoloaded launch .env.local values, so nested apps can load their own dotenv values without parent deployment variables taking precedence. (#4723)
Fixed legacy plugin validation for extension graphs that import JSON with with { type: "json" }, leaving JSON files on Bun's native loader instead of parsing them as JavaScript (#4687).
Fixed pasted terminal transcripts beginning with a shell prompt ($ ...) being mistaken for local Python shortcuts instead of being submitted as normal prompts (#4678).
Fixed wrapped OAuth copy-URL rows corrupting on paste: continuation chunks no longer carry a leading indent, so a multi-row terminal selection reassembles to the exact authorize URL (browsers strip newlines on paste but preserve or percent-encode embedded spaces, which previously corrupted the URL at every chunk boundary).
Fixed Windows browser-launch failures being unobservable: the opener now uses %SystemRoot%-resolved PowerShell Start-Process (via -EncodedCommand) instead of rundll32, which exits 0 unconditionally. Failures ShellExecute itself reports — missing target, no handler executable, access denied — now surface as non-zero exits and are logged; the encoded payload also keeps OAuth query strings (&-bearing) opaque to shell metacharacter parsing.
Fixed system prompt date rendering to use the host local calendar date instead of UTC.
Fixed bash tool timeout: 0 so it disables the command deadline instead of falling back to the minimum timeout.
Fixed read and grep refusing to access filesystem paths whose names end in a selector-shaped suffix (e.g. test:1-2, log:raw) by preferring a literal match over the trailing :<selector> peel when the raw path exists on disk (#4618).
Fixed wrapped Edit-diff rows leaking inverse video into the result card's right-edge padding: a row that broke inside an intra-line highlight left inverse active at the row end, so the frame padding after it rendered as a default-foreground block (#4616 by @chan1103)
Fixed Edit-diff continuation rows escaping into the line-number column when the row's gutter was left-padded (line number narrower than the widest in the diff) or blanked by the gutter dedup (the bare + row of a single-line replacement); such rows now wrap behind a continuation gutter, while body lines that merely start with | keep wrapping generically (#4616 by @chan1103)
Fixed live advisors continuing to use a stale modelRoles.advisor selection after /model changed the advisor model. (#4612)
Fixed Claude plugin slash commands and skills silently vanishing when the plugin manifest declares commands/slash-commands/skills as a JSON array — the shape the Claude plugins reference documents and real plugins like addyosmani/agent-skills ship. resolvePluginDir in packages/coding-agent/src/discovery/claude-plugins.ts typed those fields as string and dropped array values on the floor; it now normalizes both shapes, loads every in-root entry, and reports one out-of-plugin-root warning per bad entry. The resolver also now honours Claude's per-field merge semantic — skills adds to the default skills/ scan; commands/slash-commands replace the default commands/ — so plugins like {"skills":["./extra-skills"]} no longer lose their default skills/ folder while {"commands":["./admin"]} still replaces commands/ as documented. (#4609)
Fixed macOS Backspace on empty search not deleting sessions in the /resume picker; Fn+Backspace terminals that deliver \x7f instead of \e[3~ now reach the delete confirmation dialog. (#4580 by @JagravNaik)
Fixed /rename title arguments treating # prompt-action tokens as autocomplete triggers instead of literal session title text. (#4600)
Fixed empty session .jsonl files accumulating in ~/.omp/agent/sessions/<cwd>/ after a draft-then-clear exit cycle. SessionManager.saveDraft(text) materializes the session file so the draft sidecar has a parent; a subsequent saveDraft("") unlinked the sidecar but left the metadata-only JSONL behind (title slot + session header + startup selector entries, ~500–750 B), and #shouldHaveSessionFile() could no longer prune it once #fileIsCurrent/#forceFileCreation were latched. SessionManager.close() now drops only draft-owned metadata-only sessions with no saved draft sidecar to reattach to, while keeping real conversations, meaningful non-message entries such as handoff custom messages, explicit ensureOnDisk() sessions, drafts still pending for --resume, and never-materialized sessions untouched (#4571).
Fixed the advisor being disabled for the entire session when the advisor role resolves to a reasoning model that exposes no controllable effort surface (Devin devin/glm-5-2*: reasoning: true, thinking: undefined — Cascade routes by sibling model id rather than a wire param). #resolveAdvisorRuntimeDescriptors in packages/coding-agent/src/session/agent-session.ts used to hardcode ThinkingLevel.Medium, which tripped requireSupportedEffort on the first advisor prompt with Thinking effort medium is not supported by devin/glm-5-2. Supported efforts: (empty list). The advisor descriptor now clamps the requested effort against the resolved model via resolveThinkingLevelForModel and forwards no explicit effort when the model has no controllable efforts — matching the auto-path fix (clampAutoThinkingEffort) and the Autonomous Memory stage fix (clampThinkingLevelForModel). Explicit :off still disables reasoning, and models that support medium (e.g. Anthropic) keep receiving it (#4579).
Fixed legacy extension plugin validation failing with Export named 'calculateCost' not found in module '.../legacy-pi-ai-shim.ts' when the extension imports calculateCost (or modelsAreEqual / getBundledProviders) from @oh-my-pi/pi-ai. Those symbols were relocated to @oh-my-pi/pi-catalog/models during the catalog split but were never bridged back through the legacy pi-ai root shim; the shim now re-exports them alongside the existing getModel / getModels aliases so plugins written against pre-split pi-ai load again (#4584).
Fixed legacy extension plugin validation failing with Export named 'calculateCost' not found in module '.../legacy-pi-ai-shim.ts' when the extension imports relocated catalog symbols such as calculateCost, modelsAreEqual, getBundledProviders, getBundledModel, or getBundledModels from @oh-my-pi/pi-ai. Those symbols were relocated to @oh-my-pi/pi-catalog/models during the catalog split but were never bridged back through the legacy pi-ai root shim; the shim now re-exports them alongside the existing getModel / getModels aliases so plugins written against pre-split pi-ai load again (#4584).
Fixed legacy pi extension imports of DefaultResourceLoader from @mariozechner/pi-coding-agent / @earendil-works/pi-coding-agent by adding a compatibility loader shim that translates resourceLoader into OMP's native session discovery options. (#4567)
Fixed legacy Pi extension reloads on POSIX so loadLegacyPiModule imports the entry through a cache-busting filesystem path, refreshes load-time graph hooks when reloads add new modules, and threads the current load's ?mtime tag through the extension source graph — relative ./helper.ts siblings, #alias/* package-imports, extension-local bare dependency entries, and their relative children all rekey per reload, so same-process re-imports pick up edits across the whole graph. (#4565)
Fixed bash tool pipeline execution preserving stale upstream output when the final stage was a stripped head/tail limiter; the tool now runs the command as written so seq 1 5 | head -n2 returns only 1 and 2. (#4562)
Fixed the status-line token-rate segment rendering as <number>/s, which Ghostty auto-detected as a hyperlink on Ctrl+hover. (#4541)
Fixed retry fallback model recovery by exposing retry.fallbackChains in /settings, adding a /model action to assign the selected default fallback model, and clearing a selected model's retry cooldown marker on manual model switches. (#4533)
Fixed /handoff and auto-handoff skipping extension lifecycle hooks by emitting cancellable session_before_switch hooks and a session_switch with reason: "handoff" after the replacement session is ready (#4434).
Fixed TTSR stream interrupts so only the tool call whose stream matched a rule receives the rule-named abort result; sibling tool-call placeholders now use a neutral abort reason (#2783).

[16.3.11] - 2026-07-06

Changed

Improved session title generation reliability by moving to marker-based parsing for all models

Fixed

Fixed session titles occasionally showing raw {"title": "..."} JSON. Online title generation now always uses the <title>...</title> marker prompt instead of a forced set_title tool call — hosts that ignored or rejected forced tool_choice echoed the prompt's JSON example verbatim as the title — and JSON-shaped responses (bare, code-fenced, marker-wrapped, or truncated) are unwrapped to the bare title.
Fixed Linux startup prompt construction to read the CPU model from /proc/cpuinfo instead of os.cpus(), avoiding per-core sysfs frequency probes on many-core hosts (#4712).
Fixed llama.cpp model discovery to honor per-model architecture.input_modalities from /v1/models, so router presets that advertise image input are no longer treated as text-only (#4719).

[16.3.10] - 2026-07-06

Changed

Limited subagent HUD display to 8 items with a summary for additional running subagents
Improved TUI performance by coalescing agent registry and observer UI updates

Fixed

Fixed high-subagent sessions overwhelming the TUI by bounding the running-subagent HUD and coalescing subagent progress repaint bursts.
Fixed browser run cleanup crashing or wedging the whole omp session: run-end and tab-close aborts could reject fire-and-forget wait()/facade/tab promises with no consumer, and the global unhandled-rejection handler then exited the process, killing every subagent sharing it. Run-scoped promises are now observed at creation and routine browser teardown aborts are downgraded to log lines (#4499, #4672).
Reduced CPU use while many task subagents stream progress by disabling the obsolete task partial-result spinner repaint loop (#4424).
Capped collapsed nested subagent trees at the same per-level agent limit as top-level task rendering (failures prioritized, elided rows summarized), so deep many-subagent sessions no longer render unbounded progress trees on every repaint.
Fixed skill card headers to render a single space between the skill tag and skill name (#4662).
Fixed get_session_stats RPC responses to include context-window usage so RPC clients can render context meters.
Fixed startup of cached llama.cpp vision models so the initial default/restored model refreshes /props metadata before the session exposes it as text-only.
Fixed IRC-woken yielded subagents skipping empty-stop retry because stale yield-termination state carried into the wake turn (#4658).
Fixed irc wait skipping replies that arrived between wait calls by draining pending IRC asides before honoring queued-interrupt aborts (#4657).

[16.3.9] - 2026-07-06

Added

Added a --file flag to the say command to read input text directly from files.
Enabled streaming text synthesis in the say command for gapless, long-form audio generation.
Added voice selection validation to the say command.

Changed

Improved the say command to display the segment count and total duration upon completion.

Fixed

Fixed an issue where local llama.cpp vision models remained text-only after a model refresh, ensuring they are correctly recognized as image-capable when configured as the default or vision role.
Fixed omp commit split plans aborting when lock files (such as bun.lockb) were staged alongside their manifests by correctly pairing lock files with their corresponding commit groups and properly handling binary files during split execution.
Fixed skill loading to ensure that disabling a higher-priority provider does not drop same-named skills from enabled lower-priority providers.

[16.3.8] - 2026-07-05

Fixed

Fixed the browser tool silently launching without its Puppeteer stealth patch: the patch was keyed to puppeteer-core@25.1.0 while the resolved dependency had drifted to 25.3.0, so Bun skipped it and Chrome ran with the Runtime.enable automation tell re-enabled. Regenerated the stealth patch against 25.3.0 and pinned puppeteer-core so the exact-version patch cannot silently deactivate on future drift.

[16.3.7] - 2026-07-05

Added

Added support for reading full memory rows (working or episodic) via read memory://<id> under the mnemopi backend, returning a YAML-frontmatter header with metadata to prevent blind overwrites during edits.
Updated the URI grammar to support memory://root[/…] for file-backed summaries and memory://<memory-id> for specific mnemopi IDs.

Changed

Updated recall and memory_edit tool prompts to document truncation markers and require reading a memory ID before updating a truncated preview.
Parallelized resolution of system files (SYSTEM.md, APPEND_SYSTEM.md, and TITLE_SYSTEM.md) at startup to improve performance.
Optimized TUI performance and reduced CPU usage during streaming and live tool calls by scoping renders to changed subtrees, interning the working-message shimmer palette, and memoizing copy-selector preview highlights.
Loaded persisted Agent Hub subagents asynchronously to prevent blocking the TUI on directory walks.
Cached persisted message keys in AgentSession to avoid repeated branch walks on message completion.
Skipped TTSR delta buffering for text/thinking sources when no registered rules match.

Fixed

Fixed macOS Backspace behavior in the /resume picker for terminals delivering \x7f instead of \e[3~.
Fixed queued follow-up message rows leaking into native terminal scrollback during live repaints by anchoring the pending-messages container.
Fixed /rename title arguments treating # prompt-action tokens as autocomplete triggers instead of literal text.
Fixed empty session .jsonl files accumulating in the sessions directory after a draft-then-clear exit cycle.
Fixed advisor being disabled for the entire session when resolving to a reasoning model with no controllable effort surface (e.g., devin/glm-5-2*).
Fixed legacy extension plugin validation failures by re-exporting relocated catalog symbols (such as calculateCost, modelsAreEqual, and getBundledProviders) through the legacy pi-ai root shim.
Fixed legacy Pi extension imports of DefaultResourceLoader by adding a compatibility loader shim that translates resourceLoader into native session discovery options.
Fixed legacy Pi extension reloads on POSIX to ensure same-process re-imports pick up edits across the entire dependency graph.
Fixed bash tool pipeline execution preserving stale upstream output when the final stage was a stripped head or tail limiter.
Fixed the status-line token-rate segment rendering as a clickable hyperlink in Ghostty.
Fixed xAI web_search to prioritize xai-oauth credentials before falling back to xai API keys, and enforced result counts locally to avoid sending unsupported parameters.
Fixed token-usage badges disappearing on session resume for empty automated assistant turns.
Fixed /model search escaping the active provider tab and silently persisting a same-named model from a different provider as the default role.
Fixed Esc key behavior to immediately silence active TTS audio playback (such as Kokoro) once an assistant reply finishes streaming.
Fixed grep and bash tool guidance to instruct agents to use Rust-style patterns or grep -E instead of GNU BRE assumptions.
Fixed tool-call renderers crashing the TUI when providers send array/object path arguments before schema validation.
Fixed glob tool rejecting slash-only root searches (e.g., path:"/") with a documented error instead of returning empty results.
Fixed omp read hanging on PDFs whose inline-image binary payloads contain delimiter-looking bytes.
Fixed pi/<role> model resolution crashing on YAML list-valued modelRoles entries.
Fixed the snapcompact settings UI to expose silver16-bw and improved renderability warnings to report unsupported glyphs.
Fixed session and status usage totals to preserve provider-reported orchestration tokens separately from input and cache-hit buckets.
Fixed published .d.ts declaration files to be consumable under moduleResolution: "node16" | "nodenext" by rewriting relative specifiers to explicit .js extensions.
Fixed omitThinking settings propagation so settings-aware streams request hidden thinking summaries when explicitly enabled.
Preserved isolated branch-mode task output as a patch artifact when commitToBranch fails, surfacing the captured patch path through the evaluation failure message.
Fixed task-class subagents dropping unresolved explicit model-role selectors before startup.
Fixed large legacy snapcompact archives causing crashes on resume due to oversized archived frame payloads.
Fixed LSP diagnostics staleness by sending watched-file change notifications to running language servers before reading edit-time diagnostics.
Documented the bash tool timeout clamp (capped at 3600 seconds for async jobs) in the model-facing schema and prompt.
Fixed /fast on for custom OpenAI-compatible providers serving OpenAI models, and reported unsupported models as unavailable.
Fixed extension pasteToEditor and setEditorText prompt mutations leaving the editor visually stale by scheduling a repaint after mutations.
Fixed session title generation, commit-message generation, speech-enhancer rewrites, and classifiers silently truncating on non-reasoning-flagged models that still emit thinking output.
Fixed plan-mode "Approve and compact context" discarding queued operator turns and leaking internal plan-distillation prompts to extension hooks.
Fixed malformed pi.sendMessage custom-message payloads persisting bare session entries that crashed subsequent resumes.
Hid internal display: false session-update reminders from compact history and advisor transcripts.
Fixed idle recap crashes caused by side-channel stream malformations.
Applied WebSocket send backpressure with ordered drain retries to prevent unbounded buffer growth.
Capped docs.rs gunzip decompressed size at 256 MB to prevent zip-bomb out-of-memory crashes.
Deleted pre-created shell snapshot files when snapshot creation fails.
Aborted underlying MCP calls when proxy tool timeouts fire.
Surfaced unexpected JS eval worker exits via close listeners to prevent silent hangs.
Cached failed !command config resolutions and timed out extension dynamic model fetches after 15 seconds.

[16.3.6] - 2026-07-04

Changed

Hided abort labels for user-interruptions in the transcript to match silent abort behavior
Updated transcript streaming to declare settled rows for scrollback exactness: TranscriptContainer reports one boundary (finalized blocks plus the first live block's getTranscriptBlockSettledRows(), fed by markdown's frozen-token prefix — completed content blocks in a streaming reply render final and settle in full, so long replies and visible thinking reach terminal scrollback as exact bytes mid-stream). The heuristic commit machinery (deriveLiveCommitState append-only detection, stable-prefix ratchet, volatile cooldowns, rewrite floors), isTranscriptBlockCommitStable, and the per-renderer provisionalPendingPreview/provisionalPartialResult flags are all removed. Ephemeral block retraction (displaceable todo/job cards, IRC cards) is now gated on TranscriptContainer.isBlockUncommitted() — cards whose rows already entered scrollback seal in place as history instead of being removed.
Recovered auto-retry errors now compact in the transcript: once a retry succeeds, the superseded error rows (e.g. Anthropic 429 rate_limit_error during account rotation) render as a single dim note like rate-limited; switched account; retried across live view, rebuilds, resume, and subagent/collab transcript views, and are excluded from model context on session resume. Terminal (unrecovered) errors keep full error rendering. The persisted marker is retryRecovery on the assistant message and auto_retry_end success events carry additive recoveredErrors data.

Fixed

Fixed raw read ranges not contributing to edit seen-line provenance, so re-reading an anchor range with :raw now unblocks hashline edits without adding non-raw line prefixes.
Fixed replan-driven session title refresh updating the statusline but not the terminal window title: terminal-title updates now fire from the session-name-changed listener, so every setSessionName path (first-input titling, /rename, plan seeding, replan refresh) sets the OSC title consistently.
Fixed user-interrupt aborts rendering the persisted Interrupted by user label in assistant transcripts; replay and live views now suppress that redundant line again while preserving generic/custom abort labels.
Fixed streamed assistant text and thinking disappearing from terminal scrollback while a long turn ran, and Ctrl+O-expanded tool results showing up half-rendered above the viewport: everything that scrolls off the window now reaches native scrollback (exact bytes for settled content, a frozen what-you-saw snapshot for still-running blocks, repaired at most once when the block finalizes). Streaming eval/bash boxes no longer spray duplicated stale copies; user-initiated !/$ execution blocks report a finalization contract.
Fixed turn-ending provider errors rendering with a doubled blank gap above the Error: block (caller and error block each added a spacer).
Fixed the write tool renderer crashing when persisted runtime content is a truthy non-string value; rendering now coerces display content before Windows CR normalization. (#4495)
Fixed cmux-backend browser({action:"run"}) calls crashing the entire process with an unhandled rejection when the tab was released mid-run (e.g. a sibling subagent calling browser({action:"close", all:true}) or a session-scoped tab reap). runInTabWithSnapshot in tab-supervisor.ts creates a Promise.withResolvers() triple so releaseTab can signal in-flight runs, but the cmux branch used to await runCmuxCode(...) directly and never awaited the local promise. When releaseTab rejected that orphaned promise ("Tab ... was closed"), Bun surfaced it as an unhandled rejection and the top-level handler tore the whole session down, killing every other tab and subagent sharing it. Both backends now await the same promise (so pending.reject always has an attached handler AND the caller sees Tab "..." was closed immediately instead of blocking to the run's timeout), and a new pending.closeAc is composed into the cmux run's abort signal so wait(...), in-flight cmux socket calls, and the facade proxies unwind promptly when the tab is closed rather than leaking to their own timeout (#4499).

[16.3.5] - 2026-07-04

Fixed

Fixed Mnemopi auto-retention so protocol markers are stripped from embedding and FTS projections while stored transcripts remain readable. (#4395)
Fixed mnemopi auto-retain storing cumulative full-session transcripts on every retention interval; subsequent retains now store only newly completed user-turn suffixes. (#4396)
Fixed large artifact:// reads materializing entire MCP/tool artifacts before selector paging, preventing OOM crashes on unbounded raw reads and surfacing bounded read/search/copy guidance (#4482).
Fixed /mcp reauth and /mcp add dropping OAuth scopes advertised via WWW-Authenticate: Bearer scope="..." challenges and via protected-resource metadata (scopes_supported / scopes / scope), which caused tokens to be issued without the required scopes and reconnects to fail with insufficient_scope (#4467)
Fixed task-branch merges aborting the whole cherry-pick range when a single commit was empty against HEAD (redundant with an already-applied change, or 3-way merged to HEAD); the sequencer now --skips the empty commit and continues so later non-overlapping work still lands (#4438).
Fixed write/edit LSP diagnostics for TypeScript files outside any project root by suppressing project-resolution noise while preserving real syntax errors (#4401).
Fixed /mcp reauth against S256-only OAuth providers (Linear, and OAuth 2.1 flows generally) failing on Windows and other environments where the browser cannot auto-open. Two independent defects converged on Linear's The plain PKCE method is not allowed. Use S256 instead. error page: openPath invoked bare rundll32 and silently swallowed the resulting Executable not found in $PATH when the Windows machine PATH no longer referenced System32, and MCPAuthorizationLinkPrompt composed a Copy URL: line wider than the viewport that TUI#prepareLine then silently truncated — trimming the trailing code_challenge_method=S256 (RFC 7636 §4.3 treats a request with code_challenge and no method as plain). The MCP OAuth fallback, /login, setup wizard, auth-broker CLI, and login-dialog now advertise the short OAuthAuthInfo.launchUrl (loopback /launch route hosted by OAuthCallbackFlow) as the copy target; the OSC 8 hyperlink still carries the full URL for click-through; the MCP flow additionally stages the copy target on the clipboard via OSC 52; and the Windows opener now resolves rundll32.exe through %SystemRoot%\System32\, logging spawn failures and non-zero exits instead of dropping them (#4418).
Fixed lsp rename_file aborting when an LSP server returns duplicate byte-identical non-empty text edits for the same range, such as tsserver willRenameFiles edits through barrel re-exports (#4458).
Fixed Linux x64 PI_TINY_DEVICE=cuda tiny-model side runtimes missing ONNX Runtime CUDA provider binaries by repairing the onnxruntime-node CUDA sidecar install and preserving actionable CUDA diagnostics in omp tiny-models download output (#4475).
Improved reliability of edits when file snapshots share identical 16-bit hash tags
Fixed ACP terminal/create sending the bash tool's full shell line in command with no args, which broke spec-conformant clients that spawn command+args directly (no implicit shell) — any command containing a space, pipe, &&, redirect, or $(...) failed with ENOENT and the agent silently degraded to read-only tools. The bash tool now wraps the shell line before calling clientBridge.createTerminal, reusing the same shell binary + args the local bash-executor resolves via settings.getShellConfig() (Git Bash / bash.exe on Windows, $SHELL with sh fallback on POSIX) so bash semantics — $VAR, $(...), source, POSIX quoting, -l — are preserved on both platforms. (#4333)
Fixed inference worker subprocesses (TTS, STT, tiny-model, mnemopi embeddings) discarding stderr, which left every unexpected exit — most visibly the local Kokoro TTS worker's recurring exit code 7 crash loop — undiagnosable from the parent's logs. createWorkerSubprocess now pipes stderr without starting a live read while the worker is idle, then drains the stream after onExit, emits captured lines to logger.debug under an <exitLabel> stderr message, and keeps the last 16 KiB in a bounded ring that gets appended to the Error surfaced through onError. The exit surface is synchronized with the post-exit drain via SpawnedSubprocess.stderrDrained, so the full native trace shows up on the tts: worker error line without reintroducing event-loop liveness from unref'd workers. (#4324)
Fixed Windows session tail loss after atomic compaction rewrites by fencing append writers during full-file replacement and gating the atomic publish on a commitGuard that the storage backend checks synchronously before rename, so a concurrent flushSync (Ctrl+C / session-exit) is not overwritten by the stale body serialized before it ran. Covers post-compaction prompts, tool results, title changes, and exit diagnostics on the current JSONL path (#4338).

[16.3.4] - 2026-07-03

Fixed

Fixed omp usage hiding sibling-only limits such as Claude 7 Day (Fable) on accounts whose current report omitted that scoped bucket; the account now renders an explicit not reported row instead of looking like the usage refresh skipped the column.

[16.3.3] - 2026-07-02

Breaking Changes

Removed _input as a supported alias for the edit tool input field

Changed

Refined advisor note card visuals with a dedicated rail character and bold label headers
Deferred session_stop extension hooks and incomplete-todo reminders until all agent-owned background jobs (such as async bash or task spawns) are fully idle
Restyled the advisor note card so notes no longer blend into thinking output: bold label-tag header, heavier severity-tinted rail, and note body on the default text color instead of thinking-gray

Fixed

Fixed TUI loading animation sometimes failing to render during interactive mode status updates
Fixed race conditions and potential hangs during agent startup by improving process lifecycle management
Fixed RpcClient startup errors intermittently losing the child's stderr (e.g. the "Unknown provider" message) by waiting for the process to exit — and its stderr to fully drain — before reporting a failed start; also fixed an unhandled rejection when the agent process is killed before becoming ready
Improved reliability of file edits when snapshots share identical hash tags
Fixed terminal creation in ACP by properly wrapping shell commands and arguments, resolving execution failures (such as ENOENT errors) on Windows and POSIX platforms
Fixed inference worker subprocesses (such as TTS, STT, and embeddings) discarding stderr, ensuring unexpected crashes and exit traces are properly captured and surfaced in logs
Fixed potential session data loss on Windows during atomic compaction rewrites by preventing concurrent writes from overwriting the session file during exit or compaction
Fixed transcript scrollback boundaries to prevent duplicate rows from appearing when live blocks grow
Fixed macOS SSHFS mount detection to avoid redundant remounting attempts when the mountpoint is unavailable
Fixed SSH streamed placeholders and partial frames leaving stale rows in the TUI viewport and scrollback
Fixed Gemini web search to correctly honor the configured search model (providers.webSearchGeminiModel or GEMINI_SEARCH_MODEL) for both OAuth and Developer API grounding requests
Fixed omp install rejecting valid npm package specifications
Improved MCP OAuth reauthorization error messages when dynamic client registration is closed, directing users to configure client credentials
Fixed omp update -l to correctly support the plugin-update shorthand for upgrading marketplace plugins
Fixed /advisor on command to correctly rebuild the advisor runtime and bind to the newly configured model
Fixed edit tool failures on large source files after a structural-summary read by automatically merging unseen anchor lines into the snapshot
Fixed potential hangs and process leaks in the Debug Adapter Protocol (DAP) client by introducing write timeouts and ensuring detached adapter processes are terminated on connection failures
Fixed status-line GitHub PR lookup hangs by enforcing a command timeout
Fixed potential pipe-buffer deadlocks during plugin installation, uninstallation, and updates by concurrently draining stdout and stderr
Fixed SSH tool hangs on unreachable hosts by enforcing a 30-second timeout on pre-command connection and host probes
Fixed models.yml schema validation to surface warnings for invalid custom provider configurations instead of silently ignoring them
Fixed potential network hangs in omp update, Hindsight recall, and Smithery registry lookups by adding fetch timeouts
Reduced TUI CPU overhead during streaming and idle waits by dropping the redundant pre-render on every session event, iterating shimmer text in place instead of allocating a code-point array per animation frame, and coalescing the live-tool spinner render cadence to its glyph-advance rate (#4353).

[16.3.2] - 2026-07-02

Breaking Changes

Changed search tool paths parameter to a single semicolon-delimited path string parameter

Changed

Reduced extension startup cost, especially on Windows, by reading each extension source-graph module from disk once per load instead of twice (the graph scan now feeds the load-time rewrite hook) (#4196).

Fixed

Fixed ALL-CAPS acronyms (e.g. CNPG, ETL, JWT) being lowered to title case in auto-generated session titles. reconcileTitleCasing (packages/coding-agent/src/tiny/text.ts) now maps ALL-CAPS source tokens into an acronyms table and restores them when the model produces a title-cased artifact (Cnpg), while still declining restoration on shouty input (FIX the BUG NOW, ALL ERROR HANDLING) via a consecutive-ALL-CAPS heuristic. Title prompts also instruct the model to preserve ALL-CAPS acronyms verbatim. (#4220)
Fixed cold-start --model resolution for extension providers whose catalogs come only from fetchDynamicModels, so fresh cached runtime models are available before session startup falls back or hard-fails. (#4216)
Fixed plugin and legacy extension discovery repeatedly re-reading plugin manifests and walking extension node_modules by caching results until plugin cache invalidation. (#4197)
Fixed discoverExtensionPaths invoking every registered extension-module provider (claude, codex, gemini, opencode) on startup and discarding all non-native results. The extension-module capability is now loaded with providers: ["native"], skipping four foreign directory walks per session — noticeable on Windows where the walks are slowest (#4198).
Fixed /move overlay running an fs.statSync per directory entry per keystroke; the directory listing cache now stores Dirent[] and classifies entries without a syscall, falling back to statSync only for symlink entries (#4199).
Fixed default model switches being persisted without changing the active goal-mode session when the current context exceeded the target model window. (#4219)
Fixed live tool preview spinners staying pinned to their first frame for eval and shell-style renderers. (#4170)
Fixed isolated task merges failing when the parent working tree carried WIP for a file the isolated subagent also touched. commitPatchToBranchWorktree now tries plain apply and git apply --3way first (agent-only outcome when the WIP-side blob is tracked in HEAD), then falls back to seeding the temp worktree with the baseline WIP so the delta patch's HEAD+WIP context matches, and rewinds WIP-only files afterward so they don't leak into the branch commit. Covers untracked WIP files, staged-new WIP files, and overlaps --3way cannot resolve. (#4136)
Fixed discoverAgents() skipping agents/ subdirectories inside OMP extension packages, so agents shipped by omp plugin install-ed npm plugins (e.g. loom) and --extension/extensions: settings roots now load the same way their sibling skills/, hooks/, tools/ directories already do. The new scan goes through listOmpExtensionRoots, so Claude marketplace installs continue to flow through the claude-plugins provider without being double-counted. (#3920)
Fixed plan mode hanging without converging on ask/resolve after advisor cards, idle IRC messages, or follow-on turns. Plan-mode decision enforcement ran on only the non-synthetic prompt() return; continuation/wake paths settled via agent_end and bypassed it. Advisor cards and idle IRC are now recorded into context without waking an autonomous turn, and the ask/resolve decision is enforced at the universal agent_end terminal settle via a bounded-retry counter (provider-neutral required, both tools kept available) that reminds-then-forces a fixed number of times and then yields to the user — never looping, never silently ending plan mode un-converged. An irc send await:true to an idle plan-mode session now answers the sender through the existing ephemeral side-channel auto-reply instead of stranding it until its wait timeout, and a queued forced plan decision is dropped when its continuation is skipped or plan mode exits. (#3910)
Reduced subagent streaming CPU cost: the recent-output window no longer re-splits the full (up to 8 KB) tail on every streamed text token. Fragments without a newline extend the current last line in place, and a full recompute runs only when line boundaries actually change.
Reduced task-render CPU cost: the task result frame (repainted ~30×/sec via the spinner) previously did 7+ full passes over the result set (some/filter/reduce); a single pass now derives the status booleans, footer counts, and request total, and incremental review extraction reuses the yield data the caller already normalized instead of re-normalizing it.
Reduced model-resolution cost: resolveModelRoleValue now builds the preference context (an O(n) model-order map over all available models) once and reuses it across every fallback pattern instead of rebuilding it per pattern, and matchModel hoists the case-folded pattern once instead of .toLowerCase()-ing it for every candidate across each filter pass.
Reduced read-tool allocation: line counting counts newlines directly instead of allocating via split("\n"), and the hashline formatter no longer counts the same content twice.
Fixed the assistant-message streaming fast path dropping the transient flag, which disabled the transient render path (code-highlight skip and streaming prefix caches) on every same-shape streaming tick. In-flight renders now correctly skip per-tick syntax highlighting; highlighting applies once at message finalization.
Fixed hidden goal-mode todo context: phase names and task text are now sanitized before prompt injection (no raw newlines or control characters forging extra context lines), and the block is only rendered with tool-accurate guidance when the todo tool is active or discoverable instead of unconditionally instructing the agent to call an unavailable tool.
Fixed custom tool loading treating process.exit() from a tool module's import or factory as a host process exit instead of a recoverable load failure. Custom tools now load under the shared extension exit guard, so an exiting tool is skipped with a load error while remaining tools still load (#1704).

[16.3.1] - 2026-07-02

Breaking Changes

Changed the grep, glob, and ast_grep tools to take a single optional path argument instead of a paths array. path accepts one path or a semicolon-delimited list (src; tests); omitting it searches the workspace root (.). Multi-path search, delimited expansion, and internal-URL scopes are unchanged. (ast_edit continues to take paths.)

Added

Added speech.enhanced setting to rewrite assistant output into natural spoken prose
Added speech.enhanced setting: assistant output is rewritten into natural spoken prose by the tiny/smol model before synthesis — code blocks become one-clause descriptions, links speak their label or site name, numbers and symbols read naturally, lists become flowing sentences. Blocks are rewritten fence-aware and coalesced (bounded to two concurrent completions); any failed or timed-out rewrite falls back to the mechanical cleanup so speech never blocks on the model.

Changed

Redesigned speech vocalization for low latency and clean spoken content. Assistant markdown now runs through a speakable-text pipeline before synthesis: code blocks and tables are silent, links speak their label, bare URLs speak their host, inline-code ticks/emphasis/heading/bullet markers are stripped, and long file paths collapse to their basename. Segmentation is now parent-side and emits at sentence boundaries immediately (the previous engine-side splitter held each sentence until the next one arrived), with clause-level cuts for long sentences and an idle flush when generation stalls mid-sentence. macOS gains a gapless streaming playback backend (ffmpeg AudioToolbox, sox fallback) instead of spawning afplay per sentence.

Fixed

Fixed stuttering/latency in speech by running synthesis chunks through the player gaplessly
Fixed race condition causing EPIPE errors and broken pipes during speech playback
Fixed interrupted speech audio by ensuring segments queue and drain in order
Fixed speech vocalization starting only after the entire reply was synthesized: ONNX inference blocks the TTS worker's event loop, so per-segment IPC audio chunks queued unflushed and arrived in one burst. Streaming sends now drain the IPC channel before the next segment's inference, cutting time-to-first-audio to ~1.5s regardless of reply length.
Fixed an unhandled EPIPE: broken pipe, write rejection at the end of speech playback: the streaming player's stop() raced an un-awaited FileSink.end() against the backend SIGKILL, and mid-session writes never awaited the flush. Writes now await the flush (so a dead backend is detected and the chunk replays on the next candidate or the per-file path) and stop() swallows the expected teardown rejection.

[16.3.0] - 2026-07-02

Added

Added providers.anthropic.serverSideFallback configuration option to opt into Anthropic's server-side-fallback beta chain, allowing Claude requests to automatically retry on alternative models when blocked by classifiers.
Added task.softRequestBudgetNotice configuration option to enable subagent soft-budget wrap-up steering notices while keeping the graceful abort guard active.

Changed

Significantly optimized session loading and rendering performance, including a 10x speedup for streaming reveals on large messages, 35% faster session resumes for large files using native streaming JSONL parsing, and reduced overhead for edit-patch fallbacks.
Improved TUI responsiveness and reduced CPU usage during long-running tool sessions by throttling status-line redraws and optimizing subagent persistence checks.
Updated the tester subagent prompt to allow skipping tests for trivial changes.
Improved DuckDuckGo web search error clarity and documented datacenter/shared-egress limitations in provider settings.

Fixed

Fixed session persistence corrupting Anthropic, OpenAI, and Google signed thinking blocks and reasoning payloads, ensuring cryptographic signatures and encrypted content are preserved verbatim to prevent API replay rejections (HTTP 400).
Fixed several issues with the apply_patch and edit tools, including preventing dirty buffers on early aborts, rejecting overwrites of pre-existing files, stopping at the first failing file in multi-file operations, and pruning extremely large file snapshots to prevent session inflation.
Fixed process termination (SIGTERM, SIGHUP, uncaught exceptions) to ensure editor drafts are saved, sessions shut down cleanly, and background jobs are cleaned up.
Fixed /quit and /exit commands blocking session closure by introducing a shutdown budget and backgrounding remaining tasks.
Fixed git clone and fetch operations being killed prematurely by applying a separate 30-minute deadline for network transfers instead of the standard 5-minute local-command timeout.
Fixed git index corruption issues in mergeTaskBranches and applyNestedPatches when post-merge stash pops conflicted with the cherry-picked HEAD.
Fixed collaboration (/collab) issues, including preventing teardowns from cancelling active host dialogs, sanitizing host-delivered errors for guests, and ensuring guest UI requests are correctly routed and rendered.
Fixed RPC mode deferred shutdown (pi.shutdown()) and abort_bash commands from being blocked by active background bash processes.
Fixed model discovery ignoring NODE_EXTRA_CA_CERTS and resolved a rare Bun garbage collection segfault during discovery.
Fixed task.maxConcurrency and task.maxRecursionDepth limits being bypassed by sub-spawn paths or when queued spawns were cancelled.
Fixed various TUI and rendering issues, including macOS Ghostty Command+V image pasting, CJK history rendering across compactions, status-line redraw crashes with BigInt values, and overlapping rows in the subagent progress tree.
Fixed /copy code and /copy cmd commands being treated as normal prompts instead of copying the requested blocks.
Fixed legacy tool compatibility issues for createReadTool, createGrepTool, and extension validation failures for omp install pi-lean-ctx.
Improved robustness of MCP authentication error detection, Smithery command authorization failures, and streaming preview responsiveness for write, edit, and eval tools.
Fixed llama.cpp router/preset mode reporting 128k context in the status bar for every preset picked from /model regardless of the preset's configured --ctx-size.
Fixed post-rewind context to tell the agent the checkpoint completed and to make repeat rewind calls recover with guidance instead of a bare no-checkpoint error.
Fixed grep/ast_grep search scopes rejecting www. and collapsed-scheme (https:/host) URL spellings.
Fixed ctrl+p role-model cycling getting stuck on one transition and skipping every other role.
Fixed interactive bash status line not updating after directory changes (cd).
Fixed session title refreshes ignoring user TITLE_SYSTEM.md overrides during replans, and prevented auto-generated titles from incorrectly preserving all-caps text from user messages.
Fixed the live todo HUD going stale during long tool-use loops by adding mid-run reminders for incomplete items.
Fixed /shake and mid-stream chat rebuilds erasing active LLM output.
Fixed Tavily web search to retry without recency filters if no content is returned.
Fixed user-configured LiteLLM discovery providers keeping stale reseller display-name suffixes for up to 24 hours after upgrade by invalidating the warm model cache.

[16.2.13] - 2026-07-01

Fixed

Fixed models.yml remote compaction schema support for V2 streaming endpoint fields. (#4146)
Fixed the SSH tool to reject cwd values of ~ and ~/... before sending guaranteed-bad quoted tilde paths to remote POSIX shells. (#4002)

[16.2.12] - 2026-07-01

Breaking Changes

Removed the canonical-alias grouping and resolution layer. The equivalence key (overrides/exclude) in models.yml/models.json is now inert, and canonical-related methods have been removed from ModelRegistry.
Removed the "CANONICAL" tab from the interactive model selector and the omp models canonical subcommand.

Changed

Simplified model selection and matching by resolving bare model IDs in --model, modelRoles, and enabledModels via exact/flat-ID and provider-preference matching instead of cross-spelling canonical coalescing.
Improved cold start performance by removing the catalog-wide canonical index build from the startup path.

Fixed

Fixed the write tool not streaming execution progress to the TUI while files are being written.
Fixed live tool-call argument previews disappearing while arguments stream.
Fixed the LSP tool ignoring timeouts and abort signals during cold-starts and notification writes.
Fixed the browser tool leaking Chromium/Puppeteer processes when operations are aborted or when an agent session is disposed.
Added a configurable 30-second timeout (extensionHandlerTimeoutMs) to extension tool call handlers to prevent hung third-party extensions from blocking execution, and fixed a timer leak keeping the CLI alive.
Fixed the built-in fd tool ignoring shell cancellation (such as Ctrl-C or timeouts) during directory walks.
Fixed MCP stdio requests hanging indefinitely past their configured timeouts when the child subprocess stops draining stdin.
Fixed Python eval shell helpers buffering child-process output by streaming chunks and truncating oversized output.
Fixed cached model edit variants failing to update when changing project directories.
Fixed subagents with structured output schemas failing validation by correctly wrapping the schema in the system prompt.
Fixed MCP Streamable HTTP request and notification timeouts staying unarmed during stalled response body reads.
Fixed evaluation and task spawn defaults to respect restricted agent spawn lists.
Fixed timed-out browser.run calls leaving evaluated JavaScript continuations running.
Fixed performance degradation in session context and branch path reconstruction on deep linear histories.
Fixed agents repeating the same tool call across turns without corrective steering by wiring the cross-turn tool-call loop guard into sessions.
Fixed OpenAI-compatible model discovery (including LM Studio) reporting flat default context windows when proxies omit context length metadata, by resolving discovered IDs against the bundled model reference catalog to inherit accurate context windows, output limits, display names, modalities, and reasoning support.

[16.2.11] - 2026-07-01

Fixed

Fixed model-discovery requests (Ollama, Llama.cpp, LM Studio, OpenAI, LiteLLM, vLLM) failing to clear timeouts after completion, preventing potential memory and timer leaks.
Fixed grep and ripgrep built-in tools ignoring shell abort and timeout signals during recursive directory walks, allowing them to be interrupted immediately.
Fixed omp setup speech returning prematurely before Whisper STT downloads complete, and improved error reporting for worker download failures.
Fixed high memory usage in multi-target ast_grep searches by only retaining the final page window while preserving exact match and file counts.
Fixed async job manager disposal and task cancellation handling to properly release session semaphores when aborted.
Updated and expanded omp:// documentation coverage for managed memory, skill tools, image/speech generation, and package CLIs.

[16.2.10] - 2026-06-30

Changed

Updated grep warnings to clarify that large files are partially searched rather than entirely skipped
Renamed the filesystem walker worker count environment variable from PI_GREP_WORKERS to PI_WALK_WORKERS
Centralized filesystem traversal policy in pi-walker.
Changed the in-session /resume session picker to open as a fullscreen window on the terminal's alternate screen, matching the startup --resume picker and /settings. It borrows the alt buffer for its lifetime (the transcript is untouched underneath) and enables mouse tracking — the wheel scrolls the list and a left click resumes the row under the pointer — with the keybinding hint and bottom border pinned to the screen bottom. Previously it mounted inline in the editor slot and rendered compactly without mouse support.

Fixed

Fixed Git subcommands ignoring repository paths by stripping ambient Git environment variables
Fixed concurrent bash commands cross-killing each other on cancel/timeout. Cancellation cleanup previously walked the whole host process tree and signalled every descendant spawned since a per-run baseline, so cancelling or timing out one command could SIGTERM an unrelated command's child still running in parallel (it looked "new" relative to the canceller's baseline). Each run now tracks only the processes it actually spawned (via a brush-core spawn-observer hook) and scopes its TERM/KILL waves to that set, leaving concurrent runs untouched.
Fixed /skill:<name> invocation losing the user's prompt context when the slash token was reached mid-prompt via the autocomplete. The slash-command parser now recognizes a /skill:<name> token surrounded by whitespace in non-slash, non-local-execution drafts (in addition to the leading form) and threads the surrounding prose through to the skill as args, so the typed prompt survives both in the editor (see the TUI changelog) and in the dispatched skill message. Drafts that already begin with another slash command (/compact /skill:foo), a bash sigil (!echo /skill:foo, !!echo /skill:foo), or a python sigil ($ run.py /skill:foo, $$ run.py /skill:foo) keep their existing dispatcher precedence and are not hijacked by the mid-prompt skill parser. Applies to the interactive TUI, ACP, and RPC dispatch paths via the shared parseSkillInvocation helper in extensibility/skills (#3913).
Fixed the incomplete-todo reminder drifting to the bottom of the screen and piling up as dozens of duplicate copies in native scrollback. The reminder rendered in a dedicated anchored live-region container (todoReminderContainer) pinned above the editor, so it re-rendered in place every frame and — being taller than the viewport on short terminals while the subagent/job HUD churned below it — had its top rows committed to scrollback again on each reflow. It is now committed once into the transcript as a regular block (the same path TTSR notifications use), so it stays anchored in history where it fired.
Fixed subagent frontmatter thinkingLevel being overridden by modelRoles.task model suffixes. (#3915)
Fixed Ruff LSP auto-detection for Windows Python virtualenvs by checking .venv/Scripts, venv/Scripts, and .env/Scripts before falling back to PATH. (#3916)

[16.2.9] - 2026-06-30

Breaking Changes

Renamed the built-in quick_task subagent to sonic; update any task spawns or configurations referencing quick_task by name.

Added

Added the llama3.2:3b local model option for memory and auto-thinking tasks, utilizing a quantized ONNX model.
Added a built-in Tester subagent designed to write high-signal tests for behavior, invariants, and edge cases while avoiding redundant or low-value tests.
Added a Speech-to-Text submit trigger setting to auto-submit dictation on release, on complete sentences, or via a spoken submit command.
Added a loop-guard mechanism that detects thinking/response loops and injects a system notice during auto-retries to guide the model to break the pattern and take a concrete next step.

Changed

Renamed the Speech-to-Text (STT) setting label from "TTS Submit Trigger" to "Speech-to-Text Submit Trigger".

Fixed

Fixed an issue where mid-run compaction was incorrectly skipped when a persisted assistant display variant shared a persistence key but differed in content from the live message.
Fixed duplicate placeholder cards being created when streamed tool blocks started with an empty ID.
Fixed omp debug --profile failing on Bun by treating the optional --allow-natives-syntax flag as best-effort when v8.setFlagsFromString is unavailable.
Fixed release binaries missing the compiled tiny-model Transformers.js version pin, preventing runtime resolution issues on certain platforms like Homebrew Darwin arm64.
Fixed MCP OAuth flows silently falling back to a random redirect port when the preferred port (default 3000) was busy, which caused authentication failures with strict providers. The flow now fails fast with a configuration error when a static client ID is pinned, while dynamic registration flows continue to use fallback ports.
Fixed /mcp reauth and /mcp add commands ignoring the Escape key during OAuth authentication, allowing users to cancel the flow immediately instead of waiting for the timeout.

Removed

Removed the built-in oracle subagent.

[16.2.8] - 2026-06-30

Added

Added built-in Go coding rules including go-add-cleanup, go-bench-loop, go-exp-promoted, go-ioutil, go-join-hostport, go-new-expr, go-rand-v2, and go-range-int

Changed

Relaxed strict bash tool constraints regarding the use of search, grep, ls, and find commands

Fixed

Fixed auto-compaction dead-ends by automatically triggering a shake rescue to elide oversized tails
Improved compaction warning message to suggest running /shake images for irreducible image tails
Fixed grep/search direct execution to accept JSON-array string paths for string-or-array inputs. (#3873)
Fixed auto-compaction dead-ending with "Compaction freed too little context to make progress" when a single recent turn (large tool output, heavy fenced/XML block) is itself bigger than the recovery band — findCutPoint can't cut inside one message, so the summarizer had no lever left. The guard now runs an artifact-backed shake elide pass over the oversized tail and re-tests headroom before pausing, and the remaining warning points at /shake images for image-only tails it can't elide. (#3786)
Fixed reviewer/task subagents whose incremental yield (type: ["overall_correctness"], type: ["findings"], …) carried a value that mismatched the matching property's sub-schema being silently accepted and then post-mortem rejected with schema_violation — opaquely swapping the agent's accepted output for an error blob. The yield tool now validates each incremental section's data against its top-level property's sub-schema (items schema for array-typed labels) and surfaces the same retry feedback as terminal yields, so models like deepseek-v4-pro that emit "Correct"/"correct."/"approved" for an enum field get up to three corrective retries; the existing MAX_SCHEMA_RETRIES override then accepts the value with SUBAGENT_WARNING_SCHEMA_OVERRIDDEN instead of losing the entire result. Unknown labels stay unconstrained (#3870).
Fixed streaming tool-call previews (notably write) showing an empty body for the entire streaming phase by surfacing the partial JSON already in hand on the first reveal, then pacing only subsequent growth (#3881).
Fixed hashline edit mode preserving UTF-8 BOM bytes on edited files. (#3867)
Fixed slow local LLM streams by forwarding persisted stream timeout settings (providers.streamFirstEventTimeoutSeconds, providers.streamIdleTimeoutSeconds) into model requests, so users can widen or disable watchdogs without environment variables. (#3878)
Fixed the bash interceptor blocking echo / printf redirects to /dev/null, /dev/tty, /dev/stdout, and /dev/stderr device sinks while still directing real file writes to the write tool. (#3763)
Fixed long snapcompact sessions re-sending multi-megabyte standing image archives on every provider request by enforcing a per-request frame byte budget, letting auto-compaction fall back to context-full summaries when snapcompact output is too large, and omitting legacy over-budget frames from rebuilt LLM contexts. (#3792)

[16.2.7] - 2026-06-30

Breaking Changes

Replaced the global serviceTier and fastModeScope settings with granular, per-family settings (tier.openai, tier.anthropic, and tier.google) to control service tiers, subagents, advisors, and /fast mode targets.

Changed

Improved binary file detection and terminal handling to prevent corruption from non-UTF-8 content, and updated file summaries to explicitly note skipped binary files.
Enhanced context compaction (snapcompact) to resolve shapes contextually based on rendered text content.

Fixed

Improved reliability of DuckDuckGo web searches by updating browser request headers and parameters
Fixed an issue where CJK (Chinese, Japanese, Korean) history could become unrenderable during repeated context compactions.
Fixed a memory exhaustion bug in the TUI when using /resume on large previous sessions.
Fixed an issue where the irc inbox missed messages that arrived while the recipient agent was already running.
Fixed a startup hang caused by system-prompt GPU detection blocking and repeatedly running failed probes.
Improved error reporting for omp tiny-models download by displaying the actual worker-side download error.
Resolved status inconsistencies between /extensions, /mcp list, and the dashboard, ensuring MCP server states, allowlists/denylists, and configuration files (like mcp.json) stay fully synchronized.
Improved branch-mode task merges to preserve the agent's original commit history (messages and authors) and fixed a bug where merges were rejected due to unrelated dirty changes in the parent checkout.
Fixed an issue where the Working... loader spinner would prematurely disappear or fail to re-arm after a subagent (task) tool completed or during transient overlays (such as auto-compaction or auto-retry).

[16.2.6] - 2026-06-29

Changed

Optimized argument streaming performance by throttling JSON re-parsing for renderers that do not require raw input.

Fixed

Fixed memory leaks in the benchmark CLI by ensuring provider sessions are properly closed upon completion.
Fixed TUI rendering lag and shimmer frame starvation during streamed tool-call argument previews.
Fixed llama.cpp discovery mapping unlimited output limits (max_tokens = -1 / n_predict = -1) to the generic 32K cap instead of the actual discovered runtime context window.
Fixed memory exhaustion on Ctrl+C flush for large resumed sessions with repeated compaction summaries, and improved fork session previews starting with developer or assistant turns.
Fixed omp search applying web search provider filters before resolving its implicit provider chain.
Fixed Gemini web search to support standard GEMINI_API_KEY developer API credentials for native Google Search grounding, in addition to Cloud Code Assist OAuth.
Fixed the bash interceptor blocking echo and printf redirects to standard device sinks (such as /dev/null, /dev/tty, /dev/stdout, and /dev/stderr).
Fixed session file size inflation in the edit tool by pruning extremely large oldText and newText snapshots (over 32 KB) from tool-result details, while preserving visible diffs, paths, lines, and diagnostic metadata.
Fixed Windows MCP stdio launches for PATH-resolved npx.cmd shims by preserving the cmd.exe wrapper path to keep subprocess stdio attached.
Fixed the DuckDuckGo web_search provider returning empty results for non-encyclopedic queries by switching from the Instant Answer API to parsing the HTML frontend, and added clear error handling for bot-challenge throttling.
Fixed Windows --extension paths with spaces or \?\ prefixes being truncated or incorrectly passed to Bun import/spawn APIs.
Fixed /mcp reauth compatibility with Cloudflare by aligning OAuth prompt behavior with the reference MCP SDK and updating the client label to oh-my-pi.

[16.2.5] - 2026-06-28

Changed

Status line now collapses a linked git worktree path to the project name with a worktree icon, leaving the git segment to show the branch once instead of repeating it in the path.

Fixed

Fixed Ollama and llama.cpp discovery overestimating local context windows by using model training metadata instead of the runtime num_ctx / n_ctx, which could let compaction retry prompts that llama.cpp rejects as over context. (#3752)
Fixed isolated subagent worktrees embedding long task ids and merged in the working path; isolation now uses compact hashed segments with an m mount dir. (#3756)
Fixed recoverable context-overflow compaction keeping the failed assistant error turn in visible session history after scheduling the retry. (#3747)
Fixed editing a file read from outside the workspace (e.g. ~/.claude/settings.json) failing with "File not found": the read snapshot header now carries the full out-of-workspace path so the edit resolves it directly instead of against the working directory.
Fixed subagents spawned with an output schema (agent(..., schema=...), task with structured output) failing with schema_violation: missing required fields since the typed-yield rework: a type: "result" finalize carrying the full object was assembled as a section named result, nesting the payload one level deep. String-typed yields are now treated as terminal finalizers (their data is the complete result), only array-typed yields form accumulating sections, and a data-less finalize keeps accumulated sections instead of collapsing to the last assistant turn.
Fixed the per-provider concurrency cap (e.g. providers.ollama-cloud.maxConcurrency) bracketing the whole subagent lifecycle, which deadlocked any spawn tree wider than the cap because parents held every slot while waiting for children queued on the same cap. The semaphore now wraps each provider HTTP request, so a parent's slot frees between turns and child subagents can acquire while their parent's tool calls are running. (#3749)
Fixed Ctrl+Q / Ctrl+Enter follow-up submissions sending the literal [Paste #N] marker instead of the expanded paste body (#3737).

[16.2.4] - 2026-06-28

Fixed

Fixed todo-reminder HUD rendering outside durable chat history while preserving native collapsing and auto-clear behavior.
Fixed goal-mode continuations losing track of persisted todo progress, ensuring autonomous goal turns reconcile stale in-progress items before moving to subsequent tasks.
Fixed the /move command to correctly relocate the current session and its artifacts to the target directory, allowing /resume to work seamlessly from the new location.
Fixed omp update leaving behind stale Bun install-cache directories for globally installed packages.
Fixed a reconnection loop issue in /collab sessions caused by oversized entries (such as large tool outputs) by truncating replicated payloads that exceed 1 MB.
Fixed autolearn and local memory writes mutating Anthropic prompt-cache prefixes mid-session, ensuring prompt injections remain session-stable.

[16.2.3] - 2026-06-28

Added

Added support for multiple configurable advisors via WATCHDOG.yml/WATCHDOG.yaml files, allowing per-advisor models, tool subsets, and instructions.
Added /advisor configure, a full-screen, mouse-driven TUI to easily manage the advisor roster, configure models, toggle tool permissions, and edit instructions.
Added full unified edit diffs to advisor transcripts, allowing advisors to see changes directly without re-reading files.
Added the statusLine.compactThinkingLevel setting to render the model segment's thinking level as a single leading glyph instead of a separate text suffix.
Added support for tracking reasoning tokens in session and advisor statistics.
Added Remote Compaction V2 streaming configuration settings (compaction.remoteStreamingV2Enabled and compaction.v2RetainedMessageBudget) to control token budgets and toggle V2 streaming for remote compaction.
Added the edit.citationTags setting to emit model-facing hashline section headers as OpenAI citation markers with opaque source IDs, along with citation-marker unwrapping for hashline edit parsing, diff previews, and streaming matching.
Added mutable session titles with automatic replan title refreshes and configurable idle recaps.
Added support for incremental yield submissions with typed sections and final results for subagents.

Changed

Reduced session file size by removing redundant thinking signatures already present in payloads
Advisors can now be granted any built-in agent tool (including edit, write, and bash), removing the previous read-only restriction.
Improved the debug log and raw SSE stream viewers with a wider, bordered overlay, clearer status indicators, dynamic layouts, and mouse support for scrolling and interaction.
Updated the idle recap feature to use an LLM-generated summary of where things stand (anchored by the live goal and active todo task) instead of a static status line.
Refined interrupted thinking system instructions to encourage smoother continuation.

Fixed

Fixed array-typed output schema validation by correctly assembling incremental yields into lists.
Fixed OpenAI/Codex compatibility by removing top-level schema combinators from tool parameters.
Fixed validation errors for untyped final yields in strict-mode providers by allowing null types.
Fixed Alt+M default-role model configuration being disabled by the current session's context size.
Fixed MCP type: "sse" servers by adding the legacy HTTP+SSE endpoint handshake and streaming JSON-RPC response path.
Fixed interrupted reasoning blocks being incorrectly stripped when they contained a valid signature.
Fixed interrupted thinking being lost in LLM provider requests after user interrupts by properly stripping trailing reasoning blocks from assistant turns while preserving them in the UI and session history.
Fixed the live todo HUD going stale during long tool-use loops by introducing a mid-run reconciliation reminder that prompts the agent to update incomplete items.
Fixed resumed OpenAI and OpenAI-Codex sessions losing encrypted reasoning and native assistant turns during rehydration.
Fixed the ask tool's custom answer editor dropping the original question and option list while typing.
Fixed auto-snapcompact failing the session on local blockers (such as text-only active models, high non-ASCII transcripts, or context budget overflows) by gracefully downgrading automatic maintenance to context-full compaction.
Fixed autoresearch's before_agent_start handler crashing when the system prompt was undefined.
Fixed OMP exiting silently during startup when encountering standalone Codex hook scripts in ~/.codex/hooks/.
Fixed unreachable keyboard shortcuts in HTML session exports by changing the "toggle thinking" and "toggle tools" shortcuts from Ctrl+T and Ctrl+O to bare T and O keys.
Fixed user-invoked /skill: prompts reaching model providers as developer turns instead of user turns, including during compaction.
Fixed reasoning streaming being locked off for OpenAI-compatible providers that stream reasoning content without advertising reasoning support in model metadata.
Fixed /shake and other mid-stream chat rebuilds erasing live LLM output by preserving the in-flight streaming components and pending tools.
Fixed the time_spent status-line segment ticking continuously during idle sessions by ensuring it only accumulates active agent execution windows and resets correctly across session switches.
Fixed expanded pending SSH previews committing provisional rows to native scrollback before the final result render.
Fixed ssh:// rejecting POSIX-capable remotes whose login-shell classification was ambiguous by verifying a working transfer shell directly and gating transfers on that capability.

Removed

Removed history URI support for reading agent transcripts

[16.2.2] - 2026-06-27

Added

Added a new tiny model role for consolidated online task handling.
Added a textVerbosity setting to control OpenAI and Codex response detail.

Changed

Simplified the status line subagent display by removing running state and hub hint indicators.
Updated online title, memory, and classification tasks to prioritize the new tiny model role.

Fixed

Improved reliability of auto-retry logic for aborted requests by standardizing error classification across model adapters and ensuring stale session states are reset.
Enhanced MCP authentication error detection, header-based server discovery, and 401/403 authorization failure detection during Smithery commands and HTTP RPCs.
Fixed auto-generated session titles incorrectly preserving user all-caps text, ensuring proper sentence casing while still respecting intentional mixed-case identifiers.
Fixed Tavily web search with recency filters to automatically retry without a time range if Tavily returns an empty HTTP 200 response.
Fixed TUI thought stream stalling and ui.loop-blocked warnings during subagent-heavy runs by optimizing the mid-run compaction persistence check.
Fixed marketplace-installed plugins incorrectly appearing in both the npm plugin list and the extension-package status provider.
Fixed inconsistent OpenRouter prompt-cache hits on /advisor turns by ensuring advisor agents inherit the same provider-shaping options, hooks, and settings as the main agent.
Fixed path-scoped TTSR (Targeted Tool Safety Rules) evaluation for hashline and apply_patch edit streams, ensuring rules are correctly applied to file paths parsed from section headers and envelope markers without leaking across file scopes.

[16.2.1] - 2026-06-27

Added

Included project context files (AGENTS.md, etc.) in the advisor's system prompt to ensure adherence to user-defined project rules
Added project context files (AGENTS.md and the like) to the advisor's system prompt, so the read-only reviewer judges against the user's standing project rules the same way the main agent does.

Fixed

Fixed live ACP generate_image updates resolving OMP-internal image blob refs before sending renderable image content to clients. (#3623)
Fixed Claude marketplace plugin .mcp.json MCP servers to expand environment variables in url and headers before connecting. (#3621)

[16.2.0] - 2026-06-27

Breaking Changes

Renamed the search tool to grep and the find tool to glob. Existing user settings are automatically migrated to the new configuration keys.

Added

Added ssh://host/path support to read, search, and write tools for single text files and directory listings on pre-configured POSIX SSH hosts.
Added an interactive /move overlay with path autocompletion and directory creation prompts, starting a fresh session in the target directory while leaving the previous session resumable.
Added support for file deletion and moving within file editing operations.
Added providers.maxInFlightRequests setting to cap concurrent LLM requests per provider across local OMP processes.
Added support for discovery.type: litellm in models.yml to automatically discover model metadata from LiteLLM gateways.
Added models.yml configuration options remoteCompaction and compactionModel to run compaction on a separate model or opt into provider-native compaction.
Added project, user, and plugin-level dap.json and dap.yaml support for defining or overriding debugger adapters used by the debug tool.
Added TinyFish, DuckDuckGo, xAI, and Firecrawl web search providers.
Added an Appearance setting for OSC 9;4 native terminal progress indicators during active agent turns and context maintenance.
Added Loop Guard "Tool-Call Reminder" to automatically interrupt Gemini reasoning loops that generate excessive planning headers without acting.

Changed

Redesigned the persistent Todo HUD as a compact connector tree with fixed-budget stage previews
Anchored status and HUD containers to prevent redundant UI elements in terminal scrollback
Redesigned the persistent Todo HUD to render active stages as a connector tree, capping the number of displayed stages and tasks to keep the plan overview concise and stable.
Anchored the status row container directly between the HUD and the editor to prevent redundant UI elements from piling up in terminal scrollback.
Optimized edit tool UI: delete and single-file move operations now render as compact status rows
Improved terminal output for file-level edits (delete/move) to display accurate paths and state
Refined edit renderer to prevent misleading "No changes" messages in multi-file operations
Changed the inlineToolDescriptors setting from a boolean to a three-way enum (auto | on | off), defaulting to auto to inline descriptors only for Gemini models.
Added caching for successful document conversions (PDFs, Office documents, EPUBs) to avoid redundant conversions on repeated reads.
Moved the Working… activity indicator below the sticky todo and subagent HUDs so it sits just above the editor instead of floating atop the todo panel.

Fixed

Fixed Z.AI web search to initialize the Streamable HTTP MCP session before calling web_search_prime, preserving the returned session ID for authenticated tool calls. (#3619)
Fixed terminal hangs on Ctrl+Z (SIGTSTP) after running bash tool calls, and insulated MCP stdio servers from terminal job-control signals.
Fixed macOS Cmd+V silently dropping image-only clipboard pastes in supported terminals.
Fixed an infinite loop in auto-compaction when a single recent turn exceeded the compaction threshold.
Fixed an issue where the advisor could enter a spam loop of repeated blocker injections, polluting the transcript.
Fixed Claude API and Anthropic classifier refusals polluting the replayed session context or persisting as assistant dialogue.
Fixed MCP tool calls failing with strict-schema servers by stripping internal metadata fields at the MCP boundary.
Fixed a terminal freeze/hang when a subagent abort was triggered.
Fixed thinking blocks appearing in the UI when thinking level is "off" for providers that return them anyway.
Fixed GitLab Duo Agent namespace/project discovery and authentication flow hangs.
Hardened and improved ssh:// protocol handling, including support for IPv6 brackets, percent-encoded hosts, port validation, and POSIX shell restrictions, while preventing unsupported tools from attempting SSH connections.
Fixed TUI usage display fraction resolution and added expiry dates for banked Codex rate-limit resets to the /usage display.
Fixed a rendering issue where long-running SSH commands left stale pending headers in terminal scrollback.
Fixed garbled casing in auto-generated session titles by reconciling title tokens against the user's original messages.
Fixed IRC broadcasts rendering twice in the main agent's transcript.
Improved the persistent Todo HUD styling and progress indicators to make it self-describing and visually distinct.
Fixed browser screenshots reporting 0x0 dimensions when image headers expose real dimensions.
Fixed snapcompact compaction silently falling back to LLM summaries when local preflight rejects the archive.
Fixed snapcompact compaction silently falling back to an LLM summary when local preflight rejects the archive; manual and auto snapcompact now fail locally with the blocker instead of making provider calls. (#3599)
Fixed garbled casing in auto-generated session titles. normalizeGeneratedTitle (packages/coding-agent/src/tiny/text.ts) used to force Title Case via a \b\p{Ll} regex, capitalizing function words ("for" → "For") and amplifying stray model capitals ("dAemon" → "DAemon"). It now reconciles each title token against the user's own message: tokens typed verbatim are kept; proper nouns the user cased distinctively are restored when the model flattened them ("tinyvmm" → "TinyVMM"); lowercase words carrying a stray interior capital the user never wrote are flattened ("dAemon" → "daemon"); and model-cased PascalCase proper nouns ("GitHub", "OAuth") are left untouched. Restoration is limited to distinctively cased source tokens so a message that merely starts with "For" can't force a mid-title "for" to "For". Applies to both the local tiny-model and online pi/smol title paths.
Fixed IRC broadcasts (to: "all") rendering twice in the main agent's transcript. A subagent broadcast fans out one bus.send per live peer, and listVisibleTo always includes Main, so the main agent received the body once as its own irc:incoming card and once per other recipient as an irc:relay observation of the sibling legs (Sender → Other) — identical text shown N+1 times. IrcTool.#executeSend now sets suppressRelay on every broadcast leg when Main is among the targets (it already has the body via its direct incoming card), and IrcBus.send skips #relayToMainUi for suppressed legs. Direct sub→sub relays, direct messages to Main, and Main's own outbound sends are unaffected.
Fixed Claude API refusals polluting the replayed session context and causing later prompts to refuse again. (#3592)
Fixed browser screenshots reporting 0x0 dimensions when Bun.Image rejects an image whose PNG/JPEG header still exposes real dimensions. (#3577)
Fixed Anthropic classifier refusals being persisted as assistant dialogue after no fallback handled them; refusal stops are now displayed as errors but pruned from active and saved context before the next prompt. (#3591)
Fixed the eval tool.* bridge leaking the harness-internal i ("intent") field into MCP tools/call requests, so strict-schema servers (Linear, anything with additionalProperties:false / Zod .strict()) rejected every call with -32602 unrecognized_keys: ["i"] while the same call via the direct model tool-call path succeeded. MCPTool.execute / DeferredMCPTool.execute (packages/coding-agent/src/mcp/tool-bridge.ts) now strip INTENT_FIELD at the MCP boundary, so an MCP call behaves identically whether issued by the model directly or via the eval tool.* bridge; servers that legitimately declare i as a real parameter keep it untouched. (#3575)
Fixed the advisor entering a spam loop in which it emitted hundreds of repeated Stop., Done., and No issue; continue. <advisory severity="blocker"> injections, polluting the primary transcript and destabilizing the watched agent after the task was already complete. The advisor system prompt's rules ("at most one advise per update", "NEVER send the same advice twice") are now enforced in code by a new AdvisorEmissionGuard on the enqueueAdvice boundary in AgentSession: it normalizes each note (case-insensitive, punctuation-folded), drops content-free self-talk filler (stop/done/no issue continue/lgtm/etc.), dedupes by exact normalized text across the session (bounded FIFO history), and rate-limits to one accepted note per advisor model prompt cycle. Reset on advisor reset (compaction, session switch, /new) so a re-primed reviewer can re-raise old issues. (#3520)
Fixed auto-compaction thrashing on a session whose single most-recent kept turn already exceeds the compaction threshold. prepareCompaction keeps that turn verbatim (findCutPoint never cuts at tool results), so the rewritten context stays above threshold; the context-full / snapcompact success tail scheduled the agent-authored auto-continue (and the overflow/incomplete retry) unconditionally, so the next agent_end re-entered #checkCompaction over the same oversized tail and re-fired forever. This is the residual loop left after #3247 capped snapcompact's own frame projection — once the frame cap drops below one frame, snapcompact is skipped and the context-full summarizer path still made no headroom. #runAutoCompaction now gates the threshold auto-continue on a post-maintenance headroom check (#compactionCreatedHeadroom, sharing shake's COMPACTION_RECOVERY_BAND hysteresis from #2275) and the overflow/incomplete retry on a separate fit check (#compactionCreatedRetryFit, measured after the failed turn is dropped) so a recoverable overflow that fits the window still retries; when a pass frees too little for the relevant path it pauses automatic maintenance and emits a single warning instead of looping. The post-turn threshold check also ignores an assistant's stale pre-compaction usage so the scheduled auto-continue cannot re-trip on the kept assistant's old high token count. The headroom check now treats any residual at or below the recovery band as progress: the band sits strictly under the compaction threshold, so a stale/tool-output prune that already pushed the trigger sub-band no longer makes a residual that merely holds the line report a false "no progress" and suppress a valid auto-continue.
Fixed the advisor prompt allowing confident root-cause claims about tool-call arguments absent from its reviewed transcript, so timeout advice now has to cite observed fields instead of inventing mechanisms like paths[0] array flattening. (#3483)
Fixed eval agent() helper subagents remaining visible as idle/IRC-revivable peers after the helper call returns; one-shot eval subagents now dispose and unregister at completion. (#3407)
Fixed the parent interactive prompt wedging (frozen, 0% CPU) after a subagent's yield-triggered abort. The executor's abort monitor and abort listener each called session.abort() fire-and-forget, so runSubprocess could resolve and adopt the subagent before its abort cleanup finished clearing in-flight session state. Active-subagent aborts are now deduped through a single cached cleanup promise that the subprocess awaits (bounded) before finalizing. (#2805)
Fixed thinking blocks appearing in the UI when thinking level is "off". Some providers (MiniMax, GLM, DeepSeek) return thinking blocks even with reasoning disabled; thinking blocks are now auto-hidden when the thinking level is "off", regardless of the hideThinkingBlock setting. Toggling thinking block visibility while thinking is off shows a status message instead of silently no-op'ing. (#626)
Fixed the TUI usage display failing to resolve a used fraction for limits that only populate remainingFraction (no usedFraction, used/limit, or percent+used). The TUI's local resolveFraction was missing the inverted-remaining fallback that the shared resolveUsedFraction from @oh-my-pi/pi-ai already handles — replaced the local copy with the shared function so the TUI and CLI paths resolve fractions identically.
Fixed long-running SSH command boxes leaving a stale ⏳ SSH: [host] header above the final ⇄ SSH: [host] header in terminal scrollback. The SSH renderer now keeps its partial-result chrome on the pending icon/state and opts the block out of stream-commit while isPartial holds (via the new ToolRenderer.provisionalPartialResult flag honored by ToolExecutionComponent.isTranscriptBlockCommitStable), so the stable-prefix ratchet can't promote the partial header to native scrollback only to have the final render strand it above the settled frame (#3177).
Fixed Gemini over-planning runs that emit long chains of thinking headers (**Refining …**, ## Examining …) without ever issuing a tool call. The session now interrupts that stream, discards the partial reasoning turn, injects a hidden tool-call reminder, and continues with the corrective context instead of burning the full budget on planning.
Fixed Cmd+V on macOS silently dropping image-only clipboard pastes (screenshot via Cmd+Shift+5 "save to clipboard", Chrome image copy, …) — the user had to fall back to Ctrl+V. Follow-up to #3506: that fix handled clipboards exposing a file URL or path; the screenshot path leaves only raw image bytes on the pasteboard. macOS terminals (iTerm2, Terminal.app, Warp, Ghostty without OSC 5522, Windows Terminal forwarding, …) intercept Cmd+V and read NSPasteboardTypeString first; for an image-only clipboard that read returns "", so the terminal forwards a complete-but-empty bracketed paste (\x1b[200~\x1b[201~). CustomEditor.handleInput inserted the empty payload and the keystroke disappeared. CustomEditor now runs its own BracketedPasteHandler ahead of the inherited handler so the assembled paste payload is routed regardless of whether the start marker, payload, and end marker arrive in one stdin chunk or are fragmented across several (Windows Terminal under load, certain SSH muxes, tmux extended-keys passthrough, …). Strict-zero-length assembled payloads route to the same onPasteImage smart reader the app.clipboard.pasteImage keybind uses (attaches the clipboard image, or falls back to the #1628 smart text paste / "clipboard is empty" diagnostic); explicit image-file paths route to onPasteImagePath (the #3506 path also benefits from split-chunk assembly); everything else hands off to the base editor's pasteText so [Paste #N] markers, autocomplete, and undo state stay intact. Whitespace-only pastes are preserved as literal text. Trailing keystrokes that arrive in the same stdin read as the paste (a user hitting Enter right after Cmd+V) are queued behind the in-flight clipboard read and only dispatched once the image has reached pendingImages, so submit can't fire against an empty draft and leave the image stranded on the next prompt. (#3601)

[16.1.23] - 2026-06-26

Added

Added compaction.midTurnEnabled for mid-turn threshold auto-compaction before the next tool-loop provider request. (#3525)
Added grep -q/--quiet/--silent and -x/--line-regexp to the in-process grep builtin used by the bash tool. -q suppresses all stdout and exits 0 on the first match (short-circuiting, with match status taking precedence over read errors per GNU); -x anchors each pattern to whole lines. Unblocks shell conditionals such as grep -qx "$applet" <(strings bin).
Added plan-mode guidance (hashline edit mode only) steering the agent to revise the plan file section-by-section with SWAP.BLK/DEL.BLK/INS.BLK.POST anchored on markdown headings — a heading resolves its whole section (through nested deeper headings), so the agent can rewrite, drop, or append sections without rewriting the file.
Added tui.renderMermaid to control Mermaid fenced-block ASCII rendering; disabling it also removes the Mermaid diagram hint from the generated system prompt so Mermaid blocks fall back to ordinary highlighted code fences.
Added /resume <session-id> in the interactive command system, reusing the existing session-id/prefix resolver while bare /resume still opens the selector.
Added manual omp gc storage maintenance with gc.* defaults for blob sweeping, cold-session archiving, and SQLite WAL checkpointing.

Fixed

Fixed /resume <session-id> in the interactive TUI only searching the active cwd's session directory; id-prefix lookup now falls back to sessions from other cwd buckets like CLI --resume <session-id>.
Fixed plan mode rejecting edits to plan artifacts when models refer to them by bare filenames
Fixed absolute paths to session-owned artifacts being incorrectly routed through the editor bridge
Fixed Windows stdio MCP wrapper chains spawning visible PowerShell/cmd windows on startup after the #3544 fix. StdioTransport.connect() now probes whether OMP already has an inheritable console and resolveStdioSpawnCommand skips windowsHide/CREATE_NO_WINDOW in that case, so cmd.exe/PowerShell grandchildren reuse the terminal console instead of allocating visible conhosts during MCP startup or reconnects. (#3567)
Fixed Kimi-family models defaulting to hashline edit mode; they now fall back to replace unless edit.modelVariants, PI_EDIT_VARIANT, or PI_STRICT_EDIT_MODE explicitly opts into hashline.
Fixed MCP OAuth discovery rejecting Atlassian-style cross-host issuer metadata during the resource-server fallback probe; issuer matching now remains enforced for advertised auth-server candidates but no longer blocks fallback metadata where the resource host and authorization-server issuer differ. (#3551)
Fixed plan approval applying the wrong execution model when the model-tier slider sat on the model that exit would restore. The match check now compares the selected role's effective thinking level against the pre-plan thinking level, so picking the active planning tier is retained and picking a same-model tier with an explicit thinking suffix (e.g. default = sonnet:off while plan-mode raised thinking to high) goes through applyRoleModel instead of silently restoring the pre-plan level. (#3554)
Fixed plan approval applying the wrong execution model when the model-tier slider sat on the model that exit would restore. The match check now compares the selected role's effective thinking level against the pre-plan thinking level, and a singleton cycle (only modelRoles.plan configured, default unset, so getRoleModelCycle synthesizes a lone default entry from the active plan model and the slider stays hidden) is no longer pinned as the execution tier — approval falls through to the pre-plan restore instead of silently switching back to the plan model. (#3554)
Fixed the eval Julia kernel showing only runner-internal backtrace frames (at top-level scope (./none:N), at main (…runner-…jl:635)) with no exception type or message, making cell errors undebuggable. The host renderer (packages/coding-agent/src/eval/kernel-base.ts) displays a non-empty traceback verbatim and only falls back to ename: evalue when it is empty; the Python and Ruby runners embed the rendered error in traceback, but the Julia runner (packages/coding-agent/src/eval/jl/runner.jl) built traceback from stack frames only, so the message was dropped. emit_error now seeds traceback with the showerror output (matching the REPL's ERROR: text) ahead of the frames.
Fixed Windows stdio MCP servers timing out (and popping a visible terminal) when their direct child was a cmd.exe wrapper that itself launched a console grandchild — node wrappers, npx.cmd -y mcp-remote, similar nested shells. StdioTransport.connect() unconditionally passed detached: true to Bun.spawn, but Windows has no SIGTSTP/SIGTTIN to escape; detached only maps to CreateProcess(DETACHED_PROCESS), which strips the parent's inherited console. The hidden direct cmd.exe lost its console, so the grandchild allocated a brand-new visible conhost whose stdout no longer routed through OMP's pipe — the proxy reported the bridge was up while OMP timed out waiting for the MCP initialize response. resolveStdioSpawnCommand now returns detached: false for every Windows return shape (direct .exe, cmd.exe-wrapped batch / unresolvable command, npm cmd-shim launched through node) and keeps detached: true on POSIX, where the original SIGTSTP/SIGTTIN reason still holds; connect() consumes the resolved flag. (#3544)
Fixed the LSP tool's per-cwd config cache (getConfig in packages/coding-agent/src/lsp/index.ts) never being invalidated, so .omp/lsp.json, root markers, and plugin LSP configs added after the first LSP call stayed invisible for the remainder of the process — even after reload *. The cached observation persists across chat sessions because OMP is a single-process binary, so users were stuck on No language servers configured until they killed the process. The reload handler now drops the cwd's cache entry and re-runs loadConfig before iterating servers, so the explicit refresh request behaves as the prompt documents. (#3546)
Fixed stale snapcompact archive state being reintroduced at the AgentSession manual and auto LLM compaction save paths; preserveData.snapcompact is now stripped after hook- and result-supplied preserve data are merged, so a prior snapcompact pass can no longer leak frames into a later context-full compaction. (#3561 by @serverinspector)
Fixed the snapcompact→context-full migration shipping the prior archive's plaintext to the summarization provider without secret redaction. When secrets are configured, the migrated archive's text/textHead/textTail regions are now obfuscated alongside the previous summary, while opaque provider-replay state (OpenAI remote-compaction encrypted_content) stays byte-identical. (#3561)
Fixed fresh interactive launches ignoring modelRoles.default when the configured default lives on an extension-registered provider (e.g. an openai-compat plugin's posthog/claude-opus-4-8). createAgentSession resolved the default role before extension factories registered their providers, so the role-pointed model wasn't visible there; on a fresh launch (no -c/--resume) the post-extension fallback then went straight to pickDefaultAvailableModel, replacing the user's configured default with the first bundled provider default with auth (commonly openai/gpt-5.5 when OPENAI_API_KEY was set). The fallback now retries the default-role lookup against the post-extension allowed-model set — including its explicit thinking selector — before falling back to a bundled provider default. (#3569)

[16.1.22] - 2026-06-26

Fixed

Fixed Windows stdio MCP server launches showing a separate cmd.exe window for direct executable servers; MCP subprocesses now set windowsHide on every Windows spawn path. (#3535)
Fixed MCP OAuth still failing with Authorization failed: An unexpected error occurred against Plane's https://mcp.plane.so/http/mcp endpoint after #3502. The protected-resource metadata advertised the path-scoped issuer https://mcp.plane.so/http, but discoverOAuthEndpoints (packages/coding-agent/src/mcp/oauth-discovery.ts) probed /.well-known/oauth-authorization-server at the origin root first and accepted the metadata served there — which describes a different issuer (https://mcp.plane.so/) whose /authorize endpoint rejects every grant. Discovery now honors RFC 8414 §3.3 and skips authorization-server / OpenID Connect well-known documents whose issuer field doesn't match the queried base URL (trailing-slash insensitive). Servers that omit issuer keep today's permissive behavior, so legacy flows are unaffected. (#3537)

[16.1.21] - 2026-06-26

Fixed

Fixed autolearn.autoContinue (the "Auto-run capture at stop" toggle) letting the synthetic capture turn keep working after the learn/manage_skill call instead of yielding. With the toggle on, the controller fires autolearn-nudge.md as a single attribution: "user" message on a new turn; the prompt opened with "Before you finish:" and gave no terminal contract, so the agent treated the synthetic prompt as the user's reply to its prior pending question (e.g. "Want me to commit and push?") and continued — pushing commits, running tools, etc. — without the user ever answering. The nudge is now split into two prompts: passive mode (the reminder rides the user's real next message) keeps additive framing — "answer the user normally; the capture is in addition to" — while auto-continue mode (autolearn-nudge-autocontinue.md) is explicitly terminal — "not a user reply; do not treat this as approval; capture, then stop; wait for the user's next prompt". Attribution stays user to preserve llama.cpp warm-prefix reuse (#3456). (#3504)
Fixed the clipboard image-paste keybind dropping image-file-only pasteboards as literal text on macOS. When the clipboard exposes only the file URL (e.g. Finder Cmd+C on a .png, certain screenshot tools), arboard::get_image() returns ContentNotAvailable and pbpaste(1) returns empty (it only surfaces plain text / RTF / EPS), so InputController.handleImagePaste either fell through to the #1628 text fallback and pasted the path verbatim, or dead-ended with "Clipboard is empty". The keybind path now (1) reaches the public.file-url representation directly via a new readMacFileUrlsFromClipboard AppleScript bridge on Darwin and routes the first image-shaped path through handleImagePathPaste, (2) reuses a new extractImagePathFromText helper so any clipboard-text path that survived pbpaste is detected the same way the bracketed-paste handler already detects them — including a whole-text-as-path fallback for anchored paths whose unescaped spaces would otherwise be shredded by the bracketed-paste splitter (macOS screenshot names like /Users/me/Desktop/Screenshot 2026-06-25 at 1.23.45 PM.png), and (3) decodes file:// URLs to filesystem paths in normalizePastedPath (mirroring Codex's normalize_pasted_path in codex-rs/tui/src/clipboard_paste.rs). Keybind- and terminal-mediated paste now agree on raw image bytes, plain image paths, file:// URLs, Finder-copied image files, and screenshot filenames with spaces alike. (#3506)
Fixed compiled-binary validation for legacy pi.extensions packages whose source imports worker-only coding-agent subpaths or extension-local package subpaths such as typebox/value; omp install @charmland/pi-hyper-provider, omp plugin doctor, and runtime provider discovery now use a main-thread-safe load path. (#3508)
Fixed repeated todo updates in one TUI turn stacking full todo panels; superseded todo snapshots now stay live until the next todo update replaces them or the turn ends. (#3516)
Fixed MCP OAuth authorization failing with Authorization failed: An unexpected error occurred against authorization servers (Plane is the live example) that reject redundant fallback resource indicators. OMP now drops same-origin resources only when it synthesized them from the server URL fallback (e.g. https://mcp.plane.so/http/mcp). Provider-advertised resources from OAuth/protected-resource discovery or an embedded authorization-URL resource query parameter are preserved even when they are same-origin or origin-only, so gateway-hosted MCP services can still request the audience they advertised. The refresh-token path uses the same policy, filtered against the authorization-server origin persisted on the credential as authorizationUrl, with tokenUrl's origin as the legacy fallback when that field is absent. (#3502)

[16.1.20] - 2026-06-25

Fixed

Fixed Ctrl+Z hanging the terminal after any tool call had run: the TUI tore down (ui.stop()) but the process kept running in Sl+ state, leaving the user with a dead terminal recoverable only via kill -9. The embedded brush-core shell behind every bash tool call installs a tokio SIGTSTP listener on Process::wait (crates/brush-core-vendored/src/sys/unix/signal.rs::tstp_signal_listener → tokio::signal::unix::signal(SIGTSTP)); per tokio's contract, the first call for a SignalKind permanently replaces the kernel-default handler for the lifetime of the process. So the first bash invocation — even /usr/bin/true — silently overrode SIGTSTP's "stop" default, and InputController.handleCtrlZ's subsequent process.kill(0, "SIGTSTP") was swallowed by tokio. The handler now sends SIGSTOP (uncatchable, unblockable, unignorable) to the foreground process group, so the kernel parks omp regardless of installed handlers and the shell sees the whole job stop even when omp runs behind a wrapper (npx, pnpm exec, bunx, …) or as one stage of a pipeline. MCP stdio servers now spawn detached into their own session — they're insulated both from terminal job-control signals (which used to stop their process trees and leave the JSONL read loop blocked on silent pipes) and from the new pgid=0 suspend itself (#3461).
Fixed image-only composer submissions while the agent is streaming being treated as empty input, which dropped the image or aborted the active turn when another message was queued. Pending pasted images now count as submit content for Enter and Ctrl+Enter follow-ups. (#3467)
Fixed omp gallery --state accepting lifecycle tokens that did not match displayed state labels and rendering unknown state values as · undefined; displayed labels now work as aliases, invalid values fail with a valid-token list, and failed gallery fixtures visibly render failures. (#3473)
Fixed the bash tool's snapshotted mise() shell function dying with command: command not found: because $__MISE_EXE was empty in the replay shell. generateSnapshotScript captured the function via declare -f/typeset -f but only ever re-exported PATH, so every other env var the rc file set (notably the *_EXE sidecar mise activate exports) was lost; the function body then expanded command "$__MISE_EXE" "$@" to command "" … and died with exit 127. The snapshot script now scans captured function bodies for $VAR / ${VAR…} references and re-emits export NAME='value' for each referenced var that is currently set (with a denylist for shell-internal names like PATH/HOME/BASH_*/LC_* plus a likely-secret denylist for *TOKEN*/*SECRET*/*API_KEY*/*PASSWORD*/*PRIVATE_KEY*/*ACCESS_KEY*/*CREDENTIAL*/*SESSION_KEY*), the snapshot script umask 077s itself and the JS caller chmods the snapshot file/dir to 0600/0700 so the new export pass can't leak secrets into a shared tmp dir. Fixes mise, asdf shims, direnv-style helpers, and other activation idioms that pair a function with a helper env var. getShellConfigFile now also honours env.HOME (falling back to os.homedir()) so sandboxed callers can target a non-default rc. (#3470)
Fixed concise history:// transcript rendering for find and search so scoped paths arguments are visible instead of being hidden behind JSON fallback output or omitted when a search pattern is present. (#3482)
Fixed manual /compact leaving session.isCompacting false while active-turn abort teardown awaited, so the first steer/follow-up typed during compaction startup now routes through the compaction queue instead of being lost. (#3485)
Fixed ollama-cloud task/subagent fan-out exceeding the provider's three-request concurrency cap by adding a provider-specific subagent limiter, and let configured task/smol/advisor model roles inherit the default retry fallback chain when they do not define their own chain. (#3464)
Fixed the per-provider subagent concurrency limiter (e.g. providers.ollama-cloud.maxConcurrency) being replaced with a fresh semaphore whenever the configured limit changed, which orphaned the in-flight slots on the old instance and let a runtime or mixed limit value exceed the cap. The limiter now resizes a single shared semaphore in place — raising the ceiling admits queued waiters immediately, lowering it drains in-flight holders without admitting past the new cap. (#3464)
Fixed a background-task spawn slot leaking from the task.maxConcurrency limiter when progress reporting threw between acquiring the slot and entering the guarded run: markRunning/reportProgress now run inside the try whose finally releases the semaphore, so a failed progress report can no longer permanently shrink subagent concurrency. (#3464)
Fixed active goal runs that successfully call yield and then receive a trailing empty assistant stop skipping threshold compaction; post-yield empty-stop suppression now still anchors active-goal compaction on the yield-bearing assistant turn, so long-running tasks continue after maintenance instead of settling early.

[16.1.19] - 2026-06-25

Fixed

Fixed omp install <plugin> failing extension validation in compiled-binary mode with Cannot find module '@(scope)/pi-ai/oauth' from '<plugin>/src/oauth.ts' (and any other non-wildcard pi-* subpath import like @oh-my-pi/pi-coding-agent/tools). The bundled-registry override map seeded by __buildLegacyPiPackageRootOverrides only covered the bare package roots, so rewriteLegacyPiImports rewrote @(scope)/pi-ai/oauth to @oh-my-pi/pi-ai/oauth, fell through to Bun.resolveSync (which bunfs can't satisfy on Bun 1.3.14+), then left the original specifier alone — at which point Bun's native resolver failed because most plugins declare @(scope)/pi-ai as a peerDependency only and never materialize a real install. The new scripts/generate-legacy-pi-bundled-registry.ts reads every bundled pi-* package's non-wildcard exports field and emits both the heavy legacy-pi-bundled-registry.ts (static imports + map) and a light legacy-pi-bundled-keys.ts (statically imported by legacy-pi-compat.ts to seed the override map without the cascade through legacy-pi-coding-agent-shim → ../index → export/html/...). scripts/build-binary.ts now runs the generator before bun build --compile. (#3442)
Fixed skill:// tool resolution losing loaded session skills when a tool runs outside the session-initialization module state. Internal URL resolution now prefers the caller's session.skills snapshot before falling back to the process-global skill list, so read skill://<name> works across tool execution boundaries. (#3436)
Fixed @image mentions on OpenAI Codex Responses (chatgpt.com gpt-5.5 and siblings) failing with Codex error event: [OneOfParam] [input[N].content[…]] [invalid_enum_value] Invalid value: 'input_image'. Supported values are: 'input_text'.. convertToLlm for fileMention always emitted a developer-role message, so the auto-attached image landed in a Responses content array that the Codex backend (and OpenAI Responses generally) only allows to carry input_text. #3421's previous fix only stopped the Codex Responses Lite header from going out on image-bearing turns; the full transport kept rejecting the same payload. convertToLlm now splits a mixed-content fileMention into two messages — text-only files stay on developer (so the auto-read context keeps instruction priority), while image-bearing files ride on user (the only Responses content slot that accepts input_image). (#3443)
Fixed local:// binary attachments being decoded with Bun.file().text() before read-tool file safeguards could reject or stream them. read local://... now routes file-backed resources through the normal filesystem reader, the protocol handler refuses known binary/container resources without materializing their bytes, and the streaming reader's NUL-byte refusal now also covers binary blobs whose first newline lies beyond the byte budget (videos, archives) — those previously bypassed the per-line scan and surfaced as decoded mojibake. (#3448)

[16.1.18] - 2026-06-25

Added

Added --list flag to view stored OAuth accounts for a provider
Added --account flag to select a specific OAuth account by index for token retrieval
Added mouse support for navigation, selection, and toggling in the extensions dashboard

Changed

Enabled fullscreen alternate-screen mode for the extensions dashboard
Improved tab navigation to mute empty enabled providers while keeping disabled ones selectable
/extensions (the Extension Control Center) now opens as a fullscreen window on the terminal's alternate screen, matching /settings: it borrows the alt buffer for its lifetime (the transcript is untouched underneath) and uses the same modern chrome — a titled rounded frame, the shared TabBar for provider switching, and a divider/footer layout. The dashboard is now mouse-aware: the wheel scrolls the list (and the detail inspector, which gained its own scroll viewport), hovering highlights tabs and rows, and clicking selects a row or — when it is already selected — toggles it; clicking a provider tab switches to it. Empty enabled providers are unselectable while disabled providers stay selectable so their master switch can re-enable them.

Fixed

Fixed omp install of legacy pi extensions failing with Cannot find module '/$bunfs/root/packages/coding-agent/src/extensibility/typebox.js' on every released omp-<platform>-<arch> binary. Commit dc5c93462f removed worker entrypoints from scripts/ci-release-build-binaries.ts; the inline comment then claimed the legacy-shim and package-barrel entrypoints (typebox.ts, legacy-pi-{ai,coding-agent}-shim.ts, packages/{agent,natives,tui,utils}/...) were "still" passed to bun build --compile, but they had never been re-added. The release binaries shipped without those files in bunfs, so legacy-pi-compat.ts redirected typebox imports to a bunfs path that didn't exist. __resolveTypeBoxShimPath now mirrors __validateLegacyPiPackageRootOverrides (#2168) by dropping the override when the shim file is missing, so a missing shim falls through to native node_modules resolution instead of emitting a dead bunfs URL (#3414).
Fixed every legacy @(scope)/pi-* and @sinclair/typebox import failing to load on the omp-darwin-arm64 release binary (and any other omp built with Bun 1.3.14). __validateLegacyPiPackageRootOverrides and the rewriteLegacyPiImports emit path both depended on --compile extras being reachable as /$bunfs/root/... filesystem entries, but Bun 1.3.14 stopped exposing them through every API (fs.existsSync, Bun.file().exists(), Bun.resolveSync, await import() on the bunfs path or its file:// URL all fail; only /$bunfs/root/<binary-name> itself answers). legacy-pi-compat.ts now keeps a JS-heap reference to every bundled pi-* surface in a lazy-loaded sibling legacy-pi-bundled-registry.ts and serves them through an omp-legacy-pi-bundled: virtual namespace whose Bun.plugin().onLoad returns synthetic re-exports — no bunfs path ever leaves the module in compiled mode, and dev / source-link / installed-package modes keep the historical file:// rewrite. The matching --compile extras in scripts/build-binary.ts, the shared scripts/binary-entrypoints.ts list, and the dead BUNFS_PACKAGE_ROOT / bunfsPath / __computeBunfsPackageRoot / __joinBunfsPath helpers are gone. (#3423)

[16.1.17] - 2026-06-24

Fixed

Fixed mnemopi auto-retain extracting facts/entities from assistant-authored transcript turns. MnemopiSessionState.retainMessages still stores the full multi-role window for episodic recall, but passes only user-authored turns as extractText, so assistant prose containing always/never no longer becomes durable user Instruction: memory. (#3372)
Fixed lazy tool auto-downloads hanging when Bun.write(dest, response) receives a streaming fetch() Response; tool assets now stream the response body to disk with the existing download abort signal and remove partial files on abort. (#3369)
Fixed profile-alias installer producing backslash-separated paths for bash/zsh/fish config files on Windows. path.join was used unconditionally, producing Windows-style paths that POSIX shells can't resolve. The installer now uses path.posix.join for non-Windows platforms and normalizes script paths to forward slashes for POSIX shell alias blocks, so omp --alias works correctly in Git Bash and WSL.
Fixed pasted or dragged non-image file paths in the TUI prompt staying as inert raw text; existing files now attach as clean local://attachment-N.<ext> references while image paths keep the image attachment flow. (#3360)
Slash commands are now recorded in input history (Up Arrow recall). Previously only 4 commands (/plan, /goal, /mcp, /ssh) stored their text; all other built-in slash commands were silently skipped because executeBuiltinSlashCommand returned true before addToHistory was called. History is now centralized in the input controller after successful command dispatch. Commands that may carry secrets (/login <url> with OAuth callback params, /mcp add --token <token>) are excluded from history to prevent credential leakage (#3148)
Fixed the ask tool's "Other (type your own)" free-text editor (prompt-style HookEditorComponent) ignoring Ctrl+Q and Ctrl+Enter, so Windows Terminal users who learned the app.message.followUp chord from the main editor (#1903 / fixed by #1905) got zero feedback on submit. The hook-style and main-editor surfaces honored matchesAppFollowUp; the prompt-style handler did not, leaving plain Enter as the sole submit path and Ctrl+Enter falling through to Editor as a newline (silently swallowed by WT). #handlePromptStyleInput now checks matchesAppFollowUp first — mirroring #handleHookStyleInput — and the hint reads enter or ctrl+q submit so the chord is discoverable. (#3353)
Fixed the TUI freezing when a tool approval prompt fires while /settings (or the Extensions/Agents dashboard) is open. The fullscreen overlay's close handler restored focus to the editor it had captured at open time, but ExtensionUiController had since swapped the editor out of the editor slot for the approval prompt — so on exit the visible prompt sat unreachable while keystrokes routed to the now-unmounted editor (no Enter/Up/Down/Esc response, only Ctrl+C escaped). SelectorController now restores focus to whatever currently owns the editor slot via a focusActiveEditorArea() helper, applied to settings, extensions dashboard, and agents dashboard close paths. (#3349)
Fixed /settings coercing enum/text values to display strings before handing them to the TUI list, preventing YAML numeric enum values from reaching native truncation (#3338).
Fixed all extension loading silently failing on the cross-compiled omp-darwin-arm64 release binary (downloaded directly or via a Homebrew tap wrapper) because __computeBunfsPackageRoot mis-handled import.meta.dir = "//root/omp-darwin-arm64". Bun 1.3.14 reports <bunfs-root>/<binary-name> for the compiled entry's import.meta.dir, but the pre-fix function joined metaDir + "packages" and produced /root/omp-darwin-arm64/packages — the binary basename was baked into every bunfs path, so the TypeBox/legacy-pi shims and every @oh-my-pi/pi-* package-root override failed existsSync validation and resolveCanonicalPiSpecifier fell through to a bunfs Bun.resolveSync that also could not find the module. The function now detects the bunfs-root + binary-basename shape (path.basename(path.dirname(metaDir)) === "root") and strips the trailing binary segment by slicing the original metaDir; the production bunfs shim join path also preserves Bun's bunfs-native //root / B:\~BUN\root prefix that path.join would otherwise collapse. (#3329)
Fixed llama.cpp discovery to prefer per-model /v1/models meta.n_ctx/meta.n_ctx_train values, refresh selected models after lazy load, and bypass fresh-cache reuse so server restarts update context windows. (#3310)
Fixed task.maxConcurrency: 0 serializing subagent spawns instead of running them unbounded. The settings UI labels 0 as "Unlimited", but the session-scoped spawn Semaphore clamped max via Math.max(1, max), so the second subagent body in a batch always waited for the first to release the seat. The constructor now treats max <= 0 (and any non-finite input) as unbounded via Number.POSITIVE_INFINITY, matching the eval parallel()/pipeline() worker-pool semantics (#3305).
Fixed MCP tool calls forwarding empty optional placeholder arguments ("" and {}) to tools/call; optional placeholders are now omitted while required fields and meaningful falsy values are preserved. (#3302)
Fixed the welcome Tip: line rendering with hardcoded #b48cff / #9ccfff pastels plus a manual \x1b[2m dim, so any light theme dropped the body to ~1.5:1 contrast (well under WCAG AA). renderWelcomeTip in packages/coding-agent/src/modes/components/welcome.ts now paints the label through theme.fg("customMessageLabel", …) and the body through theme.fg("muted", …) (no manual dim), so the line tracks the active theme and stays legible on light backgrounds. (#3337)
Fixed omp usage and the /usage command duplicating provider-wide disclaimer notes (e.g. OpenCode Go's "OMP-observed spend only") once per account × limit window. Provider-level notes now render once above the per-account sections in the TUI, CLI, and ACP render paths, and identical per-limit notes are deduplicated in the TUI aggregate renderer. (#3268)
Fixed the welcome panel advertising ? for keyboard shortcuts after the ? shortcut was deliberately removed (commit dcf482c4c). The tips section now points users at /hotkeys instead. (#1614)
Fixed Devin provider models silently producing empty responses under the default defaultThinkingLevel: auto. Devin models advertise reasoning: true but no thinking.efforts (Cascade selects effort by routing to sibling model ids, not a wire param), so getSupportedEfforts(model) was empty; clampAutoThinkingEffort returned the classifier-picked effort as-is, which then tripped requireSupportedEffort in pi-ai/stream.ts with Thinking effort low is not supported by devin/<id>. Supported efforts: (silently swallowed by the TUI). clampAutoThinkingEffort now returns undefined when the model has no controllable effort surface, matching clampThinkingLevelForModel; the auto-thinking turn hook also short-circuits the classifier call for these models. (#3356)
Fixed omp tiny-models download exiting before its unref'd worker subprocess could install the runtime or download model weights. The tiny-model client now references the worker while requests are pending so standalone CLI downloads wait for Downloaded ... / Failed ... completion. (#3291)
Fixed marketplace plugin installs registering only in installed_plugins.json and never in the runtime plugin tree, leaving slash commands and extensions unavailable after omp plugin install name@marketplace. The runtime loader now also enumerates the project-scope plugins root (<projectAnchor>/.omp/plugins) so --scope project installs surface alongside user-scope installs, with project entries shadowing same-named user entries (#3244).
Fixed umans requests with more than 10 live context images still sending every image despite the provider budget; outgoing provider contexts now drop the oldest images above the active provider cap while preserving text and newest images (#3230).
Fixed snapcompact auto-compaction looping the "snapcompact could not bring the context under the limit — using an LLM summary instead" warning on every threshold tick for sub-1M-token models (Claude Sonnet 4.5, GPT-5.x, Gemini 2.x). snapcompact.compact() was called with no maxFrames override, so it defaulted to MAX_FRAMES_DEFAULT = 80; the projection in AgentSession charges FRAME_TOKEN_ESTIMATE = 5024 per frame block (the conservative high-res Anthropic ceiling), making 80 × 5024 ≈ 402k frame-token projections that always overflow a 200k budget. AgentSession.#computeSnapcompactMaxFrames now sizes the maxFrames cap from a shape-aware reserve — 2 × geometry(shape).capacity worth of verbatim text-edge chars billed at the tiktoken cl100k 4-chars/token baseline (with a 1.15 multiplier for tokenizer drift), plus a 2k summary-template allowance — mirroring what #projectSnapcompactContextTokens will charge once frames land. The shape comes from the same snapcompact.resolveShape(model, settings) call the auto and manual paths pass into snapcompact.compact(). The cap reserve applies only to the frame-cap math, not the skip decision: snapcompact is skipped outright only when kept-recent + non-message ≥ ctxWindow − reserve (no headroom at all), so the frame-less text.length <= 2 * edgeCap short-circuit in planArchive can still land a valid text-only archive when residual headroom is positive but below the cap reserve. The projection guard catches any actual frame-bearing archive that overflows. (#3247)
Fixed large-session TUI stalls by tailing appended transcript JSONL and collapsing compacted history on the live display surface (#3258).
Fixed status-line usage segment ignoring Codex subscription limits that carry a scope.tier (#2877).
Fixed extension tool_call/tool_result events for hashline edit calls to expose event.input.path for single-file edits and event.input.paths for every parsed target, so planning-mode gates can allow one markdown plan edit but still block multi-file hashline calls that cannot be represented by one path (#1678).
Fixed scripted eval agent() subagents continuing after a successful yield when a trailing empty assistant stop arrived after the executor's yield-triggered abort. The session's agent_end maintenance compared #assistantEndedWithSuccessfulYield(msg) against the trailing empty-stop message — not the prior yield-bearing one — so the empty-stop recovery path appended a retry reminder and scheduled agent.continue(), reviving the already-yielded child. The yield handler now sets a sticky #yieldTerminationPending flag (cleared on the next prompt()) that short-circuits empty-stop / unexpected-stop / compaction continuations for the rest of the run, so a successful yield is terminal regardless of trailing stops (#3389).
Fixed snapcompact rasterizing transcript frames into requests bound for GitHub Copilot business and enterprise endpoints, which then rejected the session permanently with 400 vision is not supported. The snapcompact vision gate now also short-circuits whenever model.provider === "github-copilot" and the resolved baseUrl is not the canonical personal-Copilot host, protecting cached/stale Model specs that still advertise ["text","image"] on a non-personal endpoint. (#3387)

[16.1.16] - 2026-06-23

Breaking Changes

Renamed the eval agent() helper parameters agent_type → agent and return_handle → handle across every workflow runtime (Python, JavaScript, Ruby, Julia), so the names are identical in every language (no camelCase/snake_case split) and the agent-selection parameter matches the task tool's agent. The __agent__ eval bridge wire protocol was renamed to match.
Changed the eval tool to take a single cell per call ({ language, code, title?, timeout?, reset? }) instead of a cells array. State still persists per language across separate eval calls, tool calls, and task subagents, so each call is one logical step that reuses everything earlier calls defined — the array only encouraged re-importing/re-declaring the same setup in every batch. The schema, field descriptions, examples, system eval.md/workflowz helper docs, and the [i/n] cell-counter (now hidden for single cells) were updated to match; the renderer, ACP start-text, copy-targets, and collab-web tool view still parse legacy multi-cell transcripts.

Added

Added isolated, apply, and merge options to eval agent() across every workflow runtime (Python, JavaScript, Ruby, Julia) so workflowz-driven fan-outs can request the same copy-on-write worktree isolation the task tool offers (strict opt-in via isolated: true, matching the task tool; apply: false keeps captured patches/branches without merging back; merge: false forces patch mode). Extracted the task-isolation lifecycle into task/isolation-runner.ts so the eval bridge and TaskTool share one implementation (#3196)

Changed

Made the session picker fullscreen with mouse support for clicking rows and scrolling
Pinned the session picker footer to the bottom of the screen to prevent layout flickering
Simplified eval tool to accept a single logical step (code block) instead of an array of cells
Updated eval tool documentation to emphasize incremental, single-step execution
Restricted bash tool from using ls or find, requiring the use of read or find tools
Simplified todo tool interface to accept a single operation directly instead of an array of ops
Reinforced routing of fragile, multi-step shell logic to the eval tool over bash. The system-prompt tool policy, bash.md, and eval.md now treat loops, conditionals, heredocs, inline -e/-c scripts, multi-stage pipelines, and quote/JSON escaping as the signal to write an eval cell; bash's "compute a fact" carveout is narrowed to single short pipelines, and eval.md now actively claims that territory with runtime-templated examples (only enabled backends are advertised).
Made eval an essential built-in tool (loadMode: "essential", added to the default essential tool set) so it stays active under tools.discoveryMode: "all" instead of being hidden behind search_tool_bm25.
Made the --resume session picker fullscreen on the terminal's alternate screen, so the list scrolls with the mouse wheel and a row resumes its session on left click. Rows are hit-tested against the live scroll window, and the keybinding hint + bottom border are now pinned to the screen bottom instead of drifting up and down as the visible window changes height.

Fixed

Fixed local:// URLs decoding images as corrupted text (mojibake) instead of showing the image
Fixed omp --resume hanging instead of exiting when the startup session picker is cancelled (Esc) or there are no sessions to resume. Startup arms long-lived handles (theme/appearance listeners, settings save timer, model registry), so the cancel/empty paths' bare return left the event loop alive and the process stuck after the picker cleared the alternate screen. These paths now exit cleanly via process.exit(0), matching the --version/--export early-exit convention. The in-session /resume picker is unaffected — it keeps its own cancel handler that just closes the overlay.
Fixed the /resume session picker scrolling down after a session is deleted. The delete-confirmation dialog was mounted as a sibling below the picker's bottom border, briefly growing the picker past the terminal height; the TUI committed the picker's header rows into native scrollback to fit, and when the dialog closed windowTop stayed pinned at the new commit boundary, leaving the header stranded above the viewport. The picker now hosts the SessionList in a single content slot and swaps the dialog INTO that slot (replacing the SessionList) while it is open, so the dialog only competes with the SessionList's rendered budget — not the SessionList AND the picker chrome — and the picker frame stays inside the viewport. (#3283)
Fixed the eval tool card not streaming a still-running cell's stdout: a long-running cell (e.g. a time.sleep() monitor loop) showed nothing until it returned or was interrupted, then dumped everything at once. The renderer draws cell output from details.cells[i].output, which was only populated after backend.execute() resolved — live stdout streamed into the transient result content tail (and renderContext.output), which the per-cell render branch ignores. Streamed chunks now append to the active cell's output (a dedicated per-cell tail buffer, capped like the aggregate) as they arrive, so the card shows progress live; on completion the authoritative full output overwrites the live tail. log()/phase()/display() and status ops were unaffected because they already stream via the status channel.
Fixed Escape doing nothing in the Settings text-input fields (e.g. "Python Interpreter") on terminals with the kitty keyboard protocol active (ghostty/kitty). Inside the fullscreen settings overlay the protocol reports Escape as the CSI-u sequence \x1b[27u, which the text-input submenu's raw \x1b compare missed; handleInputOrEscape now decodes Escape via matchesKey, matching every other Escape-to-cancel path.
Fixed Julia eval graph/plot visualization (Plots.jl, GraphRecipes, Makie, etc.) never rendering inline. Two bugs: (1) the runner's build_mime_bundle/emit_error dispatched show/showable/showerror directly from the long-lived main() loop, whose world age is frozen before any cell ran, so rich show(::IO, ::MIME"image/png", …) methods registered when a plotting package is using-ed inside a cell were invisible — show fell back to the default struct repr (which itself threw on Julia 1.12, aborting the whole result). These calls now route through Base.invokelatest, and the text/plain probe is guarded so a failing repr can no longer suppress the image MIME. (2) The default GR backend popped up a native gksqt GUI window on each plot; the runner now defaults GKSwstype=100 (headless, overridable) so plots render only as inline PNGs, mirroring the Python runner's MPLBACKEND=Agg default.
Fixed streaming output blocks incorrectly calculating preview height, preventing flickering banners
Fixed streaming bash/eval tool output duplicating its … (N earlier lines, showing 10 of M) (ctrl+o to expand) preview into native scrollback. The collapsed output is a sliding tail window fixed at 10 lines, so when the box outgrew the live viewport (a tall command/output under a still-live predecessor such as a parallel tool) its mutating tail scrolled above the commit window and the renderer re-committed a fresh snapshot every frame, stacking dozens of stale preview banners and chunks. The output preview is now clamped to the viewport tail (Math.min(10, previewWindowRows())) and measured in visual rows at the box's inner content width (via the new outputBlockContentWidth helper), so on short terminals the volatile tail shrinks to stay on-screen and is never committed. Fixes the duplication introduced when scroll-off commits were made loss-free.
Prevented /handoff from executing while a response is streaming to avoid session corruption
Fixed /handoff cold-missing the provider prompt cache. Handoff generation now builds its request through the same pipeline a live turn uses (convertMessagesToLlm + Agent.buildSideRequestContext + prepareSimpleStreamOptions, via the new generateHandoffFromContext), so it reuses the live system prompt, normalized tools, transformed/obfuscated message history, and — critically — a stable promptCacheKey with a unique side sessionId. Previously the oneshot sent no cache-routing key and skipped the transformContext/transformProviderContext and tool/message normalization the loop applies, so its prefix never matched what the turn populated and every handoff re-read the whole context uncached. Mirrors the cache-preserving path already used by /btw and /omfg.
Fixed /handoff (and the RPC handoff command) resetting the agent while a response was still streaming, which let the live turn keep emitting into the torn-down session. Manual handoff now refuses while a prompt is in flight (matching /fork and /move); the auto-handoff path is unaffected.
Fixed Exa web search requests firing back-to-back with no client-side pacing by adding a configurable exa.searchDelayMs delay (default 1000ms) between Exa search requests. (#3271)
Fixed ask returning (cancelled) or aborting the tool when Escape dismissed Other (type your own) custom input; it now returns to the option selector so the user can pick a listed answer instead. (#3269)
Fixed /goal threshold auto-compaction skipping real sessions through three paths: per-turn supersede/drop-useless pruning no longer deflates the threshold trigger below the last provider-billed context; active-goal text stops now attempt threshold maintenance before unexpected-stop retry continuations can return from post-turn handling; and empty toolUse stops keep the existing cleanup pass that strips the orphan assistant from active context + session history before any compaction continuation. Active-goal compaction continuations now also resolve completed retry gates before returning, preventing isRetrying from staying stuck after a retry succeeds over the threshold. Added agent_end maintenance routing and Auto-compaction threshold decision debug logs so future no-start reports identify the exact early-return branch and the billed/stored/resolved/post-maintenance token counts that fed shouldCompact. (#3174)
Fixed active /goal runs that never reached agent_end because the model kept emitting tool calls inside one agent run. Threshold maintenance now runs between tool-call turns, compacts the live loop context in place, and suppresses queued continuations that would race the still-running goal loop. (#3174)

Removed

Removed append, tree, and diff eval helper functions from Python, JavaScript, and Ruby
Removed sort, uniq, and counter text processing eval helpers from Python, JavaScript, and Ruby
Removed the append(path, content), tree(path, max_depth?, show_hidden?), and diff(a, b) eval prelude helpers from every workflow runtime (Python, JavaScript, Ruby, Julia), along with their status renderers, icon entries, and tool/docs references. Use write/read for file mutation and tool.<name>(...) for richer filesystem operations.
Removed the sort(text, reverse?, unique?), uniq(text, count?), and counter(items, limit?, reverse?) eval text helpers from the Python, JavaScript, and Ruby prelude surfaces (Julia never defined them), along with the JS HelperBundle/HelperOptions members and docs references. Sort/dedupe/count inline in cell code instead.

[16.1.15] - 2026-06-22

Added

Added a share.store setting (blob "Encrypted Blob" | gist "GitHub Gist", default blob) controlling where /share uploads the encrypted session blob — the share server (blob) or a secret GitHub gist with share-server fallback (gist).

Changed

Changed /share to upload the encrypted session blob to the share server by default instead of a secret GitHub gist. Gist-hosted shares are fetched through the unauthenticated GitHub gist API from the viewer's browser, which GitHub rate-limits to 60 requests/hour per IP, so shared NATs and repeated reloads surfaced Failed to load session: Gist fetch failed: HTTP 403 in the viewer. Opt back into gist hosting with share.store: "gist".

Fixed

Fixed nohup/& background processes started inside an auto-backgrounded (:async:) bash turn being killed when that turn's shell was torn down. The per-job shell is now retained while a background process is still running (and reaped once it exits), so backgrounded commands survive across turns while still dying with the harness process.
Fixed openai-completions provider session state surviving /model switches across different providers or base URLs. AgentSession.#closeProviderSessionsForModelSwitch only evicted openai-codex-responses and openai-responses:<provider> keys; entries keyed openai-completions:<provider>:<resolvedBaseUrl>:<modelId> (cached strict-tools disable scopes and reasoning-effort fallbacks for the old transport) lingered indefinitely. Moving away from an openai-completions backend now evicts every cached entry for the previous provider, including entries whose base URL was resolved at request time rather than read from the catalog, while same-backend model toggles keep their cached state (#3260)
Fixed MCP stdio servers failing on Windows when the launcher's PATH walk can't pin down a bare npx/yarn/pnpm-style shim (empty Bun.env.PATH under a restricted parent process, UNC mounts that reject fs.access, locked-down shells). resolveStdioSpawnCommand used to fall back to the bare command name, which Bun.spawn → CreateProcess can only resolve as <name>.exe — never .cmd/.bat — so npx -y … died ~140ms after spawn with ENOENT/EINVAL. The Windows resolver now routes any unresolvable bare command through cmd.exe /d /s /c so Windows's own PATHEXT search picks up the shim. The reporter's diagnosis ("process.env not merged") was incorrect — the merge happens at transports/stdio.ts:316-319 — but the symptom they hit is real (#3250).
Fixed bracketed paste treating a bare .png filename as an image attachment path instead of normal prompt text. (#3253)
Fixed Perplexity web-search provider hijacking the auto fallback chain when only OpenRouter auth was configured. PerplexityProvider.isAvailable() accepted hasAuth("openrouter") as a credential, so any user with an OpenRouter key (for LLM access) had every webSearch: auto request silently routed through OpenRouter's perplexity/sonar-pro — bypassing downstream providers like Gemini and producing unexpected OpenRouter billing. Auto-chain admission now requires a direct Perplexity credential (PERPLEXITY_COOKIES, OAuth, or PERPLEXITY_API_KEY); users who want the OpenRouter-backed Perplexity path can still opt in explicitly with webSearch: perplexity (#3251)
Fixed configured model discovery caches to refresh when models.yml/models.json is newer than the cached row, so updated local model metadata is not shadowed by fresh models.db entries. (#3242)
Fixed hide-secrets handling so advisor session updates are redacted before the advisor model sees them and opaque assistant thinking blocks are no longer deobfuscated.
Filtered alias definitions brush's whitespace-only expander cannot execute ((, ), |, &, ;, <, >, `) from the bash-tool shell snapshot, so user rc-files containing compound aliases like Fedora's default which='(alias; declare -f) | /usr/bin/which …' no longer poison the brush session with error: command not found: (alias; (#3234).

[16.1.14] - 2026-06-22

Added

Added a Ruby eval backend (language: "rb"): a persistent subprocess Ruby kernel that shares the standard agent tool bridge and evaluation model, supporting IRB-style auto-display of results in a persistent session
Added a Julia eval backend (language: "jl"): a persistent subprocess Julia kernel that shares the standard agent tool bridge and evaluation model, supporting Julia REPL-style auto-display of results in a persistent session
Added configuration options eval.rb and eval.jl to enable or disable the Ruby and Julia backends (both opt-in, disabled by default), as well as ruby.interpreter and julia.interpreter paths for explicit runtime control
Added a Ruby eval backend (language: "rb"): a persistent subprocess Ruby kernel modeled on the Python kernel, speaking the same NDJSON protocol with isolated frame/stdout channels, clean SIGINT cancellation that preserves kernel state, and per-owner kernel cleanup. Ships the full prelude helper surface (display/read/write/append/tree/diff/env/output/sort/uniq/counter, the tool.<name> bridge proxy, and completion/agent/parallel/pipeline/log/phase/budget) over the shared loopback tool bridge, honors local:// roots, and auto-displays the last expression IRB-style (suppressing assignments/definitions). Gated by the eval.rb setting and PI_RB env flag, with an optional ruby.interpreter override.
Added a Julia eval backend (language: "jl"): a persistent subprocess Julia kernel with the same NDJSON/result-display/tool-bridge model as the Python and Ruby eval runtimes, including eval.jl / PI_JL gating, julia.interpreter override support, persistent per-session state, clean cancellation, owner-based kernel cleanup, local:// path roots, and the current Julia prelude helper surface (display/read/write/append/tree/diff/env/output, the tool.<name> bridge proxy, and bridge-backed completion/agent/parallel/pipeline/log/phase/budget, including structured schema parsing and return_handle DAG shaping).
The eval tool advertises only the runtimes enabled for the session: a disabled backend (rb/jl are off by default) is omitted from the tool's language enum, schema field descriptions, discovery summary, and prelude documentation, so disabled backends are never surfaced to the model

Changed

Refactored internal evaluation logic to consolidate shared kernel management and IPC handling
Refactored internal draft state to consolidate images and text into the editor component
Unified transcript rendering logic for background jobs, IRC traffic, and file mentions
Unified subprocess lifecycle management for mnemopi, speech, tiny-model, and TTS workers
Softened the system-prompt and tool-prompt guidance that hard-forbade reaching for shell equivalents of dedicated tools (grep/rg, sed/perl -i, cat/head/tail, find/fd). The "Specialized Tool Priority" section, the bash/search/find/read/replace tool prompts, and the FORBIDDEN/NEVER framing now state the dedicated tools as the preferred default rather than an absolute prohibition, so the agent no longer fights itself when a quick shell command is the right call. The opt-in bashInterceptor (default off) still hard-blocks these commands for users who enable it.
Unified subprocess lifecycle management for mnemopi, speech, tiny-model, and TTS workers
Refined TUI dashboard and list component layout logic for consistent rendering
Standardized scrollbar behaviors and keyboard navigation across all list components
Optimized internal message framing logic for better performance when handling large data bursts

Removed

Removed the readHashLines setting (the "Hash Lines" toggle under Files → Reading). Hashline read/search anchors ([PATH#TAG] snapshot headers plus LINE:content) are now driven solely by edit.mode === "hashline": the toggle was redundant when off (anchors are already suppressed for non-hashline edit modes) and a footgun when on (turning it off left the default hashline edit tool with no addressable anchors, since read then skips recording the snapshot tag). Existing configs are migrated automatically by dropping the stale key.

[16.1.12] - 2026-06-21

Changed

Refined secret obfuscation to only target message roles and fields containing operator secrets
Restricted obfuscator to ignore secrets and regex matches shorter than 8 characters
Optimized obfuscation to skip static system prompts and tool schemas in provider contexts
Ensured image data bytes are never modified to prevent corrupted data URL payloads

Fixed

Fixed secret obfuscation corrupting Codex image reads (and other provider requests) with Invalid 'input[N].content[].image_url'. Expected a base64-encoded data URL ... but got an invalid base64-encoded value. The obfuscator deep-walked every string in the outbound request — including inline image base64 and opaque provider replay/signature fields — so a configured secret that happened to be a substring of the base64 (or of an ordinary word like response) injected #HASH# placeholders mid-payload. Obfuscation is now opt-in and fully typed: only user messages, tool-result messages, and user-attributed developer messages (@file mentions) are redacted; system prompts and tool schemas pass through untouched; image bytes and signature/encrypted-reasoning fields are never rewritten; and tool-call arguments are the only JSON walked. Configured plain secrets and regex matches shorter than 8 characters are now ignored to stop false matches on short words.
Fixed RPC/ACP startup clobbering explicit caller/project/global configuration for task.isolation.{mode,merge,commits}, task.eager, task.batch, task.maxConcurrency, task.maxRecursionDepth, task.disabledAgents, task.agentModelOverrides, memory.backend, memories.enabled, advisor.{enabled,subagents,syncBacklog,immuneTurns}, plus the RPC-only async.{enabled,maxJobs} and bash.autoBackground.{enabled,thresholdMs}. applyDefaultSettingOverrides re-asserted the schema default as a runtime override after settings load, regressing the isConfigured() guard added for #2598 and ignoring every explicit value the embedder, project, --config overlay, or global config had set. The guard is restored, so the host default now only fills holes (#3207).

[16.1.11] - 2026-06-21

Added

Added tab.waitForSelector for more robust element-wait behavior
Added tab.waitForNavigation to monitor and await page transitions

Changed

Standardized public HTML exports to use the brand web palette instead of the local TUI theme
Updated export logic to allow choosing between brand-web or local TUI color palettes
Clamped browser tool timeouts to fail fast with named errors instead of opaque cell timeouts
Added strict validation to block unsupported Playwright-only selector engines in browser tools

Fixed

Fixed Codex image reads re-encoding to WebP and then failing the next request with an invalid input_image.image_url; Codex-bound images now stay in PNG/JPEG-compatible formats.
Fixed prompt-template and file slash-command submissions rendering both the raw slash invocation and the expanded prompt in the interactive transcript. (#3199)
Fixed the /model thinking picker labeling the OpenAI GPT-5.5 top effort as max instead of the catalog-declared xhigh (#3194).
Fixed session-title generation silently falling back to the online smol model (and billing whatever provider held the resolved API key — OpenRouter in the reporter's case) when the user had explicitly configured a local providers.tinyModel: generateSessionTitle raced local against online with a 10s timeout and fired the online request immediately whenever the local worker returned null (unknown key, model not downloaded, transformers.js failure). Now an explicit local-model choice is honored end-to-end — on local failure the session is left untitled with a logger.warn instead of billing the smol fallback (#3187)
Fixed OpenCode MCP discovery so array commands are normalized into a stdio executable plus arguments, environment is accepted as the OpenCode env key, and argument lists are omitted when empty. (#3180)
Fixed plan mode leaving write hidden under tools.discoveryMode: "all", which made the agent attempt to create the plan file via edit and stall. Plan-mode entry now force-activates write whenever the registry built it, matching the write+edit instructions in the plan-mode prompt (#3165)
Fixed edit and patch tool writes bypassing the ACP client's open buffer when Zed (or another ACP client) advertises the fs.writeTextFile capability: all three write-mode tools (edit, patch, replace) now route through the client bridge when it is available, so the editor's TypeScript diagnostics panel updates immediately instead of requiring a workspace reload.
Fixed the agent dashboard new-agent description and the hook editor (default hook-style mode) treating Ctrl+Q as plain text on Windows Terminal, leaving Windows users unable to submit because the terminal cannot deliver a distinct Ctrl+Enter event. Both forms now submit on the shared app.message.followUp chord (Ctrl+Q or Ctrl+Enter), matching the main prompt editor and any user remap of app.message.followUp (#2118).
Fixed pi.sendMessage({ display: true }) rendering the custom message twice when fired from a session_start extension handler (or any other non-streaming dispatch before the initial transcript render). ExtensionUiController.#applyCustomMessageDisplay rebuilt the chat from the freshly-persisted session entry, and main.ts's subsequent renderInitialMessages(undefined, { preserveExistingChat: true }) then both re-rendered from session entries AND re-appended the preserved chat children — duplicating the message. The rebuild now waits until renderInitialMessages has completed at least once (tracked via InteractiveModeContext.initialChatRendered); after that, post-init extension sends still rebuild as before so messages from tool_result / agent_end / etc. surface immediately (#1955).
Fixed Git-mutating automation silently mis-handling pure Jujutsu workspaces (.jj/repo/ present, no colocated .git/). task/worktree.ts#getRepoRoot previously threw a generic "Git repository not found" for isolated subagent setup, and autoresearch/git.ts#ensureAutoresearchBranch returned a soft "Not in a git repository" warning that let /autoresearch proceed with no branch isolation or auto-commits. Both paths now detect a pure jj workspace via the new jj.isPureJjRepo helper and surface an actionable Jujutsu-specific error pointing at jj git init --colocate. Colocated jj-git (both .jj/ and .git/ at the same root) and plain Git checkouts behave exactly as before (#1935).
Fixed auto thinking mode being silently dropped when a session is resumed (--continue/--resume/in-app switch). The session log persisted only the resolved per-turn effort, not the auto selector, so resume froze the session at the last concrete level and never reclassified again. The log now records the configured selector (auto vs concrete) alongside the resolved effort, so resumed auto sessions stay in auto (shown as pending until the next turn reclassifies) and manual concrete pins still restore as concrete — including a pin whose level matches the effort auto had just resolved to.
Fixed transcript scrollback stability on terminals with eager erase risk so completed assistant messages remain stable while new streaming lines are rendering
Fixed Hindsight retain (and the shared mnemopi/Hindsight recall paths) framing assistant turns whose only content was punctuation/whitespace — most commonly the lone . some providers emit for tool-call-only or thinking-only turns — into [role: assistant]\n.\n[assistant:end] blocks that polluted the bank, wasted retain tokens, and degraded recall. prepareRetentionTranscript, extractMessages, and flattenMessagesForRecall now require at least one letter or digit per message via a shared hasSubstantiveContent predicate (#1806).
Fixed /resume rendering forked child sessions without a fork tag, making them indistinguishable from their parent when titles match (#1792).

[16.1.10] - 2026-06-21

Added

Added tab.ariaSnapshot(selector?) to the browser tool for Playwright-format ARIA-tree YAML
Added tab.ref("e5") and support for aria-ref=e5 selectors in all tab action methods

Changed

Made write and find tools essential so they are always available initially (surviving tools.discoveryMode === "all" hiding) to ensure instructions to write/find files are immediately executable (#3165)

Fixed

Fixed streamed tool-call previews freezing on their placeholder body ($ …, Write: …, an empty args tree) even after the tool finished: the pending card was created while arguments streamed, but when the closing full-arguments message_update never arrived (smooth-streaming disabled leaving the throttled arguments stale, an owned-dialect projector, or a superseded/aborted turn that still ran the call) nothing re-applied the final args. tool_execution_start — the one event every execution path emits with validated full arguments right before the result — now reconciles them onto the existing pending card and cancels any in-flight reveal so a late tick can't re-truncate the body.

[16.1.9] - 2026-06-21

Added

Added LLM request JSON export functionality to /dump

Changed

Improved browser stealth by suppressing common automation flags and patching property descriptors
Enhanced stealth for WebGL, Worker, IFrame, Screen, and Audio APIs to evade detection
Updated toString patching to register native function sources for improved fingerprint protection

Fixed

Fixed omp list and omp remove silently starting an interactive agent session (forwarding the bare verb to the model as a prompt) instead of surfacing the real omp plugin list / omp plugin uninstall <name> commands (#2935)
Fixed lazy-initialized LSP servers (basedpyright/pyright, and likely gopls/rust-analyzer) hanging on the first request: the message reader matched incoming messages against pending client requests by id before checking for a method, so a server-originated workspace/configuration pull whose id collided with an in-flight request was swallowed as a bogus response, leaving the pull unanswered and the server wedged. The reader now routes any message carrying a method as a server request before id-matching (#3001)
Fixed omp --approval-mode=yolo acp and other global option flags placed before a subcommand being rewritten to launch with the subcommand swallowed as prompt text; the CLI resolver now skips leading global flags (using the launch parser's value-consumption contract) and dispatches the real subcommand with the flags applied, so ACP mode honors the configured approval policy. (#2970)
Fixed /mcp enable and /mcp disable reconnecting unrelated MCP servers by scoping toggle reconnect/disconnect work to the named server. (#3157)
Stopped the local llama.cpp no-auth login placeholder from being sent as a discovery bearer token.

Removed

Removed /debug dump-next-request command
Removed Wafer Pass from CLI credential help; Wafer Serverless remains available.

[16.1.8] - 2026-06-20

Added

Added "Prose Only Thinking" setting to opt-out of rendering code blocks within AI thinking traces
Replaced code blocks in AI thinking traces with an ellipsis when "Prose Only Thinking" is enabled
Added a live tokens-per-second readout to the thinking pulse, counting only provider-reported tokens (no character estimate, which undercounts a summarized reasoning trace) and hiding itself whenever the rate is zero. Live numbers therefore appear only for providers that report usage mid-turn; otherwise the bare pulse keeps signalling that the model is thinking.
Updated the thinking pulse to a faster starburst whose rotation eases across each cycle (instead of ticking at a fixed rate), with a speed badge that fades from gray to the theme accent as the rate climbs.
Added live state previews for high-value built-in slash commands in TUI autocomplete.
Added a copy affordance for completed /btw answers so users can copy the visible side-answer text before branching or dismissing the panel.

Changed

Updated SnapcompactInline and SDK session context building to support asynchronous rendering
Changed side-channel turns (/btw, /omfg, and IRC auto-replies) to forward the main turn's tool catalog to preserve the prompt-cache layout, while injecting a reminder to suppress tool usage and discarding any generated tool calls.
Changed /btw, /tan, /omfg, /memory, /rename, and /move to save the typed command text to TUI prompt history so they can be recalled with the up arrow.
Changed the temporary model picker to label Alt+P selections as session-only and point users to Alt+M or /model for role model assignment. (#2952)
Replaced new Promise((resolve, reject) => ...) in AsyncDrain with Promise.withResolvers() per the repo's promise-construction convention
Changed the default advisor interrupt cooldown to 3 turns and the default auto-compaction strategy to snapcompact, with non-vision and over-budget snapcompact runs still falling back to context-full summaries.

Fixed

Fixed a TypeError in EventController tests where a missing settings property in the InteractiveModeContext mock object led to calling .get on undefined when streaming tool-call args.
Fixed auto-compaction status text to label snapcompact runs instead of reporting them as context-full maintenance.
Fixed side-channel turns failing to correctly obfuscate secrets and tools when inheriting prompt cache layout
Fixed MCP startup status to live-update from "Connecting…" to connected, still-connecting, or failed server states so completed connections do not leave a stale banner. (#3150)
Fixed session history becoming desynchronized when using the rewind tool
Fixed rewind tool output and temporary assistant side-channel data polluting the prompt cache
Fixed read against a SQLite table with many columns (e.g. 33) rendering every cell as an ellipsis and chopping the right edge. The ASCII table shrinker bottomed out at MIN_COLUMN_WIDTH=1 so every multi-char cell collapsed to …, and the final line truncation then cut off the right side. The renderer now bumps the per-column floor to 3 and falls back to a vertical column: value block layout per row when the column count exceeds the horizontal width budget. (#3107)
Fixed skill://, local://, memory://, and vault:// directory paths so read lists them and search/find can walk their backing directories. (#3116)
Fixed local tiny-model inference failures respawning __omp_worker_tiny_inference for the same failed model, and blocked the unsupported Qwen3 1.7B ONNX memory model before load. (#3132)
Fixed /join failing with timed out waiting for the host's welcome on collab sessions whose existing transcript was more than a few MB. The host now sends a small welcome frame (header + state + agents + entryCount) followed by a train of snapshot-chunk frames (SNAPSHOT_CHUNK_BYTES = 512 KB), and the guest accumulates them under a per-chunk progress timeout that resets on each chunk arrival. The first welcome lands well under one second on the default relay, so the guest's 30s first-welcome budget is no longer spent transferring the snapshot. Requires the new COLLAB_PROTO = 2 on both sides; older hosts/guests are rejected with the existing protocol-mismatch error. (#3144)
Stopped Mnemopi retention from overflowing the embedding model's context window: embed() now caps each input at MNEMOPI_EMBEDDING_MAX_INPUT_CHARS (default 8192 chars) and clips with a head/tail split so a long multi-turn MnemopiSessionState.retainMessages transcript can't make llama.cpp's /embeddings server reject the request with request (N tokens) exceeds the available context size and silently drop vector recall for that memory. The head/tail clip keeps both the opening setup and the most recent turns so later episodes don't collapse onto the same prefix vector (#3126).
Fixed /goal continuation turns that end with a successful yield skipping threshold auto-compaction while the active goal remains over compaction.thresholdTokens. (#3146)
Fixed first-turn memory recall invalidating append-only prompt caches by promoting injected memory into the stable session prompt before the next turn. (#3111)
/resume and omp --resume no longer auto-switch to the "all projects" picker when the current folder has no sessions. The picker now stays scoped to the cwd and surfaces the existing "No sessions in current folder. Press Tab to view all." hint, so users in an empty project never see other projects' session history without asking. (#3099)
Fixed /todo export and /todo import in ACP/text and TUI modes to resolve paths from the active session cwd, accept quoted paths with spaces, and report invalid path schemes without crashing.
Fixed the SDK buildSystemPrompt helper to render caller-provided tools instead of falling back to the default prompt inventory, and to keep no-tools-map skills visible when read is available through the fallback.
Fixed Esc in focused subagent views cancelling active context maintenance (compaction, handoff, and retry operations); pressing Esc now returns to the main view instead, without aborting the maintenance. The active auto-compaction and auto-retry loaders also suppress their "(esc to cancel)" hint while focused to avoid displaying a false affordance. (#2819)
Fixed /omfg saved-state copy to advertise Esc dismiss after saving and registering the rule live.
Fixed legacy Pi extensions importing getModel/getModels from the @oh-my-pi/pi-ai package root failing to load, by restoring them as compatibility aliases for @oh-my-pi/pi-catalog's getBundledModel/getBundledModels. The root StringEnum compatibility shim also accepts TypeScript enum objects in addition to value arrays (#2907).
Fixed the Alt+M model-configuration menu so role assignment remains selectable for models whose context window is smaller than the current session; the context-size disabling still applies to the Alt+P temporary active-model switch (#2861).
Fixed the advisor raising false blockers in plan mode (e.g. "don't write a plan file") because it only saw a 120-char truncation of the injected plan-mode rules, which cut off at NEVER create, edit, or delete files — excep… and hid the "except the single plan file" carve-out. The advisor delta now expands the primary agent's constraint context (plan-mode-context, plan-mode-reference) verbatim inside an XML-escaped <primary-context> wrapper instead of a one-liner, and AdvisorRuntime dedupes the re-injected prompts so an unchanged copy collapses to a marker rather than re-feeding the full rules every turn.
Fixed auto-compaction being suppressed when a before_provider_request extension shrinks the outgoing request below the real stored conversation (e.g. a context-compression proxy such as Headroom, or an aggressive obfuscator). The provider then reports deflated prompt tokens, so the threshold check never fired and the stored history grew unbounded until it overflowed the context window and could no longer be compacted at all. The compaction decision (both the pre-prompt and post-response paths) now floors the provider-reported context tokens by the agent's own local estimate of the stored conversation, so on-wire compression can no longer hide a too-large history from the auto-compactor. Context display and cost accounting still use the exact provider usage; only the compaction trigger takes the floor.
Fixed mnemopi proactive linking being configurable only through the MNEMOPI_PROACTIVE_LINKING environment variable, unlike the sibling mnemopi.polyphonicRecall / mnemopi.enhancedRecall settings: added a mnemopi.proactiveLinking config.yml setting (off by default, /settings → Memory → Mnemopi) that ingests new memories into the episodic graph as they are stored, linking them to related entities and memories; MNEMOPI_PROACTIVE_LINKING still overrides the configured value when set (#2440).

[16.1.7] - 2026-06-20

Fixed

Fixed custom models.yml providers rejecting the compat.supportsImageDetailOriginal override, so Responses-compatible proxies that reject snapcompact's native-resolution image hint can clamp frames to detail: "auto". (#3092)

[16.1.6] - 2026-06-20

Added

Enabled inline prompts with /loop commands (e.g., /loop 10 fix the bug)
Added support for compound duration formats in /loop (e.g., 1h30m)

[16.1.5] - 2026-06-19

Changed

Removed the legacy AgentSession.nextToolChoice() method. The per-turn tool-choice directive now flows solely through nextToolChoiceDirective() (which folds in the hard-choice dequeue plus active-tool filtering as a private helper), eliminating the dual entry point that let callers consume the queue while bypassing the soft pending-preview lifecycle. The underlying ToolChoiceQueue.nextToolChoice() is unchanged.

[16.1.4] - 2026-06-19

Fixed

Fixed omp bench and omp dry-balance failing to resolve models from extension providers
Improved error reporting for omp bench runs that return no output or tokens
Cache-miss marker no longer fires on a cold turn whose predecessor only wrote the prompt cache (never read it back). The session's opening request always writes the prefix with cacheRead 0, so a long-running first tool call (e.g. gh run watch) that outlived the provider's cache TTL surfaced a spurious ⊘ cache miss divider right under the opening message. The marker now requires the previous turn to have actually read a warm prefix, so it flags only a demonstrably working cache going cold — and collapses a run of consecutive cold turns to a single marker at the moment the cache broke.
Fixed omp bench and omp dry-balance failing to resolve models from extension-registered providers (pi.registerProvider(...)). Both commands build a one-shot ModelRegistry that previously only knew built-in catalog providers, so a provider/model selector for a provider contributed by an extension under ~/.omp/agent/extensions/ errored with "Model not found". A new loadCliExtensionProviders helper loads the session's extensions, drains their provider registrations into the registry, and discovers dynamic provider catalogs before resolving selectors — mirroring the interactive session and omp models paths.
omp bench now reports a run that streamed no content and measured no output tokens as a failure ("provider returned no output") instead of a misleading green check with tokens 0 / TPS 0.0.
Cleared stale pending-preview gates when resolve finds no runnable handler, and forwarded the missing peekPendingInvoker and clearPendingInvokers hooks on the production session so staged previews actually reach the resolve tool and the gate drains on dispatch failure. (#3061)
Deferred status-line usage refreshes off the render path and timeboxed the startup fetch so slow Anthropic quota lookups no longer block interactive startup. (#3057)
Fixed image paste placeholders falling through to terminal hyperlink settings before Settings.init(), so early editor rendering falls back to plain text instead of crashing. (#3064)
Fixed omp plugin install github:owner/repo silently keeping the user on a stale commit when re-run on an already-installed GitHub plugin. bun install <spec> respects the existing bun.lock pin when the spec is unchanged and never re-resolves the remote ref, so the manager now follows a git re-install with bun update <name> to refresh the lockfile pin against the upstream. The install transaction also snapshots bun.lock up front and routes feature validation, extension validation, and runtime-config save through one rollback path so a failed install can never leave the rejected commit pinned in the active tree or lockfile. First-time installs are unaffected. (#3063)

[16.1.3] - 2026-06-19

Changed

Refactored Perplexity authentication logic to prioritize cookies over OAuth in search operations
Updated token command to correctly display active Perplexity OAuth tokens when present

Fixed

Enabled auto-retry for AI "thinking loop" errors encountered during model inference
Cleared stale error banners automatically when triggered by an auto-retry recovery phase
Preserved bundled omitMaxOutputTokens policy when fresh cached provider discovery rows replace Ollama Cloud catalog models, so stale models.db entries cannot re-enable context-window-sized num_predict values. (#2984)
Normalized cached-only Ollama Cloud discovery rows to omit on-the-wire output-token caps even when the cached model id has no bundled catalog entry. (#2984)
Fixed Ollama, LM Studio, and llama.cpp (plus loopback vLLM / sglang servers) reprocessing the full prompt on every turn because provider.appendOnlyContext: auto only recognized DeepSeek and Xiaomi as prefix-cache providers. The auto-detect now enables append-only mode for ollama, ollama-cloud, lm-studio, llama.cpp, and any baseUrl resolving to a loopback/RFC1918/.local host, so the system prompt + tool catalogue + prior-turn message bytes stay byte-stable across turns and llama.cpp's KV-cache prefix reuse can hit (#3033).
Isolated mnemopi's local embedding provider in a dedicated Bun.spawn subprocess so onnxruntime-node and fastembed never load into the main agent process. Previously memory.backend: mnemopi crashed Bun on Windows — standalone binaries faulted in the NAPI process.dlopen constructor at session start, npm installs faulted in the NAPI finalizer at process teardown. Mirrors the tiny-model isolation pattern from #1607; the parent SIGKILLs the child on dispose so the destructor never runs in either address space (#3031).
Fixed image tool registration resolving image provider credentials during session startup, so broken or slow google-antigravity OAuth state no longer blocks sessions that never invoke generate_image (#3036).
Fixed LSP client returning -32601 Method not found for defined server→client requests (window/showMessageRequest, window/showDocument, workspace/{semanticTokens,inlayHint,codeLens,codeAction,diagnostic}/refresh). Servers that stall waiting for a real reply (same failure mode as #3029) now receive the spec no-op result (#3044).
Fixed WebP images being sent unchanged to local-server vision models, which can fail through llama.cpp/STB-backed decoders that do not support WebP (#2922).
Made getSettingsListTheme, getEditorTheme, getSelectListTheme, and getSymbolTheme return a plain ASCII fallback instead of crashing with "undefined is not an object (evaluating 'theme.fg')" when the global theme is undefined — e.g. when a plugin calls them before initTheme() completes or from a separate module instance under npm-global installs. (#2998)
Hardened TTS, STT, and tiny-title worker IPC send() paths against async EPIPE rejections: Subprocess.send() is now wrapped so neither a synchronous "process exited" throw nor an asynchronous EPIPE rejection (when the pipe breaks between exit being observed and the next send) can escape as a fatal unhandled rejection. A dying Kokoro/TTS/STT worker can no longer crash the whole agent session mid-task. (#2997)
Fixed Windows test failures caused by path handling: tests now use pathToFileURL, path.resolve, and path.join instead of hard-coded POSIX paths; shortenPath() normalizes backslashes to forward slashes after ~ and respects home directory boundaries; shell-escaped interpolated paths in bash tool tests to prevent Git Bash eating backslashes
Fixed HistoryStorage.resetInstance() leaking its SQLite database handle on Windows by adding a #close() method that finalizes all prepared statements and closes the database; AgentStorage gained the same resetInstance()/#close() pattern
Fixed createAgentSession leaking the internally-created AuthStorage when session construction fails before the session takes ownership, causing EBUSY on Windows temp dir cleanup
Fixed MnemopiBackend.removeDbFiles() throwing on Windows when the database handle is still being released; it is now truly best-effort (logs failures instead of silently swallowing)
Fixed Windows EBUSY test failures by replacing raw fs.rmSync/fs.rm cleanup with TempDir (which retries) and best-effort .catch(() => {}) where SQLite handles outlive the test
Fixed TempDir prefix convention: non-@ prefixes created temp dirs relative to cwd instead of os.tmpdir(), causing module resolution failures on Windows
Fixed git line-ending mismatches in autoresearch tests by setting core.autocrlf false in test repo initialization
Fixed Bedrock inference-profile ARN models being dropped from the allowed-model set when models were scoped via enabledModels, the SDK, or ACP, so an accepted ARN no longer resolves to an empty selection. (#3006)

[16.1.2] - 2026-06-19

Added

Added a welcome-screen tip for the /advisor runtime. Tips ending in a [NEW] marker now render a bold rainbow NEW! tag (it shimmers across the welcome intro's animation frames, then settles into a still rainbow) and are weighted to surface more often in the random tip rotation.

Changed

Renamed the search tool's i parameter to case and inverted its semantics to represent case-sensitive search.
Improved session history to export empty objects as {} instead of empty strings
Refined system prompt and tool documentation to improve conciseness and clarity
Simplified tool input descriptions for browser, eval, find, and memory-edit operations
Refactored authentication storage discovery to share logic with other pi-ai tools

Fixed

Fixed omp bench resolving an ambiguous model selector — a bare or canonical id shared by several providers (e.g. gpt-oss-20b or openai/gpt-oss-20b) — to a provider you have no credentials for. Bench resolves against the full catalog (credentials are ignored), so the default pick was decided by provider-priority order alone. It now redirects such selectors to an equivalent model under a provider with configured auth (honoring modelProviderOrder and canonical cross-provider variants), while an explicit provider/id selector is still benchmarked verbatim so forced/unauthenticated runs keep working.
Resuming a session whose project directory no longer exists (deleted or renamed worktree) no longer crashes with an unhandled ENOENT … chdir rejection. The resume now keeps the current working directory instead of trying to chdir into the missing path, across the in-session selector, the --resume startup picker, and SessionManager.open/continueRecent.
Fixed streaming reflowing Markdown — a fenced mermaid diagram or a GFM table — stranding stale fragments in native scrollback once the reply scrolled past the viewport (cleared only by a full repaint / Ctrl+L). While streaming, the assistant block defaulted to commit-stable, so the transcript advertised its scrolled-off rows as durable snapshot content and the renderer committed an intermediate layout to immutable terminal history; the later re-layout (a diagram reshaping, a table re-aligning its columns) then froze that superseded fragment in scrollback. A still-streaming reply whose Markdown carries a mermaid fence or a table — detected outside fenced code blocks so ordinary code snippets are unaffected — is now commit-unstable, so it stays wholly in the repaintable live region and commits once, at its final layout, when the turn finalizes.
Fixed SYSTEM.md prompt customization going through the raw system prompt override path, which dropped sections rendered by custom-system-prompt.md such as skills and rules (#3014).

[16.1.1] - 2026-06-19

Changed

Migrated legacy macOS power settings to a single sleep prevention enum
Defaulted display.cacheMissMarker setting to false to suppress the cache-miss marker by default.
Replaced the four separate power.preventIdleSleep, power.preventSystemSleep, power.declareUserActive, and power.preventDisplaySleep boolean settings with a single cumulative power.sleepPrevention enum (off → idle → display → system). Each level adds the caffeinate flags of all lower levels. Existing configs are migrated automatically.

Fixed

Cache-miss marker no longer fires on providers with implicit, best-effort prompt caching (Google/Gemini, OpenAI, Fireworks). Those report cacheWrite: 0 and drop cacheRead to zero intermittently as routine propagation noise that self-heals the next turn. Only explicit, prefix-controlled caches (Anthropic/Bedrock cache_control), which re-create the prefix on a cold turn as cacheWrite > 0, now surface a marker — where a zero cacheRead genuinely means the prefix broke.
Fixed advisor dependent settings staying visible while advisor.enabled is off (#3027).
Fixed LSP servers that dynamically register capabilities, including Expert, hanging semantic requests after client/registerCapability was rejected (#3029).

[16.1.0] - 2026-06-19

Added

Added a prompt-cache miss marker: a slim ⊘ cache miss · <n> tokens divider above an assistant turn whose request lost the provider prompt cache. Detected from per-turn usage — the previous turn cached a meaningful prefix but this request read none of it back and reprocessed the prompt (e.g. after a thinking-level / service-tier / model change, tool or system-prompt change, or a history rewrite). Reports the reprocessed token count. Toggle with the display.cacheMissMarker setting (Appearance → Display, on by default).

Changed

Optimized preview workflows by removing forced tool choices, reducing prompt cache invalidations
Unified injected message styling with consistent rounded outlines and icon-tagged headers
Refreshed branch summary messages with a uniform banner style to match compaction points
Updated skill invocation UI with a compact header, home-shortened paths, and dynamic line counts
Refined session context to utilize history blocks instead of raw images for snapcompact summaries
Refreshed the skill-prompt transcript message to match the rest of the TUI: an icon-tagged ✦ skill <name> header (with invocation args trailing), a single meta line with a home-shortened, click-to-open path in the accent color and the prompt size in muted, and a rounded outline around the card. Replaces the old [skill] label with the flat Skill:/Path:/Prompt: key/value dump (which also leaked the absolute home directory).
Refreshed the default custom/hook transcript message frame to a rounded, icon-tagged card: the [customType] bracket label is replaced by an <icon> <type> header (📦 for extension messages, 🪝 for legacy hook-role messages) with a subtle outline, matching the skill card. Covers every extension/hook custom type without a bespoke renderer.
Refreshed the branch-summary transcript message to render as the same slim divider banner as /compact, handoff, and snapcompact (⑂ branch · ctrl+o, summary revealed on expand) instead of a [branch] box.
Optimized network traffic by stripping tool descriptions from provider tool schemas
Renamed the prompt setting from repeatToolDescriptions / "Repeat Tool Descriptions" to inlineToolDescriptors / "Inline Tool Descriptors" and enabled inline descriptors by default.
Snapcompact compaction summaries now reach the model as ordered history blocks instead of one lead-in text block plus appended images: plain text at the oldest edge, an imaged middle, then plain text at the newest edge. This matches the new text-first snapcompact archive layout and preserves chronological order in the provider prompt.
Fixed /dump output repeating the tool inventory twice when inlineToolDescriptors is enabled.
Unified TUI border corners on the rounded style: tool-result frames, overlays, code fences, debug frames, and the interactive bash box now draw rounded corners (╭╮╰╯) to match the editor and message cards, instead of mixing rounded boxes with sharp (┌┐└┘) ones. boxRound now carries the sharp tee/cross junction glyphs (no rounded variant exists), so dividers still honor boxSharp.tee*/cross theme overrides. Markdown tables intentionally keep the fully sharp boxSharp set; its corner tokens now affect tables only.

[16.0.11] - 2026-06-19

Added

Added __advisor.jsonl transcript persistence for advisor model usage attribution and visibility in the Agent Hub
Added defensive reservation against naming a task __advisor to prevent filesystem collisions with internal transcripts
Added mode property to the CompactOptions interface for extension-based compaction control
Updated /compact slash command to support soft, remote, and snapcompact subcommands for per-run strategy overrides
Added /compact mode subcommands so a manual compaction can override the configured compaction.strategy/remoteEnabled for that one run: /compact soft (summarize locally, skip remote endpoints), /compact remote (summarize via the remote endpoint / provider-native compaction; warns and falls back to a local summary when no remote path is available), and /compact snapcompact (archive history onto dense bitmap images, no LLM call). Bare /compact and /compact <focus text> keep their existing behavior; soft/remote still accept trailing focus instructions, snapcompact rejects them. The subcommands are advertised to ACP clients and surface in TUI autocomplete. Extensions can pass the same selection via compact(instructions, { mode }).
The advisor now persists its own turns to a subagent-style transcript (<session>/__advisor.jsonl) so the advisor model's usage is attributed in omp stats and its transcript is observable in the Agent Hub (read-only). The transcript follows session switches (/new, resume, branch) and is drained/released before a /drop deletes the old artifacts dir. The advisor remains a non-peer: it is hidden from agent-facing rosters (irc list, history://, subagent peer prompt, broadcast targets), is not messageable (irc send / collab chat), and is not revivable/killable from the Hub or collab.

Changed

Standardized file elision markers to a compact […Nln elided…] format
Updated transcript viewer to clear cached content if the session file is deleted while opened
Refactored agent hub to use a fullscreen overlay for agent transcripts instead of inline chat
Improved transcript rendering stability by using a non-incremental builder for viewer replays
Advisor transcripts are now excluded from agent-facing surfaces like irc, history://, and peer rosters
Advisor transcripts are read-only and cannot be messaged, revived, or killed via the Agent Hub or IRC
Refined /compact argument parsing to reject focus instructions for modes that do not support them (e.g., snapcompact)
Protocol hosts (RPC/rpc-ui/ACP) now host-default the full advisor settings group — advisor.syncBacklog and advisor.immuneTurns in addition to advisor.enabled/advisor.subagents — so a host that opts the advisor in gets the default tuning instead of inheriting the user's local advisor preferences.

Fixed

Fixed memory-leaking stale transcripts in the agent viewer when underlying files are deleted
Fixed the Agent Hub transcript viewer rendering the transcript body one column right of the "Agent Hub" title (and the title appearing to shift when scrolled to the top): the fullscreen viewer added its own outer gutter on top of the transcript rows, which already carry a 1-column left pad, so the header and body no longer shared a gutter. The viewer now renders the scroll body at full width without the extra gutter, and the file-mention row carries the same 1-column pad as every other row.
Fixed the bash tool failing with pi-natives:command: syntax error at end of input on a valid &&/; chain whose later pipeline stage is a compound command, e.g. echo x && git log | while read h; do …; done | head. The minimizer's segmented-chain runner rebuilds each chain segment from the brush AST via pipeline.to_string() and re-executes that string, but simple_segment only validated the first pipeline stage — so a compound later stage (while/for/if/subshell) was re-serialized without its terminator and re-run as broken shell. Every stage is now required to be a Display-safe simple command, and — as a general guard against the recurring class of brush Display round-trip divergences (previously: quoted here-doc close tags, multi-byte char/byte offsets) — each reconstructed segment is now re-parsed and must match the original pipeline shape before the chain runner executes it; any divergence runs the command whole, unsegmented, instead of corrupting it.
Fixed Ctrl+T (toggle thinking blocks) and the /settings "Hide Thinking Blocks" toggle only collapsing/expanding thinking in the live region: blocks that had scrolled into committed native scrollback on ED3-risk terminals kept their pre-toggle snapshot, so scrolling up showed the old thinking state. Both paths now resetDisplay() after flipping each block's flag, forcing a full clear + replay of the whole transcript (matching the tool-output expansion toggle) so every block above the fold re-renders at its new height.
Fixed ACP mobile voice settings being unable to call speech.models.list by exposing the local STT/TTS model and voice catalog without triggering setup or downloads (#3011).
Fixed the collapsed inline arg preview used by tools without a custom renderer (e.g. advise, MCP tools) truncating every value at a fixed 24 columns, so a long note was cut to note="Your “stric…" even on a wide card with empty space to spare. Each value now grows into the width the card actually has, reserving only a small slice for the keys that still follow so a long leading value can't hide them.

Removed

Removed display.tabWidth setting and configurable tab width support

[16.0.10] - 2026-06-18

Added

Updated the pi option of the tools.format setting to use the new compact sigil-delimited owned tool-calling dialect (§/«…»/¤/‡‡) that uses ~46% fewer tokens than the legacy format on typical calls.
Integrated terminal QR codes directly into /collab and /collab view to display both deep links and scannable codes, and added a collab.webUrl setting for separately hosted collab web clients.

Changed

Updated description of Pi dialect in tool-calling settings to clarify compact sigil format

Fixed

Fixed legacy settings.json, models.json, and keybindings.json migration/loading to accept comments in JSON config files.
Support quoted file paths with spaces (such as default macOS screenshots) in file mentions and @-prefixed CLI arguments (#2909).
Fixed the first steering/follow-up message typed as auto-compaction begins being dropped instead of queued; the compaction abort controller (which backs isCompacting) is now installed before the auto_compaction_start event fires, so input routed during that window lands in the compaction queue rather than the core agent queue.
Fixed queued steering/follow-up messages being silently cleared by the agent reset during a handoff; they are now captured and restored across the new-session reset.
Fixed the /login provider list still offering providers listed in disabledProviders (and their credential-alias logins, e.g. openai-codex-device → openai-codex) as login targets (#2906).
Fixed an owned MCP manager not being disconnected on AgentSession.dispose(), orphaning stdio MCP subprocesses across session teardown (/exit, print mode, subagent teardown); the dispose-time disconnect is now bounded so a slow HTTP/SSE transport close cannot stall exit (#2839).
Fixed inspect_image resolving pasted image labels such as Image #1, [Image #1, WxH], and attachment://1 against current chat attachments instead of falling back to <cwd>/Image #1 (#2787).
Accepted max as an alias for the top thinking level (xhigh) in selectors and --thinking, so DeepSeek V4 Pro can be selected with its provider-facing maximum effort (#2727).
Fixed --provider=amazon-bedrock --model <application-inference-profile ARN> being rejected as an unknown model before the Bedrock provider could send the ARN to Converse Stream. (#3004)

[16.0.9] - 2026-06-18

Added

Added a --print-thoughts flag for single-shot print mode that includes sanitized thinking blocks in the text output; default print output stays the final answer only. When thinking is hidden by the hideThinkingBlock setting, --print-thoughts un-hides it for the print run so the flag is not a silent no-op (an explicit --hide-thinking still wins).

Fixed

Fixed active /goal mode being paused by internal compaction and session-switch lifecycle aborts, and made those switches persist wall-clock goal usage without charging time spent in another session to a preserved goal.

[16.0.8] - 2026-06-18

Changed

Optimized app startup time by lazily loading site-specific scraper modules
Refactored internal archive handling into a unified src/utils/zip.ts module
Centralized all fflate (ZIP) and Bun.Archive (tar/tar.gz) operations into zip.ts
Optimized archive reading by using lazy, ranged central-directory access for ZIP files
Updated internal image processing to no longer include metadata text for fetched images
Optimized omp:// documentation indexing by compressing doc bodies into a lazily-inflated blob
Changed Mermaid fenced-block ASCII rendering to use the first-party vendored renderer in @oh-my-pi/pi-utils (src/vendor/mermaid-ascii), dropping the beautiful-mermaid npm package, its transitive elkjs (~3.13MB), and the beautiful-mermaid bun patch; CJK/emoji width handling and the layout-direction override are preserved.
Changed omp:// documentation embedding to a gzipped base64 index (docs-index.generated.txt, populated at build time and reset afterward) inflated on first read, instead of a ~1.6MB raw TypeScript map. The compiled binary / npm bundle drops ~0.9MB; the dev tree and source checkouts read docs/ from disk.
Replaced the markit-ai package with a vendored in-house document engine (src/markit) for the document formats it converts: PDF (via mupdf), DOCX (via mammoth), and PPTX/XLSX/EPUB (via fast-xml-parser). Conversion output is preserved, including the PDF column/table-detection pipeline and HTML-table normalization; markit-ai's unused converters and their dependency tail are gone. Legacy .doc/.ppt/.xls/.rtf remain unsupported (a conversion error), as before.
Centralized ZIP handling behind a single src/utils/zip.ts (fflate): the new document converters, the write tool's in-place archive editing, and the read tool's ranged archive reader now share one ZIP implementation instead of mixing jszip and fflate.

Fixed

Fixed settings overlay crash when scrolling past the last row in list views
Improved tool result formatting by correctly wrapping <out> blocks in dim-ink toggles
Fixed auto-retry after transient model-stream socket closes to replay text/thinking-only partial assistant output, including turns where incomplete tool-call arguments were dropped; completed tool calls still block retry to avoid duplicating tool execution.
Fixed omp update reporting EPERM: operation not permitted, unlink '<binary>.bak' on Windows when self-replacing a standalone binary, even though the new binary had already been installed. The backed-up old executable is still the running process image and cannot be unlinked until the process exits, so the post-verify backup cleanup is now best-effort, backups use a unique per-attempt name, and stale backups are swept on the next update (#845).
Fixed plan-mode Refine plan so the internal approval abort is hidden and the editor is ready for a follow-up prompt instead of showing Operation aborted (#2971).
Fixed TUI prompts beginning with shell-style variables such as $HOME being misrouted to Python eval; Python shortcuts now require $ <code> or $$ <code>. (#2944)
Fixed LM Studio runtime discovery to use native /api/v0/models metadata so inspect_image can select VLM models. (#2945)
Fixed omp crashing at startup with Cannot find module './browser-data/browser-data.js' (e.g. on Termux/proot Linux arm64). tools/browser/attach.ts value-imported the TargetType enum from puppeteer-core, which dragged the puppeteer-core barrel (and its @puppeteer/browsers Chromium downloader) back onto the eager startup import graph — defeating the lazy-load deferral and turning any bundling quirk in that subtree into a fatal boot crash instead of a recoverable browser-tool error. The import is now import type and the target check compares against the "page" literal, so puppeteer only loads on first browser use.
Fixed the standalone compiled binary aborting at startup with error: Cannot find package 'mupdf'. The vendored document engine kept mupdf external, but a single-file bun --compile binary has no node_modules to resolve it from, and the bundler resolves the otherwise-lazy import eagerly at boot. mupdf (and its WASM module, embedded at build time via scripts/embed-mupdf-wasm.ts) is now bundled into the binary, so startup and PDF/document conversion both work in the standalone binary; npm and source installs still load mupdf from node_modules. mupdf is also imported lazily inside the PDF converter now, which fixes a compiled-bundle init-order hazard — mupdf's top-level await made the bundled markit chunk's module init async, exposing the converters before their extension tables initialized — and keeps the ~10MB WASM off non-PDF document conversions.

Removed

Removed the markit-ai, exifr, music-metadata, and direct jszip dependencies. As a side effect, fetching an image or audio URL no longer appends EXIF/audio metadata text; image inlining and resizing are unchanged, and document conversion (PDF/DOCX/PPTX/XLSX/EPUB) is unaffected.

[16.0.7] - 2026-06-18

Fixed

Fixed /model in the TUI to open the model setup picker again, leaving /switch as the temporary session model switcher (#2933).
Fixed OpenCode Go sessions recording per-request cost history so /usage can show local cap utilization. (#2942)

[16.0.6] - 2026-06-18

Added

Added model.loopGuard.enabled (default true) and model.loopGuard.checkAssistantContent (default true) settings to configure thinking and assistant prose loop detection.
Added explicit ArkType schema descriptions to parameters across all agent tools to improve model tool-calling instructions and parameter guidance
Added support for OpenRouter fallback in Perplexity web search when direct Perplexity API keys fail or are unavailable
Added support for streaming the Perplexity Responses API (/v1/responses) via the PI_PERPLEXITY_RESPONSES=1 environment variable
Added omp ttsr top-level CLI command to inspect and test Time-Traveling Stream Rules
Added omp ttsr list to enumerate all project/user-loaded TTSR rules with their conditions, scope, and source metadata
Added omp ttsr test to run snippets through the real TTSR matching pipeline with inline text, --file <path>, or stdin via --file -
Added --json output to omp ttsr test and omp ttsr list for machine-readable reporting
Added --rule, --source, --tool, --path, and --verbose options to omp ttsr test to control matching context and inspection details
Added omp ttsr subcommand for inspecting and testing Time-Traveling Stream Rules: omp ttsr list shows every TTSR-registered rule the current project/user config would load, and omp ttsr test feeds a snippet (inline, --file, or stdin) through the real TTSR matching pipeline (TtsrManager.checkSnapshot/checkAstSnapshot) and reports which rules would trigger. A positional that resolves to a file defaults to tool/edit context; --source, --tool, and --path override the inferred match context so glob/AST/scope-scoped rules evaluate the same way they do in a live session. --rule tests a single rule markdown file in isolation.
Added support for reading embedded PDF images via read <pdf>:<image>.png and listing available image members with read <pdf>:
Added a built-in ts-no-inline-cast-access TTSR rule that interrupts inline object-type assertions read immediately as a property ((x as { y: T }).y, including ?. and bracket access), steering toward schema validation, in/typeof narrowing, or a validated named type
Added startup.showSplash (default false) to show the full setup splash animation on normal interactive startup while startup.quiet still suppresses startup chrome. (#2880)
Added app.retry as an Alt+R keybinding for retrying the last failed or aborted assistant turn (#2790).
Added b branch promotion for completed /btw answers, creating a branch that preserves the side-question input and full assistant response including thinking blocks.

Changed

Enabled caching by default in the codebase search tool to improve search performance
Replaced internal schema validation and @sinclair/typebox polyfills across all agent tools and configurations from Zod to ArkType
Changed advisor model calls and overflow-compaction tasks to inherit and propagate primary telemetry spans, usage, and cost tracking
Changed PDF read output to replace  placeholders with clickable read <pdf>:<image>.png handles, including line-range and multi-range reads
Changed the built-in ts-no-any rule to recommend a schema parse at trust boundaries and in-narrowing (instead of an inline as-cast) when reading fields off unknown

Fixed

Fixed legacy plugin validation for extensions that import defineTool, StringEnum, frontmatter helpers, SettingsManager, createCodingTools, or the bare typebox package through the hosted Pi compatibility shims (#2858).
Fixed edit seen-line guard mismatch assertion message formatting to report the actual state instead of generic failure notices
Fixed hashline edit mode rendering in the TUI when setIgnoreTight triggers synchronous display rebuilds in the constructor before #editMode is assigned, which previously caused the tool envelope target path to display as … instead of the parsed hashline filename.
Fixed omp ttsr scan to discover files with gitignore-aware native globbing, skip binary/oversized files before text decoding, use a scan-specific matcher, keep default output summary-only, and avoid retaining per-file AST snapshots during scans
Fixed Perplexity web search to use shared OpenAI streaming transports while preserving streamed sources, citations, and related questions
Fixed StatusLineComponent fire-and-forget async callbacks (#isDefaultBranch, #lookupPr, fs.watch) firing #onBranchChange after dispose(), which reached the global settings proxy after tests called resetSettingsForTest() and threw "Settings not initialized" between test files; dispose() now sets a disposed flag, clears #onBranchChange, and every post-await continuation checks the flag before touching settings or the callback
Fixed read <pdf>:<member> errors for unknown PDF images to surface available extracted image names
Fixed puppeteer stealth scripts to use cached Reflect methods (Reflect_get, Reflect_apply) and Reflect.apply instead of live Reflect/Function.prototype.apply calls, preventing page tampering from leaking through proxy traps.
Fixed Perplexity API-key web search to use shared OpenAI streaming transports while preserving streamed sources and OpenRouter fallback.
Fixed subagents reporting success after a provider-error turn by preserving real run failures over earlier successful yield payloads, and retried bare OpenAI-compatible finish_reason: "error" provider failures after partial text instead of stopping immediately
Fixed MCP servers that do not implement resources/templates/list (JSON-RPC -32601) discarding their concrete resources; templates now fall back to an empty list (#2838 by @jms830)
Fixed provider setup sign-in URLs to attempt clipboard/OSC 52 copy and expose an Alt+C retry shortcut, so authentication is not blocked when TUI selection is unavailable (#2908).
Fixed ACP approval-mode documentation to describe config inheritance, omp acp --yolo/--auto-approve runtime overrides, client permission precedence, and headless prompt behavior (#2900).
Fixed /guided-goal to fall back to the current session model when neither the plan nor slow role resolves, instead of aborting during setup (#2855).
Fixed auto context-full maintenance to stop retrying the same summarization timeout before falling back to the next compaction model (#2913).
Fixed /plan <prompt> and /goal <objective> to preserve the typed slash-command line in TUI input history when entering those modes from off (#2887).
Fixed /model in the TUI to open the active-session model switcher instead of the role-assignment picker (#2846).
Fixed Perplexity web search collapsing every upstream failure to a generic 401 No authentication method available once all auth methods failed: the fallback loop now rethrows the last classified provider error (402/credits-exhausted, 429, 5xx), so quota and rate-limit failures are no longer mis-reported as authorization errors. The generic 401 is now only a defensive fallback for the no-method-ran case.

Security

Secured PDF image reads by validating requested image members against the extracted member list before opening files and refusing traversal-style names

[16.0.5] - 2026-06-17

Added

Added tui.tight setting (default false) to enable tight layout by removing the 1-character horizontal padding from terminal output.
Added a providers.antigravityEndpoint setting (auto, production, sandbox) to control google-antigravity routing for chat, search, image, and discovery calls
Added automatic endpoint-mode support for google-antigravity provider calls so users can force production-only or sandbox-only usage
Added images.describeForTextModels option (default true) to control automatic image description for attachments sent to models without vision input
Added automatic vision fallback prompts to describe images for text-only models
Added advisor.immuneTurns setting (default 1) to limit how often advisor concern/blocker notes can interrupt the primary agent.
Added a main-session session_stop extension event with continuation feedback and an 8-continuation loop cap (#2834).
Added --max-time <seconds> so CLI sessions can stop after a wall-clock deadline.

Changed

Changed google-antigravity usage report lookups to honor the selected antigravity endpoint mode when resolving the reporting base URL
Changed context usage reporting to always return numeric token counts and percentages, so status-line and footer now show estimated values instead of ? immediately after compaction
Changed context usage reporting to use anchored snapshots and pending-prompts estimates, which now keeps /context, status line, and model selector token counts in sync

Fixed

Fixed Matplotlib figure display to emit PNG output immediately when display(fig) is called, even if the figure is closed before the end-of-cell flush
Fixed persisted tool-result image payloads in details.images to externalize and resolve through the session blob store, so generated-image details survive resume without stale blob refs or truncation
Fixed duplicate Matplotlib image output by skipping the automatic end-of-request figure flush for figures that were already displayed through display(fig)
Fixed google-antigravity image generation and web search requests to fail over to the alternate antigravity endpoint on 429/server/network failures instead of stopping at the first endpoint
Fixed context usage breakdown to use a completed assistant usage anchor from the current turn instead of a pending prompt snapshot so totals no longer overcount when a large in-turn tool step returns usage
Fixed side-channel turns and advisor requests to keep using credential resolvers during retries, so Google Resource exhausted 429s can rotate to the next account instead of surfacing a terminal error banner
Fixed context token accounting to keep branch-local anchors during branching so sibling-branch messages no longer pollute context estimates
Fixed context usage consistency so /context, status line, and idle compaction logic now report the same used-token totals
Fixed status-line context cache invalidation when assistant reasoning signature data grows so displayed context usage updates accurately
Fixed the status-line context% reading inflated during long tool turns and then dropping sharply on the next message even though no compaction ran. While a request was in flight getContextBreakdown summed a cl100k estimate of the entire tail on top of the stale turn-start prompt and never re-anchored to completed in-turn steps; it now prefers the real provider prompt-token count of any step that resolves at or after the pending cutoff. The status-line memo also keys on a contextUsageRevision that bumps when the in-flight snapshot is set/cleared, so a mid-turn estimate is invalidated on turn end/abort instead of surviving into idle until the next message
Fixed image attachment handling for text-only models by saving attachments to local:// and injecting generated descriptions so they are no longer lost when the target model cannot process images
Fixed the ssh tool rejecting valid Windows identity files before invoking OpenSSH by skipping Unix mode-bit key validation on native Windows (#2850).
Fixed web_search/omp q aborting before any provider ran when the global Settings singleton was not initialized; executeSearch now reads providers.antigravityEndpoint once and tolerates an uninitialized settings store instead of throwing
Fixed the new git.enabled and images.describeForTextModels settings declaring section groups (Git, Vision) that were not registered in TAB_GROUPS, so they now render in their intended settings-panel sections
Fixed Python display(fig) for Matplotlib figures to emit PNG output immediately, even when user code closes the figure before the end-of-cell flush.
Fixed persisted tool-result image payloads stored in details.images to externalize and resolve through the session blob store, so generated-image details survive resume without stale blob refs or truncation.
Fixed the tools.format setting schema so minimax can be selected as an owned tool-calling dialect, and taught auto mode to route tool-less MiniMax-family models to the MiniMax owned dialect. (#2759)
Fixed WSL2 TUI stutter by adding a git.enabled setting and skipping footer/status-line git probes when disabled or when no git-backed status segment is visible (#2847).
Fixed JSON-mode startup notices (export/resume/session-picker messages) writing to stdout before the JSON event stream; they now route to stderr so stdout remains newline-delimited JSON.

[16.0.4] - 2026-06-17

Fixed

Fixed RPC/ACP startup forcing todo settings back to host defaults, so project-level todo.enabled, todo.reminders, and todo.eager opt-outs now suppress protocol-mode todo prompt injection; enabled todo reminders are now persisted to the JSONL transcript so the log matches the model-visible context (#2824).
Fixed default prompts to instruct the agent to read applicable skill://<name> content before starting work, so discovered skills influence broad task requests like frontend generation (#2829).
Fixed hashline visible-line validation for ACP editor reads so INS.POST anchors displayed by bridge-backed range and multi-range read output are merged into the session snapshot before edit validates them (#2773).

[16.0.3] - 2026-06-16

Added

Added support for LaTeX color commands (\textcolor, \colorbox, and \fcolorbox) in user-visible terminal prose and final chat to colorize output

Changed

Changed STT dependency setup to validate recorder and model assets per stt.modelName, so switching speech models re-runs dependency checks and downloads for the new model
Changed STT startup with cached models to warm the speech model in the background and defer full model loading until transcription begins, reducing push-to-talk start latency
Allowed user-visible terminal and final-chat responses to include LaTeX math delimiters/commands and Mermaid ```mermaid diagrams
Changed the hold-Space push-to-talk gesture to recognize a held bar from the regularity of the OS key auto-repeat rather than a raw space count or speed alone, so it no longer spams the editor, no longer eats deliberate space taps, and no longer triggers when the bar is smashed. Recording starts only after two consecutive inter-space deltas are "mechanical" — both fast (within ~120 ms) and near-identical, the metronomic signature of auto-repeat; the few pre-burst spaces typed are then tracked back out. Smashing (fast but jittery) and deliberate spacing (steady but slow) both keep typing real spaces and never start recording.
Updated markdown Mermaid rendering to color ASCII diagrams with the active theme and automatically choose a narrower layout that better fits the terminal width
Made the watched-session transcript sent to the advisor (and shown by /advisor dump) clearer: each turn now opens with a ### Session update heading; watched-agent roles render as inline **agent**: / **user**: labels instead of level-2 headings that collided with the advisor's own turns; consecutive same-role messages collapse under one label (the watched agent emits one assistant message per tool call); and batched updates are joined by a blank line rather than a --- rule.
Changed the compact transcript tool-intent prefix (history://, /advisor dump) from # to // so intent lines read as comments instead of rendering as Markdown H1 headings.
Changed the advisor advice injected into the primary transcript from a Advisor (...): - [severity] note prose block to one <advisory severity="…" guidance="weigh, don't blindly obey">…</advisory> element per note, with XML-escaped bodies. (Relocated the shared escapeXmlText helper to @oh-my-pi/pi-utils.)
Reverted /dump and /advisor dump raw to the pre-16.x full verbose dump: system prompt, model/thinking config, tool inventory with parameters, and the message transcript rendered with markdown role headings (## User, ## Assistant, ### Tool Call: <name> with the call's _i intent as a // comment under the heading and the remaining arguments as a fenced YAML block, ### Tool Result: <name>, plus ## Bash Execution/## File Mention/summary sections) instead of the model's native-dialect turn envelopes and <invoke>/<parameter> XML tool calls. Dropped the compact default and the [raw] flag on /dump; the compact → tool(...) ⇒ ok history format is no longer reachable from /dump. /advisor dump still defaults to compact, and /advisor dump raw now renders the same markdown dump (previously the model's native-dialect envelopes).

Fixed

Fixed Whisper STT cache detection to require both encoder and decoder .onnx files, so partial model downloads now trigger a proper foreground download instead of being treated as fully cached
Fixed same-process JsRuntime cleanup so disposing an older inline/direct runtime no longer deletes a newer runtime's JS helper globals; inactive cmux/direct runtimes now re-activate their globals before sequential use while overlapping cross-runtime runs fail explicitly.
Fixed magic-keyword steering notices (ultrathink-notice, orchestrate-notice, workflow-notice) to be prepended before the related user message so they influence that same turn
Fixed dequeuing or popping queued user messages to remove their preceding hidden magic-keyword notice companions, preventing orphaned queued notices
Fixed queued user steers to auto-resume after interrupts even when the transcript tail is a preserved advisor card or other non-conversational custom message
Fixed queued user follow-up messages to remain queued after an interrupt and only run on explicit resume, even when an IRC wake leaves a provider-valid tail
Fixed stranded IRC asides to wake a response turn after interruption instead of remaining pending
Fixed accepted IRC asides to be flushed into the transcript during disposal instead of being discarded
Fixed interactive submissions made while the TUI had no active input waiter: they now start a real prompt directly, with steer fallback if a background turn races in, instead of queueing behind a non-resumable idle transcript and appearing to do nothing.
Fixed pressing Esc (or Alt+Up dequeue) while agent-authored messages were queued — advisor concern/blocker notes, hidden goal/plan/budget steers, IRC/extension asides — dumping their text into the user's editor. Editor restoration (clearQueue()), pending chips (getQueuedMessages()), and popLastQueuedMessage() now surface only genuinely user-authored queued messages (plain user turns and attribution: "user" custom messages like /skill). Plain Alt+Up dequeue leaves all other queued messages in place for the continuing stream; only the Esc interrupt path keeps just advisor cards (so abort's preservation still re-records them as visible advice) and drops other internal steers, so a user interrupt can't be silently undone by an auto-resume on leftover internal context. queuedMessageCount still reflects all actual queued work (advisor cards included) so hasPendingMessages()/RPC and the empty-submit abort gate stay accurate.
Fixed advisor concern/blocker advice being withheld from the running agent and then dumped as one burst at the next user prompt after a deliberate interrupt. A user interrupt latches advisor auto-resume suppression, but a non-user resume (synthetic/auto-continue, or a queued steer draining after the abort) leaves the run streaming with that latch still set, so every interrupting note was parked hidden in the next-turn queue instead of steered into the live turn — the agent never heard the advisor mid-run and the backlog flushed all at once on the next prompt. Suppression now only withholds interrupting advice while the agent is idle (or still tearing the interrupted turn down); once a turn is streaming again the note is steered in live, since steering an active run never auto-resumes a stopped one. A concern that strands in the steer queue past the resumed turn's final poll is reclaimed as visible advice when the agent settles (mirroring abort), so it neither auto-resumes the stopped run nor lingers to flush at the next prompt.
Fixed omp --continue/-c sometimes resuming into a subagent transcript instead of the interactive session. Subagent (and HTML-export) SessionManager.open() calls run in the parent's terminal and were clobbering the per-TTY --continue breadcrumb with their own artifact-dir session file; these headless opens now suppress the breadcrumb. continueRecent() also recovers already-poisoned breadcrumbs by resolving any session file inside a parent's artifacts dir (<parent>/<agentId>.jsonl) back up to the top-level session.
Fixed the Agent Hub stacking duplicate Agent Hub · N running frames and stranding garbage rows in scrollback while navigating with subagents still streaming. The hub was a non-fullscreen overlay composited over a live transcript, so each time a running subagent's progress grew the frame and scrolled the window the previously-painted hub copy was pushed permanently into the terminal's native scrollback (which the engine can't rewrite). It now renders inline in the editor slot — the same anchored region every other selector and the ask tool use — riding the normal append-only commit path, so the transcript commits above it exactly once and the hub repaints in place instead of leaking copies. (Avoids borrowing the alternate screen.)
Fixed every subagent registering itself as its own parent in the agent registry (parentId === id), so the Agent Hub rendered each agent as sub · of <itself> and the ←← parent-navigation gesture looped on the same agent. The SDK was reusing parentTaskPrefix — the agent's own artifact/output-id prefix — as the registry parent link; spawns now pass a separate parentAgentId (the spawning agent's id: Main for top-level task spawns, the parent subagent for nested spawns and eval agent(), the focused agent for /tan) and the registry records that as the parent.
Fixed messaging a parked subagent that was restored from disk (Agent Hub scan, or a resumed/restarted session) failing with cannot be revived (no reviver registered) even though its transcript was intact. Such refs carry a session file but no in-memory reviver — the executor's live reviver closure dies with the spawning turn/process — so IRC sends and Agent Hub focus refused them. AgentLifecycleManager.ensureLive now cold-revives them through a persisted-subagent reviver factory (installed by the top-level interactive/RPC session) that rebuilds the subagent from its JSONL the way --resume rebuilds a session: it reopens the file and replays it through createAgentSession, but sources the runtime contract from a now-readable session_init record (SessionManager.peekSessionInit) so tools, system prompt, output schema, and kind are restored rather than resurrected as a default top-level session. session_init now also persists the effective spawns allowlist and read-summarization flag so a cold revive keeps the original capability surface (old files without them deny re-spawning rather than defaulting to wildcard). Isolated runs and pre-session_init files whose recorded workspace no longer exists stay transcript-only (history://).
Fixed the terminal window-title OSC writes (setTerminalTitle/pushTerminalTitle/popTerminalTitle) leaking escape sequences to a developer's terminal during bun test; they now skip when the terminal is headless (the test-runtime default), matching the ProcessTerminal render/probe suppression so interactive-mode tests no longer paint to the real terminal
Fixed empty CLI sessions being retained after opening omp and exiting without a prompt (#2800).
Fixed hooks/pre/*.ts and hooks/post/*.ts files discovered through hookCapability being registered in discovery but never loaded into the extension runner, so their tool_call handlers now run without a manual settings.json extensions entry (#2796).
Fixed startup model fallback choosing the plain OpenAI gpt-5.5 provider before the Codex OAuth provider when both shared the same default model id, which could surface a misleading OpenAI 401 despite valid Codex credentials (#2807).
Fixed local auto-thinking classification for reasoning-capable tiny models by giving them the same safe answer budget as online reasoning classifiers, with a larger local floor for non-reasoning tiny models (#2808).

Removed

Removed the built-in render_mermaid tool and its renderMermaid.enabled setting, so it can no longer be invoked directly

[16.0.2] - 2026-06-16

Added

Added the UMANS_WEBSEARCH_PROVIDER environment variable to CLI help for Umans gateway web search backend selection.

Changed

The eager task (task.eager: always) and eager todo (todo.eager: preferred/always) hidden reminders now re-fire on the auto-continuation turn after a compaction (context-full / snapcompact / handoff / shake). Compaction summarizes away the first-message prelude, so the agent would otherwise silently lose the delegate-via-tasks / phased-todo guidance mid-work; the post-compaction todo nudge is reminder-only and never forces the todo tool onto the resumed turn.

Fixed

Fixed edit-tool block operations on Emacs Lisp files: .el and .emacs paths now resolve top-level forms for SWAP.BLK, DEL.BLK, and INS.BLK.POST instead of reporting an unsupported-language block-resolution error.
Fixed PDF reads leaking recoverable MuPDF WASM warnings into the terminal TUI by routing MuPDF output through the file logger before markit-ai loads it (#2766).
Fixed /exit and /quit waiting one shutdown timeout per hanging extension by running session_shutdown handlers within a shared shutdown window (#2736).
Fixed GitHub Copilot .github/instructions/*.instructions.md discovery by loading those files as rules that honor applyTo scoping, including always-apply ** files and rule://<name> access for glob-scoped entries (#2731).
Fixed Windows bash-tool child processes defaulting interpreter pipe I/O to the ANSI codepage by adding UTF-8 encoding defaults when the inherited environment is unset (#2701).
Fixed /advisor dump raw so Opus 4.5 thinking content that already includes literal <thinking> tags is not rendered with nested thinking tags (#2700).
The plugin-extensions-discovery test suite no longer writes fixtures into — and rm -rfs the node_modules of — the developer's real ~/.omp/plugins. Its XDG_DATA_HOME isolation was a no-op on Windows (XDG is gated to Linux/macOS) and was bypassed in XDG-migrated Linux/macOS environments, so a local run could delete installed plugins. The suite now isolates the whole config root via an os.homedir() mock plus cleared XDG_* vars, with a pre-write guard that fails if resolution escapes the temp home (#2721).
Installed plugins whose extensions manifest entry points at a directory of sub-extensions (the standard pi extensions/<name>/index.ts layout, e.g. pi.extensions: ["./extensions"]) are no longer rejected at install (declared extension entry not found on disk) or silently dropped at load. The plugin manifest resolver now resolves a directory the same way as the configured-directory (-e) extension loader: the directory's own package.json omp/pi extensions (authoritative — a missing declared entry is reported instead of falling back to a decoy index), then a direct index.{ts,js,mjs,cjs}, then a one-level scan of sub-extensions (#2713).
Fixed OpenRouter @upstream routing selectors whose upstream slug also appears in the model id, so openrouter/...@deepseek:high keeps openRouterRouting.only instead of being consumed by provider-scoped fuzzy matching (#2708).
Fixed omp plugin list --json omitting locally linked plugins that exist only in omp-plugins.lock.json and node_modules symlinks. (#2742)
Fixed task subagents to install their configured ordered model candidates as child-session retry fallback chains, so retryable provider failures can advance to the next subagent model instead of failing the worker (#2750).
Fixed empty reasonless aborted assistant turns to auto-retry without switching model fallback, so transient provider-side aborts after tool results do not end headless sessions (#2685).

[16.0.1] - 2026-06-15

Breaking Changes

Settings: hooks and customTools arrays replaced with single extensions array
CLI: --hook and --tool flags replaced with --extension / -e
Directories: hooks/, tools/ → extensions/; commands/ → prompts/
Types: See type renames above
SDK: See SDK migration above
Key detection functions removed from @mariozechner/pi-tui: All isXxx() key detection functions (isEnter(), isEscape(), isCtrlC(), etc.) have been removed. Use matchesKey(data, keyId) instead (e.g., matchesKey(data, "enter"), matchesKey(data, "ctrl+c")). This affects hooks and custom tools that use ctx.ui.custom() with keyboard input handling. (#405)

Added

Added GitHub Copilot user-global discovery to the github provider: it now loads user-global instructions from ~/.copilot/copilot-instructions.md, honors the COPILOT_HOME relocation override, reads each directory listed in COPILOT_CUSTOM_INSTRUCTIONS_DIRS for an AGENTS.md and .github/instructions/**/*.instructions.md (matching Copilot CLI), scans the project .github/instructions/ tree recursively, and surfaces VS Code Copilot prompt files (*.prompt.md) from .github/prompts/ as slash commands. Previously only the project .github/ tree was scanned, so Copilot CLI users' cross-repo config was silently ignored. Closes #1913, #1915, #1916.
Added web search provider exclusions so web_search can skip configured providers without disabling them for model use (#2608).
Added an optional role field to task spawns that gives each subagent a tailored specialist identity: the role is injected as a system-prompt specialization preamble and becomes the subagent's display name and telemetry identity in the registry, IRC roster, and Agent Hub, so delegated trees are no longer clones of one generic worker (#2467)
Added a non-blocking specialization advisory to the task tool: when a spawner with remaining depth capacity spawns generic role-less workers (a task/quick_task spawn without a role, or the same agent cloned ≥2× all without roles), the result steers it toward tailored specialists. Gated so a leaf at max recursion depth is never nudged; the task-tool depth gate is extracted into a shared canSpawnAtDepth helper (#2469)
Added a work-aware IRC roster: irc list and the subagent peer roster now show each peer's current activity (a short gist of its latest intent/tool) alongside its role-derived display name, so a tree of tailored specialists is legible enough to coordinate. Backed by a new display-only activity field on the agent registry (#2470)
Added proactive IRC coordination: when one task call spawns ≥2 live siblings with spawn capacity remaining, the result suggests coordinating overlapping work via irc (composed onto the specialization advisory), and the subagent COOP/irc prompts now actively steer discovery (list), coordination (message before overlapping edits), and follow-up (replyTo/await) instead of only assuming agents resolve collisions on their own (#2471)
Added /review support for explicit GitHub pull request URLs and detected PR URLs from recent conversation context (#1743).
Added a plan.defaultOnStartup setting (Tasks tab, default off) that automatically enters plan mode at the start of a new interactive session. It applies to any session that opens with no prior history — including a --continue/auto-resume that finds no recent session and starts fresh — while sessions with restored history keep the mode reconciled from that history. It is a silent no-op when plan mode is globally disabled (plan.enabled), and the setting is hidden from the settings UI unless plan mode is enabled.
Added UMANS_AI_CODING_PLAN_API_KEY to the CLI environment help (#2636 by @oldschoola).
Added WebSearchProviderError class with HTTP status for actionable provider error messages
$ARGUMENTS syntax for custom slash commands as alternative to $@ for all arguments joined. Aligns with patterns used by Claude, Codex, and OpenCode. Both syntaxes remain fully supported. (#418 by @skuridin)
Configurable double-escape action: choose whether double-escape with empty editor opens /tree (default) or /branch. Configure via /settings or doubleEscapeAction in settings.json (#404)
Vertex AI provider (google-vertex): access Gemini models via Google Cloud Vertex AI using Application Default Credentials (#300 by @default-anton)
Built-in provider overrides in models.json: override just baseUrl to route a built-in provider through a proxy while keeping all its models, or define models to fully replace the provider (#406 by @yevhen)
Automatic image resizing: images larger than 2000x2000 are resized for better model compatibility. Original dimensions are injected into the prompt. Controlled via /settings or images.autoResize in settings.json. (#402 by @mitsuhiko)
Alt+Enter keybind to queue follow-up messages while agent is streaming
Theme and ThemeColor types now exported for hooks using ctx.ui.custom()
Terminal window title now displays "pi - dirname" to identify which project session you're in (#407 by @kaofelix)
Hook API: ctx.ui.setTitle(title) allows hooks to set the terminal window/tab title (#446 by @aliou)
Hook API: before_agent_start handlers can now return systemPromptAppend to dynamically append text to the system prompt for that turn. Multiple hooks' appends are concatenated.
Hook API: before_agent_start handlers can now return multiple messages (all are injected, not just the first)
New example hook: tools.ts - Interactive /tools command to enable/disable tools with session persistence
New example hook: pirate.ts - Demonstrates systemPromptAppend to make the agent speak like a pirate
Tool registry now contains all built-in tools (read, bash, edit, write, grep, find, ls) even when --tools limits the initially active set. Hooks can enable any tool from the registry via pi.setActiveTools().
System prompt now automatically rebuilds when tools change via setActiveTools(), updating tool descriptions and guidelines to match the new tool set
Hook errors now display full stack traces for easier debugging
Clipboard image paste support via Ctrl+V. Images are saved to a temp file and attached to the message. Works on macOS, Windows, and Linux (X11). (#419)
Configurable keybindings via ~/.pi/agent/keybindings.json. All keyboard shortcuts (editor navigation, deletion, app actions like model cycling, etc.) can now be customized. Supports multiple bindings per action. (#405 by @hjanuschka)
/quit and /exit slash commands to gracefully exit the application. Unlike double Ctrl+C, these properly await hook and custom tool cleanup handlers before exiting. (#426 by @ben-vargas)

Changed

Enriched the bundled designer, reviewer, and task agent prompts: designer gains a token-first four-phase design-system workflow (analyze → build-if-missing → compose-with-tokens → verify), reviewer gains an evidence standard (a finding is not real until you can name the triggering input; passing tests are not proof of correctness), and task gains an evidence-bound completion requirement.
Updated the session-continuation tip to mention Zellij and WezTerm panes.
Documented the role task parameter in the task-tool description (both the batch and single-spawn shapes) and made tailored specialists the default delegation guidance, so the model gives each spawn a specialist identity instead of cloning generic workers (#2468)
Changed the eval agent() helper to accept return_handle (Python) / returnHandle (JS): instead of bare text it returns a DAG node dict { text, output, handle, id, agent } whose handle is the spawned agent's recoverable agent://<id> URI, so a downstream pipeline/parallel stage can wire a large transcript by reference instead of re-inlining it. The default path is unchanged (bare text, or the parsed object under schema).
Changed isAutoresearchShCommand() to use proper command-line argument parsing instead of regex, improving accuracy for complex shell invocations
Changed autoresearch initialization prompt to display collected tradeoff metrics in the setup summary
Changed command-initialize.md template to include guidance on preflight requirements, comparability invariants, and marking measurement-critical files as off-limits
Changed command-initialize.md to instruct users to write or update autoresearch.program.md with durable heuristics and repo-specific strategy
Changed autoresearch resume guidance to emphasize continuing on the current protected branch rather than switching branches
Changed autoresearch prompt to clarify that autoresearch.md holds durable conclusions while autoresearch.ideas.md is the scratch backlog
Changed autoresearch prompt guidance to require stable measurement harness and fixed benchmark inputs unless intentionally starting a new segment
Changed autoresearch prompt to recommend keeping equal or near-equal results when they materially simplify implementation
Changed init_experiment to reset pending run state (checks, duration, ASI, artifact directory) when initializing a new segment
Changed log_experiment to set autoResumeArmed flag after successfully logging a run to enable auto-resume on next agent turn
Changed run_experiment to set autoResumeArmed flag and update dashboard after completing a run
Changed auto-resume logic to only prompt when a new pending run exists or when autoResumeArmed is explicitly set, preventing duplicate prompts
Changed path normalization in contract validation to use path.posix.normalize() for consistent path handling
Extended extension registerProvider() typing with OAuth provider support and source-aware registration metadata.
Slash commands and hook commands now work during streaming: Previously, using a slash command or hook command while the agent was streaming would crash with "Agent is already processing". Now:
- Hook commands execute immediately (they manage their own LLM interaction via pi.sendMessage())
- File-based slash commands are expanded and queued via steer/followUp
- steer() and followUp() now expand file-based slash commands and error on hook commands (hook commands cannot be queued)
- prompt() accepts new streamingBehavior option ("steer" or "followUp") to specify queueing behavior during streaming
- RPC prompt command now accepts optional streamingBehavior field (#420)
Editor component now uses word wrapping instead of character-level wrapping for better readability (#382 by @nickseelert)
Extensions can have their own package.json with dependencies (resolved via jiti)
Documentation: docs/hooks.md and docs/custom-tools.md merged into docs/extensions.md
Examples: examples/hooks/ and examples/custom-tools/ merged into examples/extensions/
README: Extensions section expanded with custom tools, commands, events, state persistence, shortcuts, flags, and UI examples
SDK: customTools option now accepts ToolDefinition[] directly (simplified from Array<{ path?, tool }>)
SDK: extensions option accepts ExtensionFactory[] for inline extensions
SDK: additionalExtensionPaths replaces both additionalHookPaths and additionalCustomToolPaths
Removed image placeholders after copy & paste, replaced with inserting image file paths directly. (#442 by @mitsuhiko)

Fixed

Fixed the advisor auto-resuming a run after the user deliberately interrupts it (Esc, or a cancel from collab/ACP/RPC/SDK/extension). A user interrupt now suppresses advisor concern/blocker auto-resume until the user next resumes (a typed message, ./c continue, or a steer/follow-up); the concern is still recorded as a visible, persisted advisor card — including one already steered into the run or arriving mid-abort — so it re-enters context on resume instead of being discarded. Natural yields are unchanged: the advisor can still steer and resume a stalled run.
Fixed /advisor on|off not being session-local by overriding the setting instead of modifying global configuration, and fixed changes not updating the TUI status line immediately.
Fixed plan.defaultOnStartup setting schema configuration missing the required ui.group property.
Fixed web_search using stale or missing provider exclusions after /move or resuming a session from another project. Provider preferences (providers.webSearchExclude, providers.webSearch, providers.image) are now reapplied when project settings reload on cwd change (#2611).
Fixed isolated task subagents with persisted transcripts resolving file tools against the parent repository instead of the isolated worktree, which made rcopy patch capture report no changes after tool-based edits.
Fixed plugin config writes against older omp-plugins.lock.json files that did not have a top-level settings object, which crashed omp plugin config set before persisting the option (#2236).
Fixed auto-retry regenerating a large write call after the provider stream timed out mid-tool-call (#2683).
Fixed soft-expired issue:// and pr:// reads to refresh live before returning stale state, with an explicit stale warning when the live refresh fails (#2684).
Fixed Claude marketplace plugin skills appearing as bare slash commands while keeping real plugin commands/ entries available (#2645).
Fixed boundary duplication warnings to always display when replacement lines match the next surviving line, even when auto-correction is disabled
Fixed secondary metrics validation to properly reject missing configured metrics and new metrics without force flag
Fixed ASI data cloning to prevent prototype pollution attacks by filtering reserved property names
Fixed CLI --api-key handling for deferred model resolution by applying runtime API key overrides after extension model selection.
Fixed extension provider registration cleanup to remove stale source-scoped custom API/OAuth providers across extension reloads.
--list-models no longer shows Google Vertex AI models without explicit authentication configured
JPEG/GIF/WebP images not displaying in terminals using Kitty graphics protocol (Kitty, Ghostty, WezTerm). The protocol requires PNG format, so non-PNG images are now converted before display.
Version check URL typo preventing update notifications from working (#423 by @skuridin)
Large images exceeding Anthropic's 5MB limit now retry with progressive quality/size reduction (#424 by @mitsuhiko)
Slash command argument substitution now processes positional arguments ($1, $2, etc.) before all-arguments ($@, $ARGUMENTS) to prevent recursive substitution when argument values contain dollar-digit patterns like $100. (#418 by @skuridin)
Edit tool diff not displaying in TUI due to race condition between async preview computation and tool execution
/model selector now opens instantly instead of waiting for OAuth token refresh. Token refresh is deferred until a model is actually used.
Shift+Space, Shift+Backspace, and Shift+Delete now work correctly in Kitty-protocol terminals (Kitty, WezTerm, etc.) instead of being silently ignored (#411 by @nathyong)
AgentSession.prompt() now throws if called while the agent is already streaming, preventing race conditions. Use steer() or followUp() to queue messages during streaming.
Ctrl+C now works like Escape in selector components, so mashing Ctrl+C will eventually close the program (#400 by @mitsuhiko)
External editor (Ctrl-G) now shows full pasted content instead of [paste #N ...] placeholders (#444 by @aliou)
Subagent example README referenced incorrect filename subagent.ts instead of index.ts (#427 by @Whamp)

[16.0.0] - 2026-06-15

Breaking Changes

Renamed the SDK tool format type and resolver from ToolCallFormat/resolveToolCallSyntax to DialectFormat/resolveDialect, and the agent option from toolCallSyntax to dialect.
Changed /dump transcript output to render messages with the selected model's native dialect turn and thinking envelopes instead of markdown role headings.

Added

Added advisor.syncBacklog setting (off, 1, 3, 5) to pause turn completion until advisor review backlog drops below the threshold, with a maximum 30-second wait
Added advisor backlog synchronization at turn end when enabled so the main session stays aligned with the advisor's pending review turns
Added automatic discovery of WATCHDOG.md files from user and project locations so advisor guidance from local watchdog instructions is appended to its system prompt
Added /advisor on, /advisor off, /advisor status, and /advisor dump [raw] slash-command subcommands to manage the advisor at runtime
Added advisor.enabled and advisor.subagents settings to enable the advisor and extend it to spawned task/eval subagents
Added advisor status badge (++ in success color) to the status line when an advisor is active
Added /dump [raw] flag to toggle between compact and legacy uncompact transcript output formats
Added /advisor on, /advisor off, and /advisor status slash-command subcommands to enable or disable the advisor at runtime and view advisor status metrics
Added a passive advisor: assign a second model to the advisor role and enable advisor.enabled to have it silently review each primary turn and inject severity-tagged advice notes via the advise tool. A nit rides the non-interrupting aside queue (batched into one card at the next step boundary), while a concern or blocker interrupts the running agent through the steering channel — aborting in-flight tools, or resuming the agent when it has already yielded — so high-severity advice is acted on immediately. Advice renders in the primary transcript as a distinct Advisor card, and the advisor gets hard-isolated read-only read/search/find access — bound to its own ToolSession so its reads never touch the primary's snapshot/seen-lines caches — to investigate the workspace before weighing in. The status line shows a ++ badge (in the success color, kept distinct from the model name) after the model name while an advisor is active, and /advisor dump copies the advisor's own transcript to the clipboard. Advisors are created only for the top-level session by default; enable advisor.subagents to extend them to spawned task/eval subagents.
Animated "thinking" pulse (·‥…‥) shown in place of a hidden thinking block while the model is actively reasoning, so streaming progress is visible even with hideThinkingBlock enabled.

Changed

Changed advisor prompting guidance to emit at most one advise per update and to prefer silence when the agent is on track
Changed /dump default output to compact markdown format; use /dump raw for the legacy uncompact format
Changed /dump and /advisor dump to default to compact transcript output and accept an optional raw flag for the legacy uncompact format
Session dump output now renders message history using the model's native dialect turn envelope instead of markdown role headings
Changed RPC, RPC-UI, and ACP hosts to default the advisor settings off instead of inheriting a user's interactive-session advisor preference.

Fixed

Fixed advisor backlog tracking so failed advice prompts do not stall catch-up indefinitely by dropping pending backlog after repeated consecutive failures
Fixed backlog accounting to decrement only after successful advisor prompts so sync waits correctly reflect advisor progress
Fixed advisor context batching to still send pending review updates when context maintenance fails instead of dropping the batch
Fixed explicit advisor enablement to clear protocol-default overrides so enabling the advisor applies immediately
Fixed advisor message card notes getting truncated to two lines when the card is collapsed.
Fixed advisor context handling to maintain its token budget by promoting or compacting/restarting advisor context while preserving advisor reasoning-off settings.
Fixed startup.quiet leaving MCP and LSP startup status events visible during launch (#2639).
Registered the Advisor group in the model settings tab so advisor settings render correctly in the settings panel.
Fixed Windows bash path handling so MSYS/Git-Bash drive aliases like /d/project and WSL-style /mnt/d/project normalize to native drive paths consistently across the bash tool cwd validation and brush filesystem builtins (#2634).

[15.13.3] - 2026-06-15

Added

Unexpected stop detection: optional tiny/smol classifier that continues the turn when the assistant says it will act but emits no tool calls.
Settings features.unexpectedStopDetection and providers.unexpectedStopModel.

Changed

Changed the job poll to return early when a steering message is queued, draining the steer immediately instead of waiting out the poll window.
Capped unexpected-stop auto-continuation to three retry attempts before giving up on repeated stops
Updated the edit tool's hashline prompt, grammar, and docs to recommend the .= inclusive range separator (SWAP 1.=3:); the legacy .. form still parses.
Normalized all internal worker argv selectors under the __omp_worker_ prefix, skipping the async worker dispatch check during normal CLI startup.

Fixed

Filtered out whitespace-only and dot-only (. or …) assistant blocks so they are treated as empty and no longer appear as visible content in message rendering, streaming reveal counts, or session export output
Filtered placeholder-only thinking content from ACP notifications and message visibility checks so dot-only reasoning_content no longer triggers turn completion or read/run updates
Fixed ModelRegistry tests making outbound network calls by automatically stubbing fetch during test execution.
Fixed eval JS cells (and browser-tab worker startup) always stalling for the full init timeout — typically the cell's whole 30s budget — before silently falling back to the slower inline worker. The self-dispatching CLI host imports the worker module dynamically from its argv dispatch, so the worker's own parentPort.on("message") attached only after Bun flushed the messages the parent posted before spawn; the synchronously-posted init handshake was dropped and never answered with ready. The host now installs a buffering parentPort inbox synchronously in the entry's sync prefix (before importing the worker module) and the worker binds it on load, replaying the buffered handshake. omp --smoke-test now also spawns the JS eval worker through the host entry and asserts it handshakes on a real worker thread.
Fixed pre-prompt context-full compaction on OpenAI Responses sessions to use provider-anchored context usage when available, so large encrypted reasoning signatures no longer trigger automatic maintenance while the visible context percentage remains below threshold (#2628).

[15.13.2] - 2026-06-15

Added

Added tool examples data to exported RPC session tool metadata, so tool dumps and other clients can receive model-call examples
Added supportsTools to model definitions and overrides so custom model configs can declare whether a model supports native tool calls
Added tools.format for choosing native tool calling or a specific owned in-band format (glm, hermes, kimi, xml), with auto falling back to GLM only for models marked as not supporting native tools.
Added a conditional easter-egg tip recommending nerd fonts when using the unicode symbol preset.
Added the tools.abortOnFabricatedResult setting (default on): with in-band tool calls, stop the model the moment it starts hallucinating a tool result mid-turn, or disable it to let the model finish and discard the fabricated continuation instead.

Changed

Changed /dump tool catalog output to render tools through the shared inventory renderer with readable TypeScript-style signatures and native-syntax <examples> blocks
Changed todo tool result rendering so that collapsed phases truncate from the beginning, showing the latest/active tasks and displaying the "more todos" summary at the top.
Expanded tools.format to support additional in-band tool-call syntaxes, including anthropic, deepseek, harmony, pi, and qwen3
Changed the 13 tools that documented hand-written <examples> blocks (eval, browser, todo, irc, ssh, ast_edit, ast_grep, debug, find, inspect_image, ask, plus the patch/apply_patch edit modes) to define examples as typed examples data on the tool (ToolExample<z.input<typeof schema>>); the AI layer now renders them in the model's native tool-call syntax and the markdown <examples> blocks were removed.
Changed the experimental owned tool-calling prompt from a GLM-only toggle to syntax-specific grammar prompts and result formats. PI_OWNED_TOOLS=1 still forces GLM; PI_OWNED_TOOLS=<syntax> forces that syntax.
Changed the large-paste menu to offer attachment XML blocks (<attachment>), local-file attachments, or inline paste as explicit actions.
Changed the system-prompt tool inventory: moved the # Inventory block to the bottom of the TOOLS section, and it now renders a compact tool-name list only when native tool calling is active and tool descriptions are not repeated; otherwise it emits full # Tool: <name> sections.

Fixed

Fixed Auto-Promote Context being pre-empted by compaction: the pre-prompt context check ran compaction directly, so snapcompact (or any strategy) fired before promotion ever got a chance. It now tries promotion to a larger-context model first — mirroring the post-turn threshold path — and only compacts when no larger-context target is available. Snapcompact (auto and manual) also falls back to a context-full LLM summary when its frame archive plus kept history would still overflow the model's usable window, instead of leaving the session over the limit.
Fixed eval JS cells intermittently failing after ~15s with exit code 1 under load (e.g. bun test --parallel, where each file runs in a worker subprocess and the eval worker is nested). The host swallowed asynchronous worker spawn/load/crash failures — new Worker reports module-load errors via an async error event, not a synchronous throw, so the spawn try/catch (and its inline-worker fallback) never fired, and no error/messageerror listener was wired — leaving a dead worker indistinguishable from a slow one and blocking the full worker-init timeout. The init handshake now rejects immediately on a worker error/messageerror event and falls back to the inline worker, and the ready listener is attached synchronously after new Worker (Bun does not buffer messages posted before a listener exists) so a fast worker's ready can no longer be dropped.

[15.13.1] - 2026-06-15

Added

Added isolated profile support via --profile <name> / OMP_PROFILE and shell alias bootstrap via --alias <command>, including launch/ACP bootstrap handling, extension-flag-safe parsing, profile-scoped user config discovery, and symlinked extension-directory discovery.
Fixed paste and image placeholders crashing when the editor renders before theme initialization.

Changed

Replaced the omp bench default prompt with a concrete query-planning trace that requires deriving selectivities, cardinalities, and I/O/CPU costs from given schema and data. The old prompt was open-ended prose recall, which rewarded not-thinking: adaptive-thinking models (Opus 4.6+/Sonnet 4.6+) minimized reasoning on the trivial task and streamed faster, skewing throughput comparisons. The new task forces multi-step reasoning so adaptive thinking engages and the benchmark measures generation under real cognitive load. The prompt also demands explicit upfront deliberation and exhaustive enumeration/costing of every join order, so adaptive-thinking models cannot short-circuit to a quick answer and each model is measured over a sustained generation up to the token cap.

Fixed

Fixed hashline edits from read, search, and ast-grep so replacements are rejected when they target lines not shown in the tool output
Fixed /tan refusing to launch while the main response was still streaming. The command exists to fork tangential work alongside an active session, so it now dispatches mid-stream and queues its handoff breadcrumb for the next turn instead of steering the in-flight one.
Fixed /tan background agents never staying in the Agent Hub. The forked clone now uses an <agentId>.jsonl session file (so the persisted-subagent scan keys it by the same id the live ref uses) and is parked rather than unregistered on completion, so it stays listed and its transcript stays readable.
Fixed the /tan dispatch breadcrumb rendering its full raw <system-notice> block in the transcript. It now shows a single compact line (Tangent dispatched [task] <jobId> — <work>), styled as a sibling of the "Background job completed" line.
Fixed profile bootstrap parsing so built-in string flags like --plan no longer consume the profile-boundary marker and drop the trailing user message
Fixed a bug where goal mode was incorrectly deactivated/set to 'none' on every wall-clock-only update (when tokenDelta <= 0) during tool execution flushes, preventing OMP from writing a mode change to 'none' in the session history database while keeping in-memory/UI state fresh.
Fixed a bug where Gemini MALFORMED_FUNCTION_CALL tool-generation errors (which are transient) surfaced as terminal error blocks. Added "malformed function call" to the transient transport error classifier so the session automatically retries the turn.
Fixed Ctrl+C teardown waiting up to 30s when an extension's session_shutdown handler hung — observed on Windows with omp-discord-presence 0.1.2 stuck on a stale Discord IPC pipe. ExtensionRunner.emit previously shared the generic 30s EXTENSION_HANDLER_TIMEOUT_MS budget for every event, including the fire-and-forget teardown event extensions cannot observe. session_shutdown now uses a dedicated 2s SESSION_SHUTDOWN_HANDLER_TIMEOUT_MS cap routed through a per-event handlerTimeoutForEvent() lookup, so a hung third-party handler can no longer hold dispose hostage. As a defence-in-depth ladder, a Ctrl+C arriving while interactive shutdown is already running now hard-exits with code 130 (the session JSONL has already been sync-flushed by the first press), surfaced via a new read-only InteractiveModeContext.isShuttingDown (#2600).
Fixed the external editor flow (Ctrl+G, plan editor, /todo edit) warning No editor configured on Windows even when the user expected the system's native editor. getEditorCommand() now falls back to notepad on win32 after consulting $VISUAL/$EDITOR (and trims those values so accidental whitespace is ignored), so Windows users get a working editor out of the box while POSIX still warns to nudge configuration (#2604).
Fixed Kokoro TTS setup loading the workspace/global @huggingface/transformers runtime before the side-installed Kokoro runtime, which could leave onnxruntime-node@1.26.0 bound to an older libonnxruntime.so.1 and fail with VERS_1.26.0 missing (#2591).
Fixed scripts/ci-release-notes.ts stranding curated changelog entries from intervening silent tags (a vX.Y.Z tag pushed without a GitHub Release, e.g. the v15.12.5/v15.12.6 casualties of the pre-#2564 release-cancellation bug). The generator now walks (latest-published-release, target] — resolved via gh release list from the release_github CI job — and merges every in-range ## [X.Y.Z] section per package, grouped by ### <category> with bullet-level dedupe so post-release changelog flattening cannot duplicate entries. Falls back to the legacy single-version extraction when no prior published release resolves, and OMP_RELEASE_NOTES_FLOOR=v15.12.4 overrides the lookup for manual rebuilds (#2596).
Fixed Test & smoke (TS) CI timeouts caused by parallel test files racing on the process-global Settings singleton. CustomEditor now accepts a magicKeywordsEnabledOverride injection point so the shimmer-gate test can assert behaviour without calling resetSettingsForTest() / Settings.init(); the "streaming tool call preview height" describe drops its gratuitous Settings reset+init. Production wiring is unchanged (#2582)
Fixed a collapsed, still-streaming tool preview (an eval/bash/ssh box with output streaming in) reading as "weirdly truncated" — top border and head rows missing — once its box outgrew the viewport, snapping back to whole only while expanded with ctrl+o and breaking again when collapsed. A streaming preview was classified commit-unstable whenever collapsed, so the transcript offered none of its rows to native scrollback; once the box outgrew the window its head fell into the gap between the commit boundary and the window top, committed nowhere and repainted nowhere. The provisionalPendingPreview flag now applies only to the pending call preview (before any result) — once a streaming result exists the result renderer is the live, top-anchored shape and the block is commit-stable in both collapsed and expanded states, so its durable head always reaches scrollback.
Fixed a crash in subagent task execution and extensions when a string (instead of a string array) was returned or set for the system prompt. Gracefully wrap string values in arrays.
Fixed the todo completion reminder escalating 1/3 → 2/3 → 3/3 within a single user pause: #checkTodoCompletion appended a <system-reminder> and then scheduled agent.continue(), so a text-only acknowledgement ("paused at your instruction") triggered another agent_end that re-ran the same check and fired the next reminder — no user input required. AgentSession now tracks #todoReminderAwaitingProgress: a reminder sets it, any toolResult (real progress) or a new user prompt clears it, and #checkTodoCompletion stays silent while it is set. Escalation through todo.reminders.max still works when the agent makes tool-level progress between stops (#2590).
Fixed profile bootstrap so an extension-shadowed --plan flag no longer swallows a following global --profile.
Fixed MCP OAuth URL-keyed credentials to stay profile-scoped under shared auth-broker storage and to clear discovered definition-only server auth during /mcp unauth.
Fixed /mcp unauth deleting another profile's MCP OAuth credential row under broker-backed auth storage: when a shared mcp.json pins an explicit auth.credentialId scoped to a different profile (mcp_oauth:profile:<other>:<url>), removal now skips ids scoped to a non-active profile, mirroring the read path that already refuses to use a foreign profile's id. Legacy url-keyed (mcp_oauth:<url>) and active-profile ids are still cleared.
Fixed auto-learn managed skills to use the active profile's agent directory, so authored profile skills keep priority over managed fallbacks.

[15.13.0] - 2026-06-14

Breaking Changes

Replaced the omp setup stt command with omp setup speech. The old stt setup component is gone (no alias); omp setup speech now provisions the full speech stack — audio recorder, speech-to-text model, and text-to-speech model.
Renamed the tts.enabled setting to speechgen.enabled (same boolean, default off; no alias). It still gates the on-demand tts speech-generation tool, now labelled "Speech Generation" in the settings panel.

Added

Added paste.largeMenuThreshold setting (0/100/250/500/1000, default 100) to control when large pasted content triggers the large-paste menu or stays as a normal [Paste] marker
Added a large-paste editor menu for pasted text over the threshold that lets users choose to wrap the paste in a fenced code block, wrap it in <pasted_text> XML tags, or save it as local://attachment-N for on-demand reading
Added snapcompact-savings.jsonl journaling for snapcompact tool-result compaction, recording session, provider, model, tool call, and estimated token savings whenever tool output is rendered as image frames
Added subagent:<id> loop-phase breadcrumbs around in-process subagent event dispatch and finalization so the TUI event-loop watchdog can attribute a main-thread stall to subagent execution (#2485)
highlightMagicKeywords(text, resetTo?, phase?) now accepts an optional phase ∈ [0, 1) that rotates the gradient cyclically; sent bubbles omit it (static palette unchanged). hasMagicKeyword(text) exported from modes/magic-keywords is the cheap shimmer-gate the editor uses on every render.
Added a fastModeScope setting (both | openai | claude, default both) controlling which providers /fast on (and the fast-mode toggle) target. both keeps the prior unscoped priority behavior; openai/claude scope fast mode to one family. /fast status now reports the active scope.
Added the mnemopi.embeddingVariant setting (en | multilingual) selecting a stronger SOTA local embedding model — en → BAAI/bge-base-en-v1.5 (768d), multilingual → intfloat/multilingual-e5-large (1024d). Resolution precedence is mnemopi.embeddingModel setting > MNEMOPI_EMBEDDING_MODEL env > variant default, so the documented env override is still honored. Changing the active model wipes and rebuilds stored embeddings on the next writable start (#2476)
Added a /guided-goal slash command that interviews you to refine an objective before enabling goal mode, then seeds goal mode with the agreed objective. The bounded interview (up to six turns) runs on the plan or slow model and falls back with a hint when the goal is still too vague (#2502).
Added a large-paste menu: when a paste reaches paste.largeMenuThreshold lines (default 100; 0 disables), the editor offers to wrap it in a code block, wrap it in <pasted_text> XML tags (both collapse to a [Paste] marker that expands on submit), or save it to the session's local:// store and insert a clean local://attachment-N reference the agent can read on demand. Esc keeps the previous inline-paste behavior, so the content is never lost.
Added 8on22-bw (leading) and 11on16-bw (tracking) options to the snapcompact.shape setting, the spacing-tuned cells that are now the per-provider defaults (Anthropic → tracking, OpenAI/Google → leading)
Added a local on-device neural TTS backend for the tts tool and a providers.tts switch (auto | local | xai, default auto). local synthesizes speech with Kokoro-82M — SoTA on-device TTS quality — via kokoro-js on the shared ONNX runtime (@huggingface/transformers + onnxruntime-node) in a subprocess worker (mirroring the tiny-model worker), keeping the model warm across calls and emitting 24 kHz WAV/PCM16 with no network call; xai keeps the existing Grok Voice cloud path; auto prefers local but routes .mp3 requests to xAI when credentials exist (no local MP3 encoder is bundled, so a local .mp3 request is written as a sibling .wav). kokoro-js is never a hard dependency: it is lazily bun installed into a version-keyed runtime dir on first use (with onnxruntime-node force-pinned to a Bun-safe version), so its transformers@3.x graph never pollutes the main tree. New tts.localModel (default kokoro) and tts.localVoice (default af_heart; American/British, female/male voices) settings select the on-device voice.
Added a unified, interactive omp setup speech command that walks one reusable flow across all three speech dependencies: it lets you pick and persist the speech-to-text (stt.modelName) and text-to-speech (tts.localModel) models from a TUI list, then downloads both models plus an audio recorder with live progress. The recorder is now auto-provisioned cross-platform (a static ffmpeg binary is fetched via the shared tools-manager when no SoX/FFmpeg/arecord is present, with the PowerShell fallback on Windows) instead of dead-ending with "install sox manually". --check and --json report recorder + STT-model + TTS-model readiness without installing.
Added omp say <text>, which synthesizes text with the local on-device TTS engine and plays it through the speakers (cross-platform: afplay on macOS, paplay/aplay/bundled ffmpeg on Linux, PowerShell Media.SoundPlayer on Windows). --out <file> writes a WAV instead of playing, --voice/--model override the tts.localVoice/tts.localModel settings, and an uninstalled model prints an actionable omp setup speech hint.
Added streaming speech vocalization: with speech.enabled on, the assistant speaks its reply through the speakers as it streams. Assistant text deltas are fed directly into the engine's incremental text input (Kokoro's TextSplitterStream via the worker) as they arrive, rather than pre-chunked in JS and synthesized one batch call per sentence — the engine owns sentence segmentation and emits one audio chunk per sentence. A single persistent player (StreamingAudioPlayer) drains those chunks gaplessly (raw 32-bit-float PCM piped to one ffmpeg→PulseAudio/ALSA process on Linux; interruptible per-file afplay/PowerShell SoundPlayer on macOS/Windows), replacing the spawn-a-player-per-sentence path that added latency and audible gaps. Overspeech is handled end to end: a new turn, a sent message, or an Esc/Ctrl+C interrupt stops playback instantly (the player process is killed rather than letting the current sentence finish); holding the push-to-talk key ducks the volume while you speak and restores it when you stop; and sequential utterances queue and drain in order instead of overlapping. speech.mode (all | assistant | yield, default assistant) picks what is spoken — all adds thinking, yield speaks only the final message at turn end — and speech.voice selects the Kokoro voice. ask-tool questions are spoken in every mode. Synthesis reuses the local Kokoro engine (tts.localModel) through a new streaming synthesis path (TtsClient.synthesizeStream) that pushes text in and streams audio chunks back over the worker protocol.
Added live (streaming) speech-to-text: with stt.enabled on, transcription now appears in the composer as you speak instead of all at once after you stop. The recorder streams raw 16 kHz mono PCM from sox/ffmpeg/arecord stdout to the warm STT worker, where an energy-based endpointer (no extra model) splits speech into segments at natural pauses; each finalized segment is committed into the editor while the in-progress segment shows a live volatile preview that refreshes in place and is kept out of the undo history. Works with both the default Parakeet (sherpa-onnx) and the Whisper (transformers.js) tiers. Recorders that cannot stream to a pipe (the Windows PowerShell mci fallback) transparently fall back to single-shot transcription.
Added skills.enableAgentsUser and skills.enableAgentsProject settings (default on) so the canonical OMP-native ~/.agent[s]/skills and project-walkup .agent[s]/skills are configurable independently from the third-party Claude/Codex/Pi toggles.
Added a read-only ctx.models facade for extensions: list() (authenticated models), current() (live session model), resolve(spec) (a model string or role alias → Model, using the same settings-backed aliases and match preferences as core selection), and family(model) (opaque canonical-identity lineage token for cross-family comparisons). Lets extension tools select models the way core does without reaching into the mutable registry (#2406)
Added RPC prompt lifecycle hints so hosts can distinguish scheduled agent turns from local-only slash commands via data.agentInvoked and prompt_result.
Added extension lifecycle events for tool approval prompts: tool_approval_requested before the approval wait and tool_approval_resolved after approve, deny, or approval prompt failure.

Changed

Changed handoff custom messages (customType: "handoff") to render in the transcript as a compaction-style expandable divider in both the main session and Agent Hub views, and expanded handoff details now show the handoff context body without <handoff-context> tags
Changed the double-tap-← gesture (empty editor, main session) to stay inert when there are no subagents to show, instead of opening an empty Agent Hub roster. The explicit Agent Hub / observe keybindings still open the empty roster. The gating reuses the hub's own row count (after its persisted-subagent scan), so it matches exactly what the hub would display.
Changed the job tool's async.pollWaitDuration setting (relabeled Max Poll Time) to add a smart value, now the default. A fixed value (5s–5m) still blocks for exactly that long; smart adapts: a blocking poll starts at a 5s floor and climbs a ladder (5s → 10s → 30s → 1m → 5m) with each back-to-back poll, so a tight poll loop backs off and stops spending turns on "still running" frames, then resets to the 5s floor after ~1 minute without polling (i.e. when the agent steps away to do real work). Escalation is tracked per agent (owner-scoped on AsyncJobManager).
Added the compat.supportsForcedToolChoice custom-model flag for OpenAI-compatible models whose endpoints accept tools but reject forced tool_choice values (#2546).
Changed speech-to-text to run fully local on-device with a tiered, multi-engine model picker. Transcription runs in a subprocess worker (mirroring the tiny-model worker; the native ONNX addons are hard-killed on shutdown to dodge the Bun NAPI-finalizer segfault) instead of shelling out to Python openai-whisper, keeps the model warm across recordings, and decodes WAV to 16 kHz mono float32 in-process. stt.modelName now selects on-device tiers across two engines: parakeet (default) — NVIDIA Parakeet TDT 0.6B v3 (25 languages) via the native sherpa-onnx-node, the Open ASR Leaderboard accuracy + throughput leader (lower WER than, and ~20× faster decoding than, Whisper large-v3) — plus fast/balanced/turbo mapping to Whisper base/small/large-v3-turbo (multilingual, up to 99 languages) via @huggingface/transformers. omp setup speech no longer mentions pip/python-whisper and reports recorder + model-cache readiness.
Changed the speech-to-text trigger from the Alt+H keybinding to a hold-Space push-to-talk gesture. Holding the space bar emits an OS auto-repeat burst; once more than 5 spaces land in the editor it recognizes the hold, deletes (tracks back) those inserted spaces, and starts recording, then stops and transcribes when the repeats stop (the space bar is released). app.stt.toggle is now unbound by default but can be rebound to a chord for press-to-toggle; the gesture is gated on stt.enabled, and Shift+Space still inserts a literal space.
task.eager ("Prefer Task Delegation") and todo.eager ("Create Todos Automatically") are now three-level enums (default / preferred / always) instead of booleans. For todo.eager, preferred renders a soft first-message reminder while always forces the todo tool (the previous "on" behavior); for task.eager, preferred adds a soft (SHOULD) delegation nudge to the system prompt while always uses hard (MUST/ONLY) wording plus a first-turn delegation reminder. Existing boolean configs migrate automatically (true → always, false → default). On models that cannot be forced to call todo, todo.eager: "always" now emits the first-turn reminder without forcing the call (previously such models received nothing) (#2539, #2540 by @metaphorics).
Fixed /model-switching to a non-default OpenRouter model returning 404 No route: POST /chat/completions when the provider is routed through the auth-gateway broker. The background catalog refresh re-ran mergeDiscoveredModel on every openrouter entry; for models whose bundled record already existed, the merge re-applied baseUrl, headers, and compat but dropped transport: pi-native because the raw /v1/models payload carries no transport hint. The next /model switch then picked the now-transport-less entry and routed through the default openai-completions client to ${baseUrl}/chat/completions — a path the auth-gateway never serves. mergeDiscoveredModel now propagates the override/existing transport on the rediscovery branch (#2555).

Fixed

Fixed npm plugin installs to reject packages whose declared extension entry points cannot load because imports or nested dependencies are unresolved (#2312).
Fixed the deferred MCP discovery banner (Connecting to MCP servers: …) overdrawing the chat input bar. onMCPConnecting wrote the banner straight to process.stderr while the TUI owned the terminal; it now emits on an mcp:connecting event channel that InteractiveMode renders through showStatus (the status container), mirroring the existing LSP-startup pattern so the banner can never paint over the input box border (#2483)
Fixed Win+Shift+S screenshot paste on Windows dead-ending on Image not found: Windows Terminal forwards a bracketed paste of the Snipping Tool's transient …\MicrosoftWindows.Client.Core_*\TempState\… file path, which is already gone (or never materialized) by the time omp reads it, so handleImagePathPaste failed with ENOENT even though the screenshot bitmap was still on the clipboard. The handler now falls back to the clipboard image across every local read failure (missing file, undecodable file, or generic error) before degrading, and the fallback is skipped over SSH where the clipboard lives on the remote host rather than the terminal holding the screenshot.
Fixed npm prebuilt extension compatibility shims deriving their own package root through bare @oh-my-pi/pi-coding-agent resolution, which could select an older Bun cache copy in global installs and reintroduce mixed-runtime plugin loading stack overflows.
Honor the context_length reported by OpenAI-compatible /v1/models discovery (discovery: { type: "proxy" } and discovery: { type: "openai-models-list" }) when present, so aggregator-reported windows override the stale bundled reference; the value is validated through the positive-number guard so a 0/negative/stringly-typed upstream value cleanly falls back to the bundled reference (then 128000) instead of pinning a broken window (#2466 by @androw).
Fixed web_search SearXNG fallback when HTTP 200 responses contain no usable results plus unresponsive_engines; SearXNG now raises a transient provider error, and the fallback loop rejects any provider response with no renderable content before formatting an invisible success (#2571).
Fixed read on a GitHub commit URL (github.com/<owner>/<repo>/commit/<sha>) returning the raw commit HTML page instead of structured content. parseGitHubUrl had no commit case, so commit URLs fell through to generic HTML rendering; they now resolve via the commits API and render as markdown (subject, author, stats, parents, full commit message, and a per-file unified diff), matching the existing blob/tree/issue/PR handling.
Fixed release runs being silently cancelled by a later main push, which left tagged versions (v15.12.6 in the wild) without a GitHub Release or npm publish. The CI workflow's concurrency group was ${{ github.workflow }}-${{ github.ref }}, and since the release-script commit + v* tag are pushed atomically to refs/heads/main, the release run shared the CI-refs/heads/main group with every subsequent push; cancel-in-progress: true then killed it before release_binary / release_github / release_npm could run, and no future run carried the release tag at HEAD. The group now resolves to a per-sha release-<sha> slot with cancel-in-progress: false whenever the push subject matches chore: bump version to (the release-script convention) or github.ref is a v* tag (workflow_dispatch recovery), so release runs are isolated from PR/main churn (#2564).
Allowed compat.streamIdleTimeoutMs: 0 in models.yml. The schema was .positive(), so the documented "set to 0 to disable" escape hatch was only reachable via the global env var (#2422)
Fixed LaTeX math delimiters ($/$$) and commands (such as \text, \times) rendering raw in the terminal by instructing the model to write equations as plain text / Unicode in its replies. The instruction is scoped to conversational output, so it does not constrain LaTeX or Markdown/KaTeX content the agent is asked to write to files (#2550 by @usr-bin-roygbiv).
Fixed the MCP stdio transport close() potentially hanging while awaiting a read loop that can block indefinitely; teardown now detaches the read loop instead of awaiting it (#2550 by @usr-bin-roygbiv).
Fixed per-project memory isolation pulling other projects' memories into recall. Legacy (pre-#2412) mnemopi banks were rescued into recall when any working-memory row tagged the active cwd, but recall reads a bank wholesale and cannot filter rows by cwd, so a mixed-cwd bank leaked unrelated projects' memories. A legacy bank is now rescued only when every row tags the active cwd (#2412).
Fixed a prompt template whose name collides with a builtin slash-command alias (e.g. a models template beside the builtin /model, which owns the models alias) appearing as a duplicate entry in the slash-command autocomplete picker. The picker's reserved-name filter now also excludes command aliases, matching runtime resolution (slash commands expand before prompt templates, so the template was already unreachable).
Fixed the tool-result renderer re-shaping on every invalidate() (spinner tick, stream chunk, resize, keystroke), which made large grep/find/read results block the main thread for seconds and made typing sluggish. ToolExecutionComponent.#updateDisplay() now memoizes on a dirty key (result version, expand state, partial flag, spinner frame, image visibility, theme epoch, background-task freeze state, the resolved terminal image protocol, and a display-input version that covers streamed call args, the async edit-diff preview, and Kitty image conversions) and a #displayBuilt guard that also fast-paths the #contentText fallback, so the O(result-size) shaping runs once per change instead of every frame without freezing streamed args, previews, converted images, a backgrounded task settling to its static form, or images that arrive before the async image-protocol probe resolves. Image-bearing results also re-shape on terminal resize (keyed on the resolved image dimensions only when images are present) so inline images rescale, while image-free results never re-shape on resize (#2484)
Fixed setTheme() not bumping the theme epoch on its invalid-theme fallback path: a failed theme load swaps the active theme to the dark fallback, so memoized renderers (the tool-result renderer above) must re-shape — previously they kept the failed theme's stale colors until some other state changed (#2484)
Fixed tool-call spinners animating out of phase across parallel tool calls — each live tool block advanced its glyph from its own per-instance start time, so concurrent spinners showed different frames. Glyphs now derive from a single shared monotonic clock (sharedSpinnerFrame), keeping every live block in lockstep.
Fixed the per-turn token-usage row (display.showTokenUsage) churning and duplicating in scrollback, most visibly with parallel tool calls. The row was rendered inside the assistant block above the turn's tool blocks, so finalizing the block was deferred and late appends recommitted the already-committed tool rows. The assistant block now always finalizes as soon as a tool-call appears, and the usage row is emitted as a standalone finalized block below the turn's tool blocks across all three render paths (live, transcript rebuild, agent-hub).
Fixed the editor input box claiming a disproportionate share of small terminals (<=18 rows): the editor max-height floor (6 rows) ignored the available space. The height now yields to terminal size (EDITOR_MIN_CHROME_ROWS), reserving rows for the transcript and status line whenever the terminal can host both; on terminals too small for both, the editor collapses to its real bordered minimum instead of overshooting a fictitious cap.
Fixed ultrathink / orchestrate / workflowz keywords not glowing while typing — the editor appended a CURSOR_MARKER (ESC-prefixed) after each render, so the magic-keyword regex's right-boundary (?!\S) tripped on ESC and dropped the gradient until a trailing character was typed. The marker-aware editor decorate (packages/tui) plus a phase-aware highlightMagicKeywords overload restore the live glow and add a Claude-Code-style shimmer while the prompt is focused, gated on magicKeywords.enabled (#2475).
Fixed the compaction flow (/compact and plan-mode "Approve and compact context") leaving UI artifacts. executeCompaction added a Spacer(1) to the transcript that the sibling handoff path never adds and that leaked as an orphan blank line whenever compaction was cancelled or failed; that spacer is removed. On success the compaction loader is now stopped and the status container cleared before the transcript is rebuilt, so the live loader row no longer flickers over the reconciled transcript near the status seam (the idempotent finally still covers the cancel/fail paths) (#2486)
Fixed plan approval's "Approve and compact context" running the compaction summarizer on the pre-plan model instead of the plan model, cold-missing the plan model's prompt cache. Compaction now runs on the plan model (warm cache); the switch to the execution/pre-plan model happens only after a successful compaction and before any input queued during compaction is dispatched, so the queued turn runs on the post-compaction model. A cancelled compaction now also restores the pre-plan model (it previously stranded the session on the plan model), while a failed compaction stays on the plan model with its context intact.
Fixed Alt+Up (dequeue) reporting "No queued messages to restore" for messages — including skills — typed while the session was compacting. restoreQueuedMessagesToEditor now drains compactionQueuedMessages alongside the agent queue, so the Alt+Up to edit hint restores every pending message it advertises.
Fixed restoreQueuedMessagesToEditor (Alt+Up dequeue and Esc-abort) producing colliding [Image #N] markers when the editor draft already held pending image(s): queued text was prepended but queued images were appended, so positional marker → image lookup at submit time resolved to the wrong image. Each queued message's image markers are now renumbered by the running pending-image count before merge so the combined text stays aligned with the merged pendingImages order (#2531).
Fixed AgentBusyError ("Agent is already processing. Use steer() or followUp()...") surfacing on mode transitions — as Failed to finalize approved plan: ... when a plan was approved while the agent was still streaming the post-resolve continuation (or a turn started by the approve-time compaction/clear), and as an error toast when a loop auto-submit or goal continuation fired during a streaming/compaction race. Plan approval now aborts any in-flight turn before dispatching the executor's first prompt, and submitInteractiveInput routes streaming-time loop, goal-continuation, and manual submissions through the follow-up queue (streamingBehavior: "followUp") instead of throwing (synthetic continue-shortcuts stay developer-attributed and keep their prior behavior). Extends the manual-/goal fix in #2454 to the continuation and plan-approval paths.
Fixed /plan cycling between plan and plan_paused with no path back to mode none. handlePlanModeCommand had branches for entering and pausing but fell through to #enterPlanMode() when invoked from the paused state, so once a session entered plan mode the only operator-visible toggle re-entered it. The handler now matches planModePaused and fully exits — clearing planModeHasEntered and appending a mode_change to "none" — so /goal (and any other mode gated on planModeEnabled || planModePaused) can run again after a third /plan (#2510).
Fixed HTML session export rendering empty text tokens (text, userMessageText, customMessageText, toolTitle) as the dark-theme grey #e5e5e7 on every theme not literally named light, making transcripts illegible on custom light themes like sandstone, limestone, and porcelain. getResolvedThemeColors and the standalone isLightTheme helper now classify against the resolved statusLineBg luminance (the same surface Theme.isLight uses), so the HTML defaultText falls back to #000000 on light themes and the standalone helper stays in lockstep with Theme.isLight (#2516).
Fixed the Agent Hub opening on its own from a stray mouse click. The double-tap-← gesture (empty editor) fired on any two left keys within 500ms, but terminals with "click to move cursor" / pointer features (iTerm2 option-click, WezTerm, kitty, tmux) synthesize a burst of arrow keys on click — delivered sub-millisecond apart in one stdin read — so a single click could pop the hub with no key ever pressed. The gesture now requires the second tap to land a human-plausible interval after the first (≥40ms, <500ms) and ignores any third-or-later rapid tap, so synthesized bursts are rejected while a deliberate double-tap still works. The same hardening applies to the focused-subagent ←← "return to main" gesture.
Fixed the ctrl+p model-role cycle indicator (the default / gpt / fable / … chip track) stacking duplicate copies in the scrollback when other chat activity landed between two cycles. The track was emitted through showStatus, whose back-to-back coalescing only merges when the previous status is still the last transcript child; any interleaved append broke that identity check and appended a second track. It now renders into a dedicated anchored container above the editor (cleared and rebuilt in place each cycle, like the Todos HUD) and auto-clears after a short linger, so rapid presses or concurrent activity can never duplicate it.
Fixed the Working… loader vanishing for the rest of a turn after an auto-compaction (context-overflow recovery) or auto-retry. Those overlays took over the shared status container with a bare statusContainer.clear(), which detached the working loader but left loadingAnimation set; the resumed turn's agent_start → ensureLoadingAnimation() is guarded by if (!this.loadingAnimation), so it skipped re-attaching the loader and the spinner stayed gone while the agent kept streaming. The overlay handlers now fully tear the working loader down (stop + dereference) via #stopWorkingLoader(), so the next agent_start recreates and re-attaches it.
Fixed JS eval helper optional arguments rejecting Python-style positional calls. read(path, offset, limit) now works alongside read(path, { offset, limit }), null/undefined skip optional positional slots, and non-local URI reads such as artifact://... delegate through the read tool so line slicing works on spilled artifacts.
Fixed legacy extension compatibility remapping for @*-pi-ai/utils/oauth imports so background workers load the relocated @oh-my-pi/pi-ai/oauth exports instead of resolving missing src/utils/oauth/* files (#2566).
Fixed eager todo initialization prompting GPT-5.5 to emit unsupported task metadata fields, which could leave fresh sessions stuck on the forced first todo call (#2561).
Fixed Windows plan-mode task fan-out crashing the TUI when nested async task progress formed a cycle; task rendering now cuts recursive snapshots and long Windows local:// roots are shortened under temp storage (#2551).
Fixed a band of streaming assistant output being lost from native scrollback — committed nowhere, repainted nowhere — once a reply grew taller than the viewport. Markdown whose layout keeps changing above the streaming tail (most visibly a table whose columns re-align as rows arrive) never earns a byte-stable commit-safe end, so as its head scrolled above the window the rows fell into the gap between the commit boundary and the window top and vanished. TranscriptContainer now reports a getNativeScrollbackSnapshotSafeEnd() for commit-stable live blocks (their whole body is durable content), and the renderer commits those scrolled-off rows audit-exempt — a later layout change of an already-committed row freezes a slightly-stale row in scrollback (duplication never loss) instead of dropping it. Provisional blocks (collapsing tool/edit previews) are unaffected.
Fixed unknown ---prefixed flags being silently consumed as prompt text, which let a stale or typoed flag start a real agent session (connecting to MCP servers, waiting on the model) instead of failing fast. parseArgs now tracks unrecognized flag-shaped tokens and runRootCommand calls reportUnrecognizedFlags immediately after the post-extension reparse, exiting 2 with Error: unknown flag: --… before any session, MCP, or initial-message work runs. Extension-registered flags still pass cleanly since the validation runs after the extension-aware reparse, and -- is honored as a POSIX positional separator so flag-shaped prompts (omp -p -- --explain-this) survive the new guard (#2459).
Fixed /goal <objective> and /goal set <objective> during streaming so goal context is steered immediately but objective submission waits for the active turn to finish instead of spamming AgentBusyError. The interactive goal-continuation timer is now streaming-aware too: if a turn starts inside the 800 ms idle window the timer was scheduled in, it drops the tick instead of submitting a stale goal-continuation that would resurface the same AgentBusyError; the next agent_end reschedules (#2454).
Fixed ~/.agent[s]/skills not appearing as /skill:<name> commands when every named source toggle (skills.enableCodexUser, skills.enableClaudeUser, skills.enableClaudeProject, skills.enablePiUser, skills.enablePiProject) was off: loadSkills gated the agents provider on anyBuiltInSkillSourceEnabled, so a user who turned off the Claude/Codex/Pi sources to clean noise also lost their own canonical OMP-native skills. The agents provider now reads the dedicated enableAgentsUser/enableAgentsProject toggles, and the unknown-third-party fall-through gate is restricted to the named third-party toggles so keeping the default agents toggles on no longer silently re-enables opencode/github/claude-plugins/gemini skill sources (#2401).
Fixed Claude Code marketplace plugin skills installed under skills/<name>/SKILL.md to also appear as bare slash commands such as /understand, matching Claude-native plugin docs. The slash command name is taken from the skill directory basename so display-style frontmatter names like name: Understand Anything still resolve to /understand (#2415).
Fixed ACP /move builtin test expectations to compare the resolved destination path so the test is portable on Windows and Unix (#2381 by @oldschoola).
Fixed vLLM discovery so providers.vllm.baseUrl drives the built-in endpoint, additional OpenAI-compatible vLLM provider IDs work through openai-models-list, and discovered max_model_len or fallback context_length values set context windows instead of falling back to 128k.
Fixed extension discovery ignoring package directories symlinked into an extensions/ directory.

Removed

Removed the Python openai-whisper dependency and pip install path from speech-to-text — the bundled transcribe.py and all Python/whisper probes in omp setup speech are gone; the recorder (SoX/FFmpeg/arecord) remains the only external tool.
Changed the speech-to-text trigger from the Alt+H keybinding to a hold-Space push-to-talk gesture. Holding the space bar emits an OS auto-repeat burst; once more than 10 spaces land in the editor it recognizes the hold, deletes (tracks back) those inserted spaces, and starts recording, then stops and transcribes when the repeats stop (the space bar is released). app.stt.toggle is now unbound by default but can be rebound to a chord for press-to-toggle; the gesture is gated on stt.enabled, and Shift+Space still inserts a literal space.
Added an experimental, opt-in auto-learn loop (default off, autolearn.enabled). When enabled, after the agent stops it is nudged to capture reusable lessons: durable facts go to long-term memory and repeatable procedures become managed skills — SKILL.md files written to an isolated ~/.omp/agent/managed-skills directory that is discovered and surfaced like authored skills but never overwrites user-authored skills (authored names always win). Two tools back this: manage_skill (create/update/delete managed skills) and learn (record a lesson, optionally minting a managed skill in the same call). learn works with the hindsight, mnemopi, or file-based local memory backend; under local, lessons append to a learned.md in the project's memory root (kept separate from the consolidation artifacts so they survive a consolidation pass) and are injected into future sessions. The nudge is passive by default (a hidden reminder rides the next turn); autolearn.autoContinue instead auto-runs one capture turn at stop, and autolearn.minToolCalls (default 5) gates trivial turns. Plan/goal-mode turns and subagents are never nudged.
Fixed /plan cycling between plan and plan_paused with no path back to mode none, while preserving prompted paused-mode requests. The no-arg third toggle now fully exits — clearing planModeHasEntered and appending a mode_change to "none" — and /plan <prompt> from plan_paused re-enters plan mode and submits the prompt as the first turn (#2510).
Fixed HTML session export rendering empty text tokens (text, userMessageText, customMessageText, toolTitle) as the dark-theme grey #e5e5e7 on every theme not literally named light, making transcripts illegible on custom light themes like sandstone, limestone, and porcelain. isLightTheme now classifies against the resolved statusLineBg luminance (the same surface Theme.isLight uses), while HTML defaultText contrasts the actual export surface (export.cardBg / export.pageBg / derived userMessageBg) so light-status themes with dark export cards keep light transcript text (#2516).

[15.12.6] - 2026-06-14

Breaking Changes

Removed the writeLine and writeLineSync methods from the public SessionStorageWriter contract, requiring custom SessionStorage backends to switch to the append API

Added

Added package-level exports for SessionContext, session entry types, session listing/loader helpers, and migration APIs via session/session-context, session/session-entries, session/session-listing, session/session-loader, and session/session-migrations
Added asynchronous session-write append(...)-based persistence API in session storage implementations so callers can stream writes without sync line-appending methods

Changed

Changed session persistence internals to expose writeTextAtomic(...) on session storage writers for atomic whole-file replacements
Changed online session-title generation to support tool-choice-less title models. Providers/models that cannot be forced to call a tool (chat-completions hosts without tool_choice support such as DeepSeek V4, and Claude Fable/Mythos) are now prompted to wrap the title in <title>...</title> markers instead of the set_title tool call; extraction is lenient, accepting a plain sentence or a truncated/unclosed tag. A TITLE_SYSTEM.md override is reused in this mode with the marker instruction appended.

Fixed

Fixed session JSONL persistence so the first assistant turn materializes the file synchronously, leaves the append writer open, and writes later entries with a sync append writer even during writer-close races instead of waiting on a queued rewrite.
Fixed submitted user messages emitting OSC 133 command-start markers without a matching command-finished marker, which made some terminals group later transcript output under the first prompt instead of appending it normally.
Fixed queued forced tool choices being rejected and requeued or dropped when their named tool is no longer active for the upcoming turn, preventing eager todo and pending-action reminders from forcing unavailable tools. (#1701)

Removed

Removed the re-roots past a cwd-less legacy session in a shared explicit sessionDir relocation test case and the stores symlink-equivalent home cwd sessions under home-relative directories file-operations test case.

[15.12.5] - 2026-06-13

Changed

Terminal resize now repaints only the viewport while a drag is in flight and defers the full transcript replay until the drag settles. Outside a multiplexer, every SIGWINCH used to erase and replay the entire transcript at the new width — re-laying-out (and, for markdown, re-lexing) all of history on each event, work thrown away the instant the next event arrived and re-done dozens of times a second during a drag. The TUI now composes and paints only the visible tail mid-drag — a new ViewportTailProvider fast path that the transcript implements by rendering blocks bottom-up and skipping everything above the fold, touching no commit/scrollback state — then runs the single authoritative rewrap + native-scrollback rebuild ~120 ms after the last resize event.

Fixed

Updated docs/models.md and docs/providers.md to reference the omp models subcommand (and omp models canonical / omp models find) instead of the removed top-level --list-models flag (#2458)
Fixed unknown ---prefixed flags being silently consumed as prompt text, which let a stale or typoed flag start a real agent session (connecting to MCP servers, waiting on the model) instead of failing fast. parseArgs now tracks unrecognized flag-shaped tokens and runRootCommand calls reportUnrecognizedFlags immediately after the post-extension reparse, exiting 2 with Error: unknown flag: --… before any session, MCP, or initial-message work runs. Extension-registered flags still pass cleanly since the validation runs after the extension-aware reparse (#2459).
Fixed prompt templates discovered from cwd/.omp/prompts/ and the agent prompts directory not appearing in the slash-command autocomplete picker. InteractiveMode.refreshSlashCommandState() now feeds session.promptTemplates into the autocomplete provider alongside builtins and file-based slash commands; templates whose names collide with an existing command are skipped to mirror the runtime expansion order (expandSlashCommand precedes expandPromptTemplate) (#2462).
Fixed empty-Enter steering interrupts leaving the newest queued steer visible after aborting an auto-continued queued turn; queued turns now re-drain after every abort/settle cycle.
Added an agent-visible notice when the write tool auto-marks shebang files executable, so agents do not redundantly run chmod +x.

[15.12.4] - 2026-06-13

Breaking Changes

Removed the top-level --list-models flag path and migrated model listing to the new omp models command

Added

Added omp models command to list and manage models with ls, find, canonical, and refresh actions
Added --json output plus -e/--extension, --no-extensions, and --config controls to omp models listings
Added skills.enableAgentsUser and skills.enableAgentsProject settings (default on) so the canonical OMP-native ~/.agent[s]/skills and project-walkup .agent[s]/skills are configurable independently from the third-party Claude/Codex/Pi toggles.

Changed

Model registry merge and omp models / model picker handle unknown context/output limits (null) — unknown limits render as - instead of a fake 222K/8.9K.
Changed omp models to use cached provider data by default and require omp models refresh for a forced online re-fetch
Updated model-resolution errors to point to omp models when a provider or model is not found
Upgraded workspace catalog packages to their latest versions as of 3 days ago, and refactored the ACP agent implementation to be compatible with @agentclientprotocol/sdk version 0.25.0.
Made the zod version requirement in the workspace catalog more tolerant (^4.0.0 instead of 4.4.3), and aligned type definitions in coding-agent extensibility modules.
Changed /logout to pick a stored account after the provider, so multi-account OAuth providers can remove one credential without logging out every account.
Changed the status-line context% to report the provider's real prompt-token count — anchored on the last assistant response, matching the /context panel and the collab host broadcast — instead of an independent cl100k estimate of the whole conversation. The estimate could read several points high and climb past 100% on subscription/Codex models (whose advertised window, e.g. 272K, is already the input budget after reserving max output) while the request was still well within the real limit. Right after compaction the segment now shows ? until the next response re-establishes the true count, and the redundant per-message estimate cache was dropped in favor of memoizing getContextUsage().

Fixed

Fixed ACP thinking-delta mapping to tolerate live chunks that only carry delta text.
Fixed image input to Ollama (local ollama, ollama-cloud, and any ollama-chat model) failing with an opaque HTTP 400 when an attached image was encoded as WebP. Ollama decodes images through llama.cpp / stb_image, which is built without WebP support, so the resize pipeline now auto-excludes WebP for those models — the automatic equivalent of OMP_NO_WEBP=1, applied across every image path (@file mentions and prompt/paste/CLI attachments, the read/inspect_image tools, eval display images, fetched images, and browser screenshots). Other providers are unaffected and still honor OMP_NO_WEBP.
Fixed queued steering/follow-up display to derive from the agent-core queue, so queued chips clear when the core dequeues them and no longer strand after empty-Enter aborts.
Fixed model auth gateway probing to avoid skipping candidates with unknown maxTokens limits (null)
Fixed model listings so providers registered via extensions are now included from -e and configured extensions sources
Fixed /mcp reauth, /mcp test, and /mcp unauth to find and operate on MCP servers reported by /mcp list even when they are only runtime-discovered and not stored in writable config, including namespaced plugin servers like cloudflare:cloudflare-api
Fixed MCP server name validation so colon-namespaced server IDs are accepted when persisting reauth overrides so namespaced OAuth MCP servers can be stored in user config as server:subserver entries
Retried assistant turns that stop with reasoning/thinking only and no final text or tool call, so Gemini/Antigravity thought-only STOP responses continue instead of silently ending the session.
Fixed ~/.agent[s]/skills not appearing as /skill:<name> commands when every named source toggle (skills.enableCodexUser, skills.enableClaudeUser, skills.enableClaudeProject, skills.enablePiUser, skills.enablePiProject) was off: loadSkills gated the agents provider on anyBuiltInSkillSourceEnabled, so a user who turned off the Claude/Codex/Pi sources to clean noise also lost their own canonical OMP-native skills. The agents provider now reads the dedicated enableAgentsUser/enableAgentsProject toggles, decoupled from the third-party fall-through (#2401).
Fixed Windows PowerShell image paste so Ctrl+V can fall back to the PowerShell clipboard bridge when the native clipboard reader reports no image (#2429).
Fixed misaligned box borders in Mermaid ASCII rendering for CJK (Korean/Japanese/Chinese) and emoji labels — affects both fenced mermaid code blocks in assistant messages and the render_mermaid tool. beautiful-mermaid@1.1.3 measures label width in UTF-16 code units while terminals render East Asian characters 2 columns wide; a patchedDependencies entry rebuilds its ASCII renderer to measure terminal display columns (grapheme-cluster aware, wcwidth-style policy). The patch mirrors the upstream PR (lukilabs/beautiful-mermaid#128) and should be dropped once it ships in a release.
Fixed interrupt loader state getting stuck after queued-message aborts by removing the session-layer flush/latch path; empty Enter now aborts the active turn and lets the existing post-unwind queue drain resume normally.
Fixed /goal <objective> and /goal set <objective> during streaming so goal context is steered immediately but objective submission waits for the active turn to finish instead of spamming AgentBusyError (#2454).
Fixed concurrent omp --session startups (e.g. cmux pane restore after an unclean shutdown) crashing with SQLITE_BUSY_RECOVERY while the agent SQLite databases were still under WAL recovery. The auth credential store and AgentStorage.open() retry the SQLITE_BUSY family with bounded backoff, and every shared SQLite open path (AgentStorage, history, autoresearch, memories, github cache, auto-QA grievances, catalog model cache, stats) now installs the busy handler before the first lock-taking statement so transient WAL recovery contention waits instead of crashing (#2421).
Mnemopi per-project / per-project-tagged bank derivation is now stable for one cwd, ignoring the surrounding git layout. Previously the bank id was hashed from git.repo.resolveSync(cwd)?.repoRoot ?? path.resolve(cwd), so adding or removing a .git anywhere above the working directory silently repointed the same conversation to a new bank and stranded its memories (e.g. /home/x/projects/repo flipping between projects-… and repo-…). The derivation in packages/coding-agent/src/mnemopi/config.ts now hashes path.resolve(cwd) directly, and session startup widens the recall set with any sibling bank under <dbDir>/banks/ whose working_memory rows already carry the active cwd in metadata_json.$.cwd, so memories stranded by the old, less-stable derivation become visible again on the next session without manual migration (#2412).
Fixed model switching (Ctrl+P role cycling and the alt+p / /switch / /models selector) intermittently freezing the UI for several seconds. AgentSession.setModel/setModelTemporary ran an eager await modelRegistry.getApiKey(model) purely as an existence pre-flight and discarded the value — but getApiKey does real work: it synchronously executes command-backed key programs (apiKey: "!cmd", execSync with a 10s timeout, blocking the event loop) and refreshes OAuth tokens over the network when one crosses the expiry window (the "fine for a few switches, then a multi-second stall" symptom). Switching now uses the synchronous, side-effect-free ModelRegistry.hasConfiguredAuth check; the concrete key (command execution + OAuth refresh) is still resolved lazily per request via the existing resolver, so an unconfigured provider still fails fast with No API key while a healthy switch never touches the network or spawns a subprocess. hasConfiguredAuth no longer runs the command program or refreshes tokens either, matching its documented "probe without resolving an API key" contract.
Fixed session resume (pi -c / --continue / --session) hanging for ~10s at startup — surfaced by the watchdog as Still starting … phase: createAgentSession > restoreSessionModel — when an OAuth token needed refreshing or the auth broker (OMP_AUTH_BROKER_URL) was unreachable. Picking which saved model to restore is a pure selection that only needs to know whether auth is configured, but restoreSessionModel probed each candidate with the async getApiKey, which refreshes OAuth tokens over the network, executes command-backed key programs, and issues auth-broker requests — so a slow or unreachable endpoint stalled resume for the full refresh timeout per candidate. Startup model selection now uses the synchronous, side-effect-free ModelRegistry.hasConfiguredAuth probe (the same fix already applied to interactive model switching); the concrete key is still resolved lazily on the first request via the resolver.

[15.12.3] - 2026-06-12

Fixed

Restored the intended double-Esc default to /tree and fixed the Esc-opened selector path to reset the terminal display instead of preserving stale scrollback when the selector opens.

[15.12.2] - 2026-06-12

Fixed

Collab links now dot-join the room secret (<roomId>.<key>, host/r/<roomId>.<key>) instead of using a second #: RFC 3986 forbids a raw # inside a URL fragment, so macOS Foundation (behind terminal click-to-open) percent-encoded the browser deep link's second # to %23 and the web client rejected the session. /join, omp join, and the web client still accept legacy #-joined links and leniently decode %23-mangled ones
Fixed brew install can1357/tap/omp failing on macOS with sandbox-exec … exited with 1: the generated Formula/omp.rb (rendered by scripts/ci-update-brew-formula.ts) now stamps using: :nounzip on every per-platform url so Homebrew leaves the bare Mach-O/ELF asset in the staging CWD (the previous Dir["omp-*"].first returned nil because UnpackStrategy::Uncompressed#extract_nestedly nested the file outside it), and wraps generate_completions_from_executable in with_env(HOME: buildpath) so the popened binary's ~/.omp lookup goes to the writable staging dir instead of the sandbox-denied real home (#2398)

[15.12.1] - 2026-06-12

Added

Added the compaction.dropUseless setting (default on): tool results flagged contextually useless are elided by the per-turn cache-aware prune pass and the pre-compaction threshold prune, replaced with [Uneventful result elided]. Built-in tools flag their uneventful outcomes — zero-match/zero-result search/find/ast_grep/recall (warnings included; the follow-up call has already corrected course), empty LSP lookups, job polls where everything is still running (plus nothing-to-wait-for and unknown-id polls), irc wait timeouts and empty inbox drains, and zero-result github searches / run-watch give-ups

Fixed

Restored the default double-Esc behavior to open the editable message-history selector instead of /tree so pressing Esc twice can edit a previous message again (#2396).

[15.12.0] - 2026-06-12

Added

Added /collab view command to create a read-only spectator join link
Added read-only hints and status text for guest-only participation in collab sessions
Added share.serverUrl and share.redactSecrets settings: the share server/viewer base for /share links (default https://my.omp.sh/s) and a toggle (default on) that runs the secret obfuscator over the shared snapshot before upload
HTML session exports now embed subagent transcripts: sub-session files stored next to the session (<session>/<AgentId>.jsonl, recursively) ride along in the export payload, agent ids in task tool cards become drill-down links, and a breadcrumbed overlay renders each subagent's full transcript — including its own tool cards and deeper nested agents — with Esc/backdrop navigation
Added Agent Hub focus mode: Enter on a local agent now retargets the main view to that live subagent session with regular transcript rendering, steering, Esc-to-main return, ←← parent navigation, and a ghost status-line badge.

Changed

Blocked cycling model and thinking presets while a focused subagent session is active and now prompts users to return to the main session first
Changed codexResets.autoRedeem from a boolean to unset/yes/no: unset runs the eligibility check and asks before spending a saved Codex reset, yes redeems without prompting, and no skips the check entirely
Changed collab links so full links with a write token grant mutation rights while links without a token now join as read-only
/share no longer uploads a plaintext HTML export to a gist for gistpreview. It now snapshots the session JSON, gzips and seals it with a fresh AES-256-GCM key, and pushes the blob to a secret gist (when gh is authenticated) or to the share server (capped at 1 MB; oversized sessions are trimmed — images first, then long strings, then oldest entries). The share link is https://my.omp.sh/s/<id>#<key>: the viewer fetches the blob (hex ids from the gist API, others from the relay store) and decrypts it in-browser, so the key never leaves the client. Configured secrets are additionally redacted from the snapshot unless share.redactSecrets is off. Custom ~/.omp/agent/share.{ts,js,mjs} handlers keep the legacy HTML-file contract; /share also works for in-memory (--no-session) sessions now
/collab now prints a join hint with both link forms: the compact omp join link for terminals and a click-to-join browser deep link (https://<relay-host>/#<link>, displayed scheme-less, OSC 8-linked) — the relay serves the collab web client at /, and the room id + key ride in the URL fragment, so they never appear in any HTTP request. /join, omp join, and the web connect screen accept either form
npm installs no longer download fastembed's ~270MB ONNX native dependency tree eagerly: fastembed and onnxruntime-node are external to the bundle and optional peers of @oh-my-pi/pi-mnemopi, fetched on demand only when Mnemopi local embeddings are first used
The job tool prompt now documents that omitting poll waits on all running jobs, so agents stop enumerating every job id to poll everything
Collapsed task tool blocks now cap the per-agent list at 4 rows: the live progress view keeps the running/pending tail visible behind a … N more agents (…) summary line with per-status counts, finalized batches keep failed/aborted rows visible while folding the slowest successes, and the streaming call preview caps at the same 4 (expand shows the full list)
The anchored Subagents HUD above the editor now lists only detached background spawns: sync task calls (the parent turn is blocked and the inline tool block already renders live progress) and eval agent() helpers (rendered by their eval cell's own progress tree) no longer appear in the list
Completed rows in the job tool output now show the standard done checkmark instead of the watch glyph that replaced the spinner
HTML session exports render tool calls through the same React tool renderers the collab web client uses: per-tool views for all built-in tools (bash, edit diffs, todo boards, eval cells, task batches, LSP, search, browser screenshots, …) are bundled as an <omp-tool-view> web component into the export instead of the previous string-built dummy renderers
Modernized the HTML export page chrome to match the tool-card design language: hairline borders, dense mono typography, compact role-tinted message cards, refined tree sidebar and filter controls, themed scrollbars, collapsed-by-default thinking blocks, and mobile sidebar drawer — derived from the active theme's variables so light and dark themes both render correctly
Agent Hub's embedded chat overlay is now reserved for collab guest viewing; local agents open in the main session renderer instead.

Fixed

Filtered silent abort markers from Agent Hub assistant output so failed turns now render as normal error text
Reset focus-session state when switching transcript rendering between the main session and a focused subagent so streaming/progress context no longer leaks across sessions
Fixed read-only collab sessions so prompting, interrupts, and other write actions are blocked with a read-only warning instead of being applied
Fixed Mnemopi local embeddings in bundled and compiled installs failing with Cannot find module '../bin/napi-v3/.../onnxruntime_binding.node': the Bun bundle inlined fastembed's loader so its relative native require resolved against dist/cli.js. fastembed/onnxruntime-node are no longer bundled; on first use Mnemopi bun installs the pinned pair into ~/.omp/cache/fastembed-runtime/<version-key> and loads the binding from there (#2389)
Fixed the interactive Model scope startup banner so models without an explicit thinking level do not show :undefined, and entries that were scoped without a :level are no longer rendered with the global default thinking level (which applyRootSessionOptions pre-fills on the cycling array for Ctrl+P) (#2385).

[15.11.8] - 2026-06-12

Added

Updated collab link handling to accept compact roomId#key links and relay hosts without explicit scheme when joining or starting sessions
Added /collab stop and /collab status options to control and inspect active shared sessions
Added /collab, /join, and /leave for live session sharing: the host shares an end-to-end encrypted link (AES-256-GCM key only in the link fragment; the relay sees opaque bytes) and guests render the session natively in their own TUI — streaming text, tool cards, footer state, ctrl+o expansion, /dump — and can prompt or interrupt the host's agent. Guest prompts render with an author badge; session-mutating commands stay host-only. Defaults to the public relay.omp.sh relay (collab.relayUrl); a self-hostable Go relay lives in the pi-www repo as omp-collab-relay
Added collab.relayUrl and collab.displayName settings plus a collab status-line segment showing the participant count (host) or guest role. Guests mirror the host's real model/thinking state into their replica agent and the host's subagent ecosystem end-to-end: the live subagent HUD, the Agent Hub table with live progress, hub chat/kill/revive (routed to the host), and on-demand subagent transcript viewing
Added omp join <link> subcommand that launches the interactive TUI and immediately runs /join <link>

Changed

Moved collab live-session wire contracts into @oh-my-pi/pi-wire and stopped broadcasting unsupported session events or entries to collab guests.
The Ctrl+P role-cycle track and the plan-approval model slider now color segments by track position from the theme's own palette (accent/success/warning/error + markdown/syntax hues, deduplicated per theme since many themes alias them) instead of role-keyed colors with a gray fallback for custom roles
Interactive boot no longer blocks first paint on MCP server discovery. For UI sessions, createAgentSession returns immediately and connects to configured MCP servers in the background; their tools and slash commands stream in through the existing live-refresh channel once connected. An explicitly requested MCP tool whose server has not finished connecting resolves to a deterministic "still connecting" placeholder instead of an "unknown tool" error, and each server's instructions join the system prompt once its background connection completes — carried in on the same live-refresh that registers its tools — so server-provided instructions are preserved, not dropped. tools.discoveryMode: "auto" is re-resolved once the background connect reports the real MCP tool count, so a toolset large enough to cross the threshold flips discovery on (registering search_tool_bm25) instead of force-activating every tool, and a session disposed while servers are still connecting disconnects them instead of resurrecting their tools onto the dead session. Non-UI modes (print/rpc/acp) keep the blocking discovery path. Measured ~290 ms (≈24% of cold boot) off the first-paint critical path with MCP servers configured.
Assistant-message streaming is cheaper per token. During a stream the component now reuses its Markdown subtree across reveal ticks — only the growing block is re-rendered — instead of tearing down and re-lexing every block on each ~30 fps tick, and grapheme counting in the reveal controller is memoized. A completed thinking block that precedes a still-streaming answer is no longer re-highlighted every frame (~66% less render work on think-then-answer streams in benchmarks; single-block streams are unchanged).
Cold boot no longer builds the model catalog's canonical-equivalence index on the first-paint critical path. The ModelRegistry constructor built buildCanonicalModelIndex over the entire ~3,200-model catalog synchronously (~210 ms); it is now built lazily on first read (getCanonicalModels/getCanonicalVariants/getCanonicalId, reached by the model picker and by enabledModels/default-role pattern resolution), which a default interactive launch never touches before paint. Measured ~244 ms (≈16% of cold-boot wall) off first paint; the picker pays the one-time build on first open.
Repeat read summaries of an unchanged file no longer re-run the tree-sitter parse. The per-session summary is memoized on the content hash of the freshly-read bytes — the file is still read fresh on every call, so results stay correct without a staleness window — dropping a repeated same-file summary read from ~17 ms to ~2.5 ms.
Attributed the previously-unlabeled synchronous boot region in the PI_TIMING startup table with modelRegistry:init, buildCanonicalModelIndex, and initTelemetryExport spans.

Fixed

Fixed remote (SSH) image attachment silently inserting the local-machine path as plain text: when omp runs over SSH and the terminal forwards a bracketed-paste image path, the path is unreachable from the remote host. The path-as-text fallback is gone for unreachable paths and the status now points the user at the clipboard image-paste shortcut so the bytes actually cross the connection (#2375).

Security

Rejected non-local ws:// relay URLs and invalid room keys when parsing collab links to prevent insecure or malformed session joins

[15.11.7] - 2026-06-12

Added

Added mouse support to the setup wizard, including hover, wheel scrolling, and click-to-select for provider sign-in, web-search provider, glyph, and theme option lists
Added pointer support for the providers tab bar so tabs can be selected by click and wheel scrolling works inside the active panel
Added left-click navigation in setup wizard splash/outro screens to advance or exit without pressing Enter
Added bench CLI command to benchmark model selectors with a shared prompt and report time-to-first-token and throughput
Added omp bench options --runs, --max-tokens, --prompt, and --json to control run count, response length, prompt text, and machine-readable output
Added benchmark failure reporting that shows per-run errors, flags failed runs in the summary, and exits with a non-zero status when any model benchmark fails
Added snapcompact.shape setting to pick the frame variant snapcompact prints text with — auto (each provider's eval winner) or any research-eval variant: the square grids (8x8r/8x8u/6x6u/5x8 × sentence-hue/black ink) and the per-model winners (6x12-dim, 8x13-bw, 8on16-bw, doc-8on16-bw, doc-8on16-sent, doc-8on16-sent-dim); applies to both the compaction archive and inline system-prompt/tool-result imaging

Changed

Updated snapcompact.shape option help text so the auto mode and 8on16-bw descriptions now reflect OpenAI’s current default and GPT winner variants
Model selectors keyed by retired effort-tier variant ids keep resolving after catalog collapsing: reference resolution and bare-id matching consult the hand-table aliases (gemini-3.5-flash-low, gemini-pro-agent, recycled gemini-3-flash) plus the X-thinking → X grammar for auto-collapsed pairs, with exact matches always winning while a raw id is live; explicit :effort suffixes transfer unchanged. models.yml modelOverrides and rate-limit selector suppressions keyed by raw member ids re-key onto the collapsed model
Custom/config provider model lists now collapse X/X-thinking twins at registry rebuild (collapseBuiltModelVariants), folding config-defined twins into one entry whose thinking toggle routes to the -thinking backing id

Fixed

Fixed Ctrl+T thinking-block toggles clearing the pending user message and loader before the assistant stream starts (#2370).
Fixed Windows CodeGraph MCP launches from npm .cmd shims by resolving the shim's Node entry point and spawning it directly, keeping stdio attached to CodeGraph instead of the transient cmd.exe wrapper (#2367).
Fixed snapcompact archives going partially unreadable on OpenRouter, which hard-caps requests at 8 images and silently drops the excess: compaction now passes a provider-aware frame budget (providerFrameBudget) so the archive never exceeds the cap (unknown providers get a safe floor of 5), with overflow kept as a text tail on the summary instead of rendering frames the gateway would drop; inline system-prompt/tool-result imaging shares the same per-provider budget (providerImageBudget), so once existing images exhaust it (e.g. 8 archive frames on OpenRouter) tool results ship verbatim as text
Fixed Hindsight retains to send offset-aware local timestamps instead of UTC Z strings so extraction prompts keep the user's local time-of-day context (#2363).
Fixed tool calls taller than the viewport reading as cut off while streaming (the head reappeared only once the result landed): the 15.11.6 stranded-preview fix marked every collapsed pending tool preview commit-unstable, which also blocked durable top-anchored streams — e.g. a task call's context/assignment markdown — from reaching native scrollback mid-run. Commit stability is now classified per renderer (ToolRenderer.provisionalPendingPreview): only the tail-window previews the result render re-anchors (edit/apply_patch streamed-diff tails, bash/ssh command caps, eval cells with interleaved outputs) stay provisional; every other pending preview commits its settled head mid-stream again
Fixed omp bench reporting "tokens 0, TPS 0.0" successes on repeated OpenRouter runs: pi-ai opts every OpenRouter request into response caching, so bench's byte-identical request replayed a cached generation with zeroed usage at edge speed. Bench now sends X-OpenRouter-Cache: false so every run measures a fresh generation
Fixed omp bench failing with HTTP 400 {"detail":"Instructions are required"} against openai-codex models: bench requests now carry a minimal default system prompt (same guard as eval's completion bridge)
Fixed pressing Ctrl+T (toggle thinking blocks) — or hitting any other rebuild path such as the theme/preset selector — during the pre-streaming window after a submission erasing the just-submitted user message until the first assistant token arrived: the user message is rendered optimistically before session.prompt(...) lands it in session entries, so rebuildChatFromMessages() had no record of it; it now replays an in-flight optimistic submission after rebuilding the transcript, gated on optimisticUserMessageSignature (cleared by EventController once the real message_start lands) so it cannot duplicate post-streaming (#2372).

[15.11.6] - 2026-06-12

Changed

Renamed the saved Codex reset command from /reset-usage to /usage reset and added /usage show as an explicit alias for plain /usage.

Fixed

Fixed the Subagents HUD order so progress updates no longer move active rows; batched subagents keep their task order.
Fixed /settings dirty-value highlighting so changed values stay orange while the row has keyboard focus.

[15.11.5] - 2026-06-12

Added

Added codexResets.autoRedeem, codexResets.minBlockedMinutes, and codexResets.keepCredits settings so Codex users can opt in to automatic spending of saved rate-limit resets
Added automatic Codex reset redemption flow so a blocked weekly usage-limit request can be retried immediately after consuming a saved reset when auto-redeem is enabled
Added --history mode to omp usage to show recorded usage-limit history instead of a single live snapshot
Added --days (-d) support for omp usage --history to control the history window (default 7 days)
Added historical usage output for omp usage --history, including per-limit sparklines with latest and peak percentages
Added --json support in omp usage --history to emit timestamped usage-history snapshot data
Added guidance and exit code 1 when no usage history is available for the selected window
Added an anchored Subagents HUD above the editor (next to the Todos block) listing every running subagent as Id: description; rows appear on spawn and the block clears itself when the last subagent finishes
Added /reset-usage command to spend a saved Codex rate-limit reset by running /reset-usage <active|account id|email>, with a TUI selector flow for interactive reset redemption
Added snapcompact.systemPrompt enum modes none, agents-md, and all, so users can disable system-prompt imaging, image only loaded AGENTS.md/context-file instruction sections, or image the full system prompt
Added agents-md snapcompact mode that rasterizes AGENTS.md-style context sections from the system prompt into frames and appends them to the first user message
/context (TUI panel and ACP report) now shows estimated snapcompact wire savings when snapcompact.systemPrompt or snapcompact.toolResults is enabled — per-feature text → frames token deltas, the reason a swap does not apply (savings margin, image budget, or text-only model), and the estimated size of the next request. The estimate and the live provider-request transform share one planner (planInlineSwaps) so displayed numbers cannot drift from wire behavior.
Added /debug dump-request and /debug next-request as aliases for /debug dump-next-request when arming a one-shot AI provider request dump
Added /debug dump-next-request <path> to dump the next AI provider HTTP request JSON to a chosen file.

Changed

Changed usage views (/usage reports, status summaries, and account headers) to show available saved Codex rate-limit resets and point to /reset-usage for spending them
Changed /context system-prompt savings lines to show the scope as AGENTS.md or all, so savings estimates now indicate whether imaging applied only to context-file instructions or the entire prompt
Changed /debug handling in interactive mode so /debug with arguments now executes the requested debug subcommand instead of always opening the debug selector
Changed /debug dump-next-request path handling to expand ~ and resolve relative paths against the current working directory
Changed the task tool's TUI block: the header now shows the task dispatch glyph (tool.task) while agents are in flight instead of a spinner (async spawns return immediately, so a spinner misread the call as blocking), and per-agent rows use one static dot for every state — completed rows keep the same dot and settle from accent to the plain foreground color instead of switching to a different status glyph
Compressed the eval, browser, read, and irc tool prompts (6156 → 4932 tokens o200k, −20%): deduplicated claims across sections, tightened helper reference descriptions, trimmed redundant examples; input grammar, examples, and template conditionals (py/js/spawns, read display modes) unchanged
Changed the Nerd Font tool.task icon to the Octicons tasklist glyph.

Fixed

Fixed settings search to rank matching tabs by relevance so exact matches appear before incidental matches
Fixed the /settings search bar ignoring regular editor hotkeys: the query is now a real single-line input with cursor movement (←/→, word jumps, Home/End), word deletion (Alt+Backspace, Ctrl+W), kill/yank, undo, and paste support
Fixed snapcompact toolResults imaging confusing models into reporting tool malfunctions: the note prepended to rasterized tool results now tells the model the result is in the PNG frame(s) below and that the image delivery is deliberate, not a tool error, instead of the bare [Rasterized] marker
Fixed tool-call boxes rendering "inside themselves" after slow tool runs: a pending collapsed preview that sat byte-static past the stable-prefix window (e.g. an edit's tail-window diff while the apply + LSP pass ran) had its settled head committed to native scrollback, and the result render then stranded that stale call-box fragment above the final block. Pending collapsed tool previews are now provisional (isTranscriptBlockCommitStable) and never enter scrollback mid-run; expanded top-anchored streams keep committing mid-stream
Fixed the higher-level Responses retry classifier to recognize OpenAI Zero Data Retention rejections (Previous response cannot be used for this organization due to Zero Data Retention) as stale-replay errors. The provider-level fix in @oh-my-pi/pi-ai covers the common case in a single turn; this layer additionally guarantees that any ZDR error that still bubbles up resets the Responses session and retries at zero backoff instead of falling back to a different model (#2341).

Removed

Removed the todo task notes feature: op: "note", the text payload, task notes state, markdown note blocks, and note rendering are no longer supported.

[15.11.4] - 2026-06-12

Added

Added mouse-driven interaction to /settings, including tab and setting row hover highlighting, wheel scrolling, and left-click activation for entries and submenus
Added fullscreen /settings mouse-event handling so scrolling and clicks work in an alternate-screen overlay
ModelRegistry.resolver now accepts a model directly — resolver(model, sessionId) — deriving provider, baseUrl, and modelId from it; all model-scoped call sites migrated from the verbose resolver(model.provider, { sessionId, baseUrl, modelId }) form.
Added experimental snapcompact.systemPrompt and snapcompact.toolResults settings (off by default, /settings → Context → Experimental) that render the system prompt and large historical tool results as dense snapcompact PNG frames on vision-capable models to cut token cost. Frames are built per-request in the provider-context transform, cached across turns, capped by a per-provider image budget, and gated on a token-savings estimate — they never reach session.jsonl.
Added a Personality selector to /settings (Model → Prompt): default (the previous built-in reply style), friendly, pragmatic, or none. The selected spec renders into a dedicated <personality> system-prompt block (extracted from the former <reply-guidelines> section) and applies to the live session immediately; subagents always omit the block.
Added mnemopi.polyphonicRecall and mnemopi.enhancedRecall config.yml settings (off by default, /settings → Memory → Mnemopi) that enable the mnemopi 4-voice polyphonic recall engine and the tiered query result cache without environment variables; MNEMOPI_POLYPHONIC_RECALL / MNEMOPI_ENHANCED_RECALL still override the configured values when set (#2323).

Changed

Changed /settings keyboard navigation so Tab and Shift+Tab toggle focus between section headings and setting rows in sectioned lists, with ↑/↓ jumping sections, ←/→ switching tabs, and status text reflecting the active controls
The task tool description now mandates batching parallel spawns into one call's tasks[] (sharing context once) instead of presenting multiple task calls per message as an equal alternative; separate calls are reserved for different agent types or unrelated context
Updated /settings on-screen navigation hints to match the new section-focus behavior (↑/↓ and Tab/Enter) and tab-switching arrows
Changed /settings to open as a full-screen overlay on the alternate screen so it no longer shares space with the underlying transcript
Codex, Gemini, and Perplexity web search now route their OAuth bearers through the new withOAuthAccess driver: a 401 or usage-limit force-refreshes the same account and then rotates to a sibling instead of failing the search, while identity metadata (chatgpt-account-id, Google projectId) is re-derived from the refreshed credential on every retry.
Kagi web search, the xAI TTS tool, and model-discovery list fetches now resolve their bearers through withAuth with an auth-storage resolver instead of a one-shot key snapshot, gaining the same force-refresh + rotate retry on 401.
Compaction, handoff, and branch-summarization call sites now hand the compactor a per-candidate API-key resolver (availability still gated on a key snapshot), so credential refresh happens before the #986 fallback-model loop advances.
The mnemopi backend now passes an OpenRouter resolver for default embedding/extraction setups (AuthStorage-stored keys included), keeping pinned literal keys and custom endpoints unchanged.
Reorganized the /settings panel for findability: every tab now has titled sections backed by a per-tab layout contract (TAB_GROUPS); on wide terminals the panel renders a section sidebar with the active section's settings beside it (narrow terminals keep a flat list with inline headings), and PgUp/PgDn jump section-to-section. The Editing tab became Files (edit/read/LSP) and a new Shell tab hosts bash, eval, and Python settings. Misplaced settings were rehomed: bash toggles united under Shell, tool approval mode and policies together under Interaction → Approvals, marketplace auto-update next to startup update checks, and the todo auto-clear delay beside the other todo settings.
Restyled the /settings chrome: the tab strip moved below the content as a label-less footer, the panel keeps one constant height across navigation, value changes, and submenus (no more viewport jumps after each change), and typing now runs a global cross-tab search — results group under per-tab headings, the footer shows live match counts with non-matching tabs muted, Tab hops between matching tabs, and Esc exits search landing on the selected result's tab.
Normalized /settings labels and descriptions: consistent Title Case labels (e.g. "Todo Auto-Clear Delay", "GitHub View Cache"), uniform unit placement, articles and verb-first phrasing in descriptions ("If false…"/"Whether to…" rewritten), and a stale browser-tool description (Ulixee Hero) corrected to the actual puppeteer/Chromium implementation.
/settings section headings are now underlined — the active section's heading stays bold, and headings outside the active section render dim with the same underline — so section boundaries read at a glance.
/settings: Tab now toggles keyboard focus between section headings and the setting rows — while headings are focused, ↑/↓ jump whole sections and Enter/Esc drop back into the rows — instead of cycling tabs. ←/→ still switch tabs everywhere, tabs without sections (e.g. Plugins) keep Tab as tab-switching, and the footer hint follows the focus state.
Image-generation (Antigravity, xAI, OpenRouter, Gemini) and xAI TTS request failures now throw pi-ai's typed ProviderHttpError carrying status and response headers instead of Object.assign-patched Errors.
Collapsed bash, ssh, and eval previews now cap the command/code section to a viewport-sized tail window (terminal rows minus a chrome reserve) that renders identically while streaming and after completion, with ctrl+o as the only way to uncap. Previously bash/ssh capped the command only while streaming and snapped it fully open the moment the tool finished, and eval never capped cell code at all.
The job renderer now shimmers running-job labels while the poll block is live, freezing to static once the block seals, and drops the id column when the label repeats it — task jobs label themselves with their agent id, so rows read ⟨task⟩ SessionTree instead of SessionTree ⟨task⟩ SessionTree.
Task progress rows now render static text with the task icon instead of shimmering the description or showing the pending/hourglass status glyph.
app.clipboard.pasteImage (Ctrl+V) now falls back to pasting clipboard text when no image is present, so hosts that deliver only that chord (VS Code's integrated terminal forwarding Ctrl+V, Windows clipboard history via Win+V) cover both payload kinds; WSL text reads now reach the Windows clipboard through host PowerShell like image reads already did (#1628).
Changed /usage to show the OAuth account currently selected for the active model provider when usage reports include multiple accounts, making multi-account sessions easier to verify without marking unrelated providers.

Fixed

Fixed session tree connector rendering in /tree and the HTML export: flattened chain rows under a last-sibling (└─) branch drew a dangling │ in the corner column itself and shifted columns once the chain branched deeper; the chain anchor now sits one level right, below the branch head's content, so connectors terminate at └─ and indentation stays stable (#2325).
Fixed prompt and image input typed while the agent is idle between turns from being dropped so steering now queues and auto-resumes processing once the session is ready
Fixed image-bearing prompts queued during compaction registering their local-submission signature without the image count, so the delivery event treated them as foreign messages — clearing any draft typed since queuing and leaking the stale signature.
Fixed Esc/Alt+Up queue restoration dropping attached images: clearQueue() and popLastQueuedMessage() now hand queued images back alongside the text, and the restore paths return them to the pending-image buffer instead of silently discarding them.
Fixed a failed steer submission in the no-input-waiter path discarding the message: the text and images are restored to the editor for retry instead of being lost to history-only recovery.
Fixed local memory consolidation on Responses-style models that reject user-only requests by sending a dedicated stage-two system prompt.
Fixed the settings sidebar divider alignment in /settings by locking sidebar width to the widest group name so switching tabs no longer shifts the layout
Fixed plan-review overlay mouse hit-testing so wheel, hover, TOC, and body clicks map to the intended regions
Fixed lsp.formatOnWrite sending a hardcoded tabSize: 3, insertSpaces: true on every textDocument/formatting request, which silently re-indented 2-space YAML (and any LSP-formatted file) to 3-space on every write/edit through formatter-aware servers like yaml-language-server. Format options are now resolved per-file from .editorconfig (indent_size, indent_style, tab_width), falling back to the indent sniffed from the in-memory content the agent is about to write, then to a 2-space default. The duplicate DEFAULT_FORMAT_OPTIONS constant in lsp/index.ts and lsp/clients/lsp-linter-client.ts is replaced with a single shared resolveFormatOptions helper (#2329).
Fixed stale OpenAI Responses replay failures such as Item with id 'rs_...' not found. by resetting the provider session and retrying without advancing fallback chains.
Fixed /settings Escape handling so an open submenu receives Esc and returns to the settings list before a second Esc closes the panel (#2331).
Fixed Escape not closing /settings on the Plugins tab while the async plugin list is still loading: PluginSettingsComponent now closes on Esc while no child view is mounted, and an npm plugin registry listing failure is caught (like marketplace failures) so a bad registry no longer leaves the tab permanently blank (#2331).
Fixed unconfigured pi/smol, pi/slow, and pi/designer agent model roles using cloud-priority defaults before the user's configured modelRoles.default, which could route local-default setups to authenticated paid providers (#2336).
Fixed issue:// reads failing on older GitHub CLI releases that reject the optional stateReason issue JSON field; single issue reads now retry without it and issue listings no longer request it (#2333).
Fixed image generation ignoring /login-stored OpenRouter and Google API keys: provider selection and requests now resolve through the model registry (with env-var fallback) instead of environment variables only.
Fixed detached (task async spawn) progress snapshots repainting commit-eligible rows after the block leaves the live transcript region; later partial snapshots are dropped while the final completion snapshot still applies.
Fixed five consumers treating Zod tool-parameter schemas as plain JSON Schema objects, leaking schema-instance internals (def, shape, methods stringified as undefined) or silently reading nothing: /dump and the RPC get_state dumpTools payload now convert parameters through the same wire-schema conversion providers receive; context-usage token estimation no longer stringifies the Zod def tree (overcounting tool schema tokens in the status line); tool-discovery search indexing recovers schemaKeys from Zod tools (previously empty, weakening BM25 ranking); and the extension inspector panel renders Zod tool arguments instead of "(no arguments)".

[15.11.3] - 2026-06-11

Fixed

Fixed stale Working… loader rows being committed to native scrollback above the live loader: the interactive status container now reports a live-region seam while it has mounted content (#2328 by @35844493).
Fixed Mnemopi memory consolidation never running on session shutdown: MnemopiSessionState.dispose() now drains pending fact extractions and runs sleepAllSessions on every owned bank before closing handles (and AgentSession.dispose() awaits the result), matching the /memory enqueue slash command. mnemopiBackend.clear opts out via dispose({ consolidate: false }) so the destructive /memory clear path does not spend tokens consolidating memories that are wiped on the next line. consolidate() deliberately keeps no aliasOf short-circuit so /memory enqueue from a subagent still flushes and sleeps the parent's shared banks (the alias guard lives in dispose for lifecycle, not in consolidate for content). Without this, episodic_memory, gists, consolidation_log, graph_edges, and triples stayed empty for every deployment because the SHMR/beam pipeline only ran when a user typed /memory enqueue|rebuild (#2320).
Fixed multi-path search collapsing distinct scopes into one walk rooted at their unrequested common ancestor: paths: [".", "~/.gitconfig"] scanned the entire home directory (typically until the 30s grep timeout), and explicit file entries inside walker-pruned directories (e.g. .git/config, gitignored files) silently never matched because they were folded into the directory walk's glob union. Multi-path scopes now fan out into per-target scans when the common ancestor is not itself a requested path, plain-file entries are read directly as their own targets, and matches from overlapping targets are deduplicated.
Fixed lsp.formatOnWrite sending a hardcoded tabSize: 3, insertSpaces: true on every textDocument/formatting request, which silently re-indented 2-space YAML (and any LSP-formatted file) to 3-space on every write/edit through formatter-aware servers like yaml-language-server. Format options are now resolved per-file from .editorconfig (indent_size, indent_style, tab_width), falling back to the indent sniffed from the in-memory content the agent is about to write, then to a 2-space default. The duplicate DEFAULT_FORMAT_OPTIONS constant in lsp/index.ts and lsp/clients/lsp-linter-client.ts is replaced with a single shared resolveFormatOptions helper (#2329).

[15.11.2] - 2026-06-11

Added

Added the Expert Elixir language server (expert, invoked as expert --stdio) to the built-in LSP server list, auto-detected for Mix projects (mix.exs/mix.lock). When both are installed, elixir-ls remains the primary navigation server (Expert is ordered after it).
Added magicKeywords.enabled and per-keyword magicKeywords.ultrathink, magicKeywords.orchestrate, and magicKeywords.workflow settings to disable hidden magic-keyword notices and ultrathink auto-thinking escalation (#1796).
Added external-editor support for Plan Review section annotations, preserving multiline feedback for Refine plan (#2305).
Added plain-RPC slash command discovery with command source metadata and startup/update notifications (#2261).

Changed

Bash tool calls in one assistant message now run in parallel instead of serializing the batch: non-pty bash is scheduled as a shared tool (pty: true stays exclusive), and overlapping calls on the same shell session key degrade to isolated one-shot shells so they cannot queue behind or abort each other

Fixed

Fixed CDP attach target selection for Chromium/Edge endpoints that expose page targets through discovery before browser.pages() is populated, and improved ws:// app.cdp_url diagnostics (#2246).
Fixed local memory consolidation on Responses-style models that reject user-only requests by sending a dedicated stage-two system prompt.
Fixed extension, custom-tool, custom-command, and hook loaders injecting pi through bare @oh-my-pi/pi-coding-agent self-imports, which could pull a different cached package version into global installs during plugin load and trigger mixed-runtime stack overflows.
Marked unsmoothed assistant streaming renders as transient so streamed code blocks can avoid synchronous syntax highlighting until the message finalizes.
Routed LSP hover code rendering through the shared cached highlighter instead of calling the native highlighter directly on each render.
Kept smooth assistant streaming renders transient until message_end so catch-up frames do not synchronously re-highlight still-growing code blocks.
Fixed the job tool's poll header repeating the running count ("waiting on 19 of 19 19 running"): the title now reads "waiting on N jobs" (or "waiting on N of M jobs" when some settled) and the dim meta lists only settled categories (done/failed/cancelled)
Fixed irc send await:true to a busy peer stranding the sender until timeout when async execution is disabled: the recipient (e.g. Main blocked in a synchronous task spawn awaiting the sender's own batch) can never reach a step boundary to reply, so it now generates an ephemeral side-channel auto-reply from its context (the pre-mailbox respondAsBackground behavior), records it as an irc:autoreply aside in its own history, and delivers it back over the bus with replyTo threading
Fixed Windows-style backslash paths and globs being parsed as literal POSIX path segments by search/find tools (#2245).

[15.11.1] - 2026-06-11

Added

Added the statusLine.transparent appearance setting (default off): when enabled, the status line skips the theme's statusLineBg fill and powerline end caps so the bar inherits the terminal's default background — useful in Ghostty and other terminals whose theme background does not match the theme's hardcoded status-line color (#2306)

Changed

Changed display.smoothStreaming to animate streamed tool-call arguments (including write/edit/bash previews) at the smooth streaming cadence so partial tool input now appears progressively instead of in large jumps
Changed streamed updates for custom wire tools to render incremental raw input as { input } prefixes for preview
async.enabled now defaults to true, keeping background task execution and async bash available out of the box; disable it to force blocking subtasks (#2301).
Shortened the status line's background-job indicator to a dedicated gear icon plus count (e.g. ⚙ 1 instead of 👥 1 job running) and moved it left of the session name in the right segment group

Fixed

Fixed the task tool's streaming call preview rendering the per-agent list above the shared context/assignment briefs: agent rows now render below them, matching the result layout, so the list no longer pushes the brief down as items stream in and then jumps below it once the first progress snapshot arrives
Fixed interrupted or ending tool calls to flush streamed argument previews to the full received payload instead of freezing on a partial reveal chunk
Fixed tool-argument reveal to avoid splitting UTF-16 surrogate pairs at frame boundaries
Fixed input lag in long sessions whenever a spinner was visible: loader ticks forced a full TUI re-compose (re-walking every transcript block) 12.5 times per second. Animations that only change their own rows (status loaders, welcome intro, voice mic glyph, /btw streaming panel) now use the new component-scoped render path (TUI.requestComponentRender), which reuses every other root subtree's rows; resize, overlays, inline images, forced repaints, root-list changes, or any concurrent full request still compose the full frame
Fixed transcript rendering so blocks with rows committed to scrollback while still streaming rerender on the first finalized frame, so the final committed output is visible instead of stale in-flight lines
Fixed committed transcript bypass logic to re-render finalized blocks when their version changes, so post-finalization updates like restored inline errors and late tool-result content now appear in scrollback
Cached stable edit/write preview highlighting inside each tool component so steady-state frame renders do not re-enter the native syntax highlighter.
Skipped rendering finalized transcript blocks whose rows are already committed to native scrollback, keeping long restored sessions from re-rendering historical tool blocks every frame.
Fixed MCP OAuth authorization and token requests to include the required resource indicator for the target MCP server.
Restored async.enabled=false as the task tool's blocking mode, so subagents no longer become background jobs when async execution is disabled (#2301).
Fixed the /tree selector and HTML session export dropping the inherited │ gutter for chain rows under a last-sibling branch, so the conversation flow under a └─ branch stays visually anchored to its parent message (#2298).
Fixed Anthropic web search requests to include metadata.user_id, matching the main Messages path: API-key requests forward the active session id verbatim, OAuth requests build the Claude-Code-shaped {session_id, account_uuid?, device_id} JSON envelope so gateways see consistent attribution (#2295).
Fixed Anthropic classifier refusals to switch through configured retry fallback chains without same-model retries, and to keep the fallback pinned for the conversation. (#2290)
Fixed the edit-tool hashline prompt to stop steering agents toward insert after block N: on closing delimiter lines; opener-only block anchors now point visible closing-line insertions to plain insert after M:. (#2292)

[15.11.0] - 2026-06-10

Breaking Changes

Removed the resume option from the task tool API and its resume execution path; continue work on finished subagents by sending follow-up messages via irc instead
Removed the irc.enabled setting: irc availability is now derived — the tool exists exactly when there is someone to message (the session can spawn subagents through task, or it is a subagent itself). A stale irc.enabled key in config is ignored
The task tool was reworked to always run spawns in the background as independent, persistent agents: results arrive as async job deliveries (block with job poll only when genuinely needed). The wire schema is now shape-swapped by the new task.batch setting (default on): { agent, context, tasks[] } — one subagent per task item, per-item isolated, and a required shared context — or, when disabled, a flat single-spawn shape { agent, id?, description?, assignment, isolated? } with shared background passed via local:// files instead
Removed the task.simple setting and the task tool's per-call schema parameter outright: structured subagent output now comes only from the agent definition's output frontmatter or the inherited session schema, and ad-hoc structured workflows use eval agent(prompt, schema). A stale task.simple key in config is migrated away
Reworked irc to send/wait/inbox/list ops over a per-agent mailbox bus: the blocking awaitReply auto-reply turn is removed — send is fire-and-forget with delivery receipts, and replies are real turns by the recipient observed via wait (or the send await: true sugar)
Removed the context argument from eval agent() in both the JS and Python preludes: pass shared background via a local:// file referenced in the prompt
Replaced the standalone session-observer overlay with the Agent Hub: app.session.observe (ctrl+s) now opens the hub, whose chat view absorbed the observer's transcript renderer

Added

Snapcompact compaction now passes the session model so frames render in the provider-optimal shape (unscii 8x8r-bw for Anthropic-family/unknown APIs, 8x8r-sent for Google, Lanczos-stretched 6x6u-sent with detail: "original" for OpenAI), per the snapcompact 200k-token evals
Added per-turn supersede pruning of stale read results: when a file is re-read, older copies of the same path/selector are pruned from context at cache-favorable moments (small suffix, idle gap, or alongside overflow pruning). Gated by the new compaction.supersedeReads setting (default on)
Added soft request budgets for task subagents (explore/quick_task 40, others 90, configurable via task.softRequestBudget, 0 disables): crossing the budget injects a one-time wrap-up steer into the child; crossing 1.5× aborts the run gracefully
Added cancelled/aborted subagent salvage: instead of (no output), merged task results now carry the child's last activity snippet plus request/token stats, and per-child stats lines include request counts
Added a repeat-read notice to the read tool: the third and later reads of the same file in a session append a one-line note suggesting range re-reads or the context echoed in edit results
Added a hard inline byte cap (~50KB) at the bash and browser tool-result boundaries with head/tail elision and an artifact:// footer for the full output, closing paths that previously let 100KB+ results land inline
Added the Agent Hub overlay (ctrl+s, alt+a, or double-tap left arrow on an empty editor): a live table of registered subagents (status, unread IRC count, current task, last activity) with per-agent chat — Enter opens a transcript + input line that steers a running agent, prompts an idle one, and revives a parked one; r revives and x aborts/releases the selected agent
Added the snapcompact compaction strategy (compaction.strategy: "snapcompact"): history is archived onto dense bitmap "snapcompact" frames a vision model reads back directly, instead of an LLM-generated summary — instant, free, and verbatim. Auto compaction (including overflow recovery) and manual /compact both honor it; falls back to context-full with a visible warning notice when the current model is text-only (e.g. Codex API surfaces) or when /compact is given custom instructions. Frames survive context rebuilds and later compactions (budget eviction is middle-out: the session-head frame is pinned); the expanded compaction message notes the attached frame count
Added a persistent subagent lifecycle: finished subagents stay live as idle, are parked to disk after task.agentIdleTtlMs (default 7 minutes; 0 keeps them live until exit), and are revived automatically when messaged or prompted from the Agent Hub
Added the history:// protocol: history:// lists every registered agent and history://<agentId> renders a concise markdown transcript (tool calls collapsed to one line each, thinking elided) for live and parked agents alike
Added an IRC mailbox bus with bounded per-agent inboxes: irc wait blocks until a matching message arrives, inbox drains or peeks pending messages, and sending to an idle or parked agent wakes or revives it for a real turn
Added a dedicated TUI renderer for the irc tool: directional send/receive headers with delivery-outcome coloring, quoted message bodies with expand-aware truncation, per-recipient receipt trees for broadcasts and failures, and status-badged peer listings with unread counts
Added the task.batch setting (default on): the task tool's batch shape { agent, context, tasks[] } spawns one subagent per item — each its own independent background job with the normal idle/parked lifecycle and optional per-item isolation — and prepends the required shared context to every spawned subagent's system prompt; disabling it restores the flat single-spawn schema

Changed

Changed task-tool sync execution to fan out multiple tasks[] items in parallel and return a merged result payload when no async job manager is available
Changed the compaction UX so the conversation no longer visually restarts: the TUI renders the full-history display transcript (buildSessionContext({ transcript: true })), with each compaction shown as a slim inline divider — ── 📷 compacted · ctrl+o ── — at the point it fired; expanding (ctrl+o) reveals the summary and snapcompact frame count. Applies to live compaction, /compact, /tree navigation, and session resume
Changed async.enabled to gate async bash commands only — the task tool now runs asynchronously regardless of the setting
Changed irc.timeoutMs to be the default timeout for irc wait and send with await: true
Moved the grouped path-tree helpers (buildPathTree, walkPathTree, find's grouped output formatter — now formatGroupedPaths) to @oh-my-pi/pi-utils so compaction summaries can render file lists with the same prefix-folded tree as find/search; tools/find no longer exports formatFindGroupedOutput
Changed TTSR rule notifications to combine rules into one block: a multi-rule match renders name: description rows (collapsed view caps at 4 rules with a +N more hint, ctrl+o expands), and consecutive notifications merge into the previous block while it is still the live transcript tail

Removed

Removed the pre-initialization startup splash and input buffer, so commands typed during launch are no longer queued and are handled only after the interactive TUI initializes

Fixed

Fixed irc live message delivery so successfully handed-off messages are no longer enqueued as mailbox mail, so they do not inflate unread irc counts
Fixed irc send with await: true to wait for a fresh reply to the current call instead of consuming previously buffered messages
Fixed main-session chat output to stop duplicating outbound irc sends from the main agent as relay cards
Fixed task-tool runtime compatibility so legacy flat task calls (agent, assignment) still execute under task.batch even though the wire schema is batch-first
Fixed the job tool's TUI preview leaking the model-facing <task-result> envelope for settled task jobs — the preview now shows the inner output body, and pretty-printed JSON bodies are flattened onto one line instead of previewing a lone {
Fixed npm CLI distribution bundles by embedding the stats dashboard client bundle so dashboard assets are served in prebuilt installs
Fixed the resolve tool's result block turning white after the leading icon: the accent-styled symbol embedded a foreground reset inside the inverse-rendered line, dropping the block color for the rest of the row
Fixed the CLI smoke-test command to start the stats server and verify dashboard HTML is served, catching bundled-asset regressions
Added verification of a <div id="root"></div> and index.js in smoke-test dashboard responses
Restored the checkmark glyph on ask-tool custom answers and the multi-select "Done selecting" option, which a status-glyph sweep had swapped for the ask tool icon
Fixed the thinking.autoPending statusbar indicator using question-mark glyphs (▣?, nf-md-help_box, [?]) in every symbol preset, which made the auto-thinking pending state indistinguishable from a terminal missing-glyph fallback. Replaced with clear loading indicators (⟳, fa-circle-o-notch, [~]) (#2267).
Fixed tab.screenshot({ save }) ignoring the save path's extension: an explicit .webp/.jpg destination received hardcoded PNG bytes behind a mismatched name. The full-res capture format is now derived from the save path (png/jpeg/webp, puppeteer-native), and the reported mime type follows the bytes actually written; unknown or missing extensions still capture PNG
Fixed an infinite compaction.strategy: shake auto-continue loop in thinking-heavy sessions: the post-shake check now uses the provider-anchored trigger metric (instead of a local estimate that undercounts thinkingSignature payloads) and only treats pressure as resolved when residual context lands inside an 80% recovery band, so shake reliably falls back to context-full compaction when it cannot create real headroom (#2275).

[15.10.12] - 2026-06-10

Added

Added RPC subagent subscription frames, snapshots, and transcript catch-up APIs for desktop clients embedding omp --mode rpc.
Added opt-in shellMinimizer.sourceOutlineLevel and shellMinimizer.legacyFilters settings so shell minimization can tune source outlining and selectively fall back to conservative legacy routing.
Added repeatable --config <path> CLI overlays for temporary config.yml-style settings without editing the persistent global config (#1733).
Added python.interpreter to pin eval's Python backend to an explicit interpreter and skip automatic runtime discovery (#1802).
Added !command resolution for models.yml provider apiKey values and provider/model headers (#1888).
Documented the oMLX setup path through existing OpenAI-compatible local discovery (#1957).
Added TITLE_SYSTEM.md discovery so users can override the automatic session-title generation prompt for online and local tiny title models without patching installed prompt files. The override is re-discovered when the session working directory changes via /cwd.
Added a structured memory runtime surface for extensions and UI integrations to query backend status, search memories, and save explicit memories across the configured memory backend.
Added support for Git repositories using the reftable storage format by detecting extensions.refStorage = reftable in the repository configuration and falling back to shelling out to Git commands (git symbolic-ref, git rev-parse) for reference and HEAD resolution.
Added /setup providers (also available as /setup or /providers) to reopen the interactive provider setup scene from an active TUI session, letting users sign in and choose a web search provider without rerunning the full onboarding flow.

Changed

Bash execution now preserves minimized shell output inline while saving the untouched capture as an artifact://… footer when shell minimization rewrites a command's output.
Task tool live progress now renders finished subagents first and keeps unfinished (pending/running) ones pinned at the bottom of the list.
OutputSink artifact files (~/.omp/agent/artifacts/<id>.<tool>.log) are unbounded by default again, so artifact://<id> references preserve the complete raw stream. The head + rolling-tail capping machinery from #2081 (with its [ARTIFACT TRUNCATED: …] close notice) remains available as an opt-in via artifactMaxBytes, and the head window now closes permanently on first overflow so later small chunks cannot be written out of order before the tail replay.

Fixed

Fixed Ollama chat turns using the :off thinking selector so requests explicitly send reasoning disablement instead of falling back to the provider default (#2239).
Fixed long-running sessions becoming sluggish because the status line recomputed context usage by walking the full message history on every refresh. Message-token totals are now cached incrementally, and status lines that do not render context segments skip context accounting entirely. (#2089)
Fixed extension-registered slash commands being allowed to shadow builtin command names in the ACP/RPC command list: both the TUI and getSessionSlashCommands() now filter against the shared builtin reserved-name registry.
Fixed ACP turns for extension slash commands finishing before nested pi.sendUserMessage() prompts ran: the turn now drains scheduled extension prompts and prompt-event handlers before reporting end_turn, and prompts queued on a closed or disposed session fail fast with an ACP_SESSION_CLOSED error instead of running against a dead session.
Fixed hide-secrets placeholders leaking into ephemeral side-channel turns (IRC replies, summary prompts): streamed text deltas are now deobfuscated before emission (withheld while a partial placeholder is still streaming) and the final assistant message is deobfuscated before reuse. Provider-context obfuscation moved into the agent loop's transformProviderContext hook so request telemetry also captures the redacted context (#2146).
Fixed a failed Settings.init() permanently poisoning subsequent initialization: the cached init promise is now cleared on error so a retry can succeed.
Fixed exiting plan mode without confirmation only when neither the default plan file nor slug-named local plan files contain draft content (#2024).
Fixed enabledModels being ignored by the ACP model picker (Zed and other ACP clients) — AgentSession.getAvailableModels() now applies the configured allow-list, so only the models listed in enabledModels appear in the UI. Also applies consistently to the RPC get_available_models endpoint and the /model slash command. Glob selectors (anthropic/*), provider-scoped fuzzy patterns, and substring selectors now resolve in this synchronous path too, using the same scope semantics as startup resolution instead of being skipped with a warning.
ACP sessions now skip the client permission gate for bash/edit/delete/move when the user explicitly opts into yolo approval mode (--yolo/--auto-approve or a configured tools.approvalMode: yolo) and the effective per-tool policy is "allow"; default-config sessions keep the gate (#2097 by @Mokto)
Fixed hidden thinking blocks leaving placeholder Thinking... lines in the transcript (#2068).
Fixed Hindsight per-project-tagged scoping siloing retains/recalls per linked git worktree: projectLabel() now resolves the primary checkout root (or shared bare-repo common dir) via the new sync git.repo.primaryRootSync helper, so every worktree of one repo shares the same project:<name> tag and per-project bank id (#2232).
Fixed MCP OAuth flows accepting pasted redirect URLs or authorization codes through /login in headless environments (#2122).
Forwarded model ids through ModelRegistry API-key resolvers and Antigravity usage-limit rotation so pi-ai can apply model-family-scoped OAuth quota backoff instead of treating all google-antigravity counters as credential-wide. (#2198)
Fixed bare omp extensions being treated as a chat prompt instead of returning an actionable plugin-command error (#2089).
Fixed hide-secrets redaction so configured secrets are scrubbed from provider-facing system prompts, tool definitions, developer/system-reminder messages, and assistant tool-call arguments before model requests (#2146).
Fixed subagents looping indefinitely on byte-identical no-op edit calls. The hashline executor previously surfaced a soft "your body row(s) are byte-identical to the file" hint that some models ignored; one captured session emitted 182 such repeats in 205 calls over 16 minutes before the user aborted. A new per-ToolSession noopLoopGuard now tracks consecutive identical no-op payloads per canonical path and escalates to a thrown ToolError after NOOP_HARD_LIMIT (3) repeats, so the agent loop sees a tool failure and breaks the cycle (#2081).
Fixed the bash result renderer recomputing styled output (split / replaceTabs / truncateToVisualLines) on every TUI repaint, which scaled with both transcript length and per-row output size. With a long captured session every keystroke walked hundreds of bash rows; the reporter on issue #2081 observed Ctrl+X/Ctrl+C feeling unresponsive because the main thread was pinned re-styling scrollback. The result renderer now caches its produced lines keyed by (width, previewLines, expanded, rawOutput, isPartial), mirroring the existing eval-renderer cache; invalidate() clears the cache as before. Hot-path repaints with unchanged inputs are now O(1) (#2081).
Fixed ACP available_commands_update to include extension-registered slash commands so clients like Zed surface them in the slash-command palette.
Fixed ACP cancel button leaving the session in a stuck state — a new prompt sent while a turn is still in-flight (e.g. immediately after pressing Stop in Zed before session/cancel is processed) now implicitly cancels the running turn and queues the new message, instead of throwing an error that blocks further interaction.
Fixed interactive !/!! shell shortcuts to run non-bash commands through the configured user shell, including interactive startup for zsh/fish aliases and functions (#1816).

[15.10.11] - 2026-06-10

Added

Added supportsReasoningParams, alwaysSendMaxTokens, strictResponsesPairing, and a recursive whenThinking overlay (alongside streamIdleTimeoutMs/supportsLongPromptCacheRetention/requiresToolResultId/replayUnsignedThinking) to the OpenAI/Anthropic compat schema so custom model entries can configure those provider-specific capabilities
Custom model thinking config now uses the catalog's explicit vocabulary: efforts (ordered list) plus optional defaultLevel, effortMap, and supportsDisplay overrides; the legacy minLevel/maxLevel/levels range shape is still accepted and normalized at parse time. Wire facts (effortMap/supportsDisplay) are backfilled from model identity when not set, so existing claude-proxy configs keep the 5-tier adaptive scale and summarized display without changes.
New omp usage command: a detailed per-account breakdown of provider usage limits (bars, windows, reset times, plan metadata) covering every stored credential — accounts with no usage endpoint are listed as "no usage data" rows. Each provider section ends with per-window capacity stats ("capacity: 5h → 2.40/5 accounts used (2.60× quota left)"). Flags: --provider to filter, --json for the broker-shaped report payload, and --redact to mask account emails/ids down to a two-char anchor plus a minimal middle-out differentiator (ca*9*) for screenshot-safe sharing.
Startup hangs are now self-diagnosing (speculative fix for the "zero output, hangs even on omp -h" report class): a watchdog prints a stderr line every 10s naming the deepest in-flight startup phase (via logger.openSpanPath()) until a mode runner takes over, pausing around legitimate interactive waits (fork/move prompts, the --resume session picker); PI_DEBUG_STARTUP is restored as streaming synchronous [startup] phase markers covering command-module imports and the native addon load, which the post-startup PI_TIMING tree structurally cannot show for a hang; and waiting on piped-stdin EOF announces itself after 1s instead of blocking silently.
npm installs now execute a prebundled single-file entry: the published bin.omp points at dist/cli.js (built by scripts/bundle-dist.ts during prepack, ~18MB minified, natives/transformers/mupdf external), cutting npm-install cold start by roughly 3x versus transpiling the raw TypeScript graph per launch; src/** stays published for SDK consumers and worker fallbacks. The on-repo manifest keeps bin.omp at src/cli.ts — release rewrites it via the publishBin override in scripts/ci-release-publish.ts — so source installs (bun link, install.sh --source) keep working without a build step
Plain interactive TTY launches render the full welcome box (logo held on the intro's first frame, model, tips, LSP servers, recent-sessions loading placeholder) before session construction, clearing the screen so the TUI's first paint replaces it in place; the welcome box now reserves fixed slot counts (4 recent sessions, 4 LSP servers) so its height no longer shifts between the splash, loading, and loaded states. First-run launches keep the dim two-line splash (omp <version> / Initializing session…); resume/fork/continue flows, quiet mode, PI_TIMING, and non-TTY stdio still skip it
Added /stats to launch the local stats dashboard from an active session, syncing session files first and opening the same browser dashboard as omp stats.
/settings now supports type-to-search filtering on setting labels, paths, descriptions, and values; Escape clears an active search before closing the panel.
Added a read-only view op to the todo tool that echoes the current list without mutating state, so the agent can recover exact task text instead of guessing it from memory.

Changed

Centralized model-identity logic in the new @oh-my-pi/pi-catalog package: config/model-equivalence.ts, config/model-id-affixes.ts, and config/model-provider-priority.ts were removed in favor of @oh-my-pi/pi-catalog/identity, and the registry's proxy-reference lookup now shares the catalog's single lazily-built bundled-model walk (@oh-my-pi/pi-catalog/identity bundled accessors) with the canonical-equivalence index instead of walking the ~12K bundled models twice into duplicate maps
Split the configured/implicit provider discovery protocols (Ollama, llama.cpp, LM Studio, openai-models-list, proxy) out of config/model-registry.ts into config/model-discovery.ts; the registry keeps orchestration (caching, status tracking, merging) while the protocol clients take an injected fetch/auth context
Catalog values (bundled models, modelsAreEqual, clampThinkingLevelForModel, getSupportedEfforts, DEFAULT_MODEL_PER_PROVIDER, Gemini/Antigravity wire headers) are now imported from @oh-my-pi/pi-catalog/<module> instead of the @oh-my-pi/pi-ai barrel, which no longer re-exports them; the resolver's defaultModelPerProvider alias was removed and its duplicated default-model fallback / scoped-model dedupe blocks were factored into pickDefaultAvailableModel and a shared addScopedModel helper
Cached custom model alias maps and built them lazily on first custom model reference lookup, avoiding unnecessary startup model-registry initialization
Cached resolved auth broker configuration and snapshot reads for the process lifetime so repeated startup paths reuse the same OMP_AUTH_BROKER_* resolution instead of re-running config/token discovery
Reused task-agent discovery results for repeated TaskTool.create calls in the same working directory to avoid repeated plugin scans during subagent startup
SSH tool creation now formats host descriptions from synchronous host-info cache reads (memory hit or cached JSON) instead of per-host async reads — hosts without cached info render the existing placeholder; warm-cache descriptions are byte-identical
Deferred heavy dependencies off the startup import graph to first feature use: linkedom (web fetch feed parsing and scrapers), puppeteer-core/@puppeteer/browsers (browser launch), @mozilla/readability (page extraction), @xterm/headless (interactive bash PTY), @babel/parser (JS eval import rewriting), and the mnemopi memory engine (backend/state construction)
Renamed the PI_TIMING startup phase discoverModels to discoverAuthStorage — the timer only ever wrapped auth storage discovery
Worker threads (stats sync, browser tab, JS eval) and the tiny-model subprocess now re-enter the CLI entrypoint with hidden argv selectors (__omp_*, --tiny-worker) via the declared worker-host entry (workerHostEntry()), collapsing the per-distribution spawn branches; outside a CLI host (bun test, SDK embedding) spawn sites fall back to loading the worker module directly, and both binary build scripts dropped their per-worker --compile entrypoint lists
The CLI entry no longer top-level-awaits runCli — the floating call reports rejections to stderr and exits 1, keeping the entry module CJS-lowerable and the bundle parse-friendly
Tightened the system prompt and tool prompts: deduped restated warnings (bash "catch yourself" list, search/find shell-fallback recaps, read instruction/critical overlap, the AST metavariable primer duplicated across both ast tool descriptions), factored the repeated repo-default clause in the gh search ops, dropped a dead rsed reference and an internal tool-timeouts.ts pointer, and pruned internal mechanism the agent can't act on (screenshot temp-file/downscaling pipeline, browser spawn lifecycle, gh "replaces former op" history and run-watch grace period, output-minimizer heuristics, BM25 ranking name, task.maxConcurrency pointer)
Extended the prompt-efficiency pass to the full prompt surface (subagent/plan-mode/notice/title/commit system prompts, agent definitions, goals, memories, review and autoresearch prompts): RFC-keyed prescriptive prose, fixed garbled grammar and a stale <PLAN_TITLE> placeholder in the plan-approval reminder, deduped intra-file restatements, and corrected the todo op table's claim that rm requires a task/phase (bare rm clears the whole list)
Replace tool prompt no longer recommends sed -i/cat-heredoc commands that the bash interceptor blocks; its bash-alternatives table now only lists non-intercepted commands
Capped concurrent IRC cards in the transcript's live region at 4: cards landing below a still-running tool cannot commit to native scrollback, so an unbounded burst pushed the live block's uncommitted rows above the window top (content read as cut off until the cards expired). The oldest live-region card now retires as soon as a new one would exceed the cap.
Interactive PTY mode (pty: true) no longer injects the non-interactive environment (TERM=dumb, GIT_EDITOR=true, PAGER=cat, NO_COLOR=1) that defeated its purpose — the PTY child now gets a real TERM=xterm-256color; and when a PTY is requested but unavailable (headless/RPC), the result now carries an explicit downgrade notice instead of silently running through a dumb pipe.
Raw sqlite ?q= queries are now capped at 1000 rows with an "add a LIMIT clause" notice — statement.all() on a multi-million-row table previously materialized every row, blocking the process for minutes.
Plain-file range reads no longer scan to EOF on files over 4MB just to count total lines (the count is reported as approximate), and multi-range reads slice from a single pass instead of re-streaming the file once per range.
gh run_watch now polls adaptively (3s for the first minute, then 15s), survives rate-limit errors with backoff instead of dying and discarding accumulated context, reuses job data for completed runs, and gives up with a clear message after ~90s when a commit has no workflow runs at all (previously an infinite 3-second poll loop).
The legacy patch-mode fuzzy matcher pre-normalizes file and pattern lines once per seek with a Levenshtein lower-bound bail, replacing the per-position re-normalization that made a single mismatched hunk against a 10k-line file cost multi-second synchronous stalls; the streaming hashline preview also caches file text and tree-sitter block resolution across ticks instead of re-reading and re-parsing every target file per streamed chunk.
The DAP client reader now uses chunk-list buffering and the output buffer is a chunk deque with a running byte count — debugging a chatty program previously cost O(n²) Buffer.concat per chunk plus whole-buffer byte-scans per 1KB trim, freezing the session.
GitHub caching: the per-lookup auth key is memoized against hosts.yml mtime (was a blocking readFileSync on every issue:///pr:// read including cache hits), background refreshes are deduped by row identity, and PR diffs are stored once per row instead of twice (unified + rendered copies).
Task progress snapshots shallow-copy per-agent progress instead of structuredClone-ing nested tool payloads (up to 500KB) on every progress event; streaming assistant-message reveal caches per-block grapheme counts and skips the markdown render LRU for in-flight partials, eliminating 2-3 full Intl.Segmenter walks per 33ms tick and tens of MB of retained stale partial snapshots on long replies.
Python eval cells: the availability probe is cached per cwd (was two interpreter spawns per cell even with a hot kernel), and stdout frames coalesce per write instead of one locked+flushed JSON frame each.
Multi-entry edits now stop at the first failing entry and report exactly which entries were applied and which were not — continuing after a failure applied later entries authored against line numbers that assumed the failed entry succeeded, and a retry of the whole batch then double-applied the survivors.
Decomposed config/model-registry.ts further: model roles (MODEL_ROLES, getRoleInfo, getKnownRoleIds) moved to config/model-roles.ts, the models.json config handle and provider validation moved to config/models-config.ts, the two provider+id merge scaffolds collapsed into one mergeByModelKey helper, the four ~15-field override/overlay enumerations now share a ModelPatch type applied by a single applyModelPatch(base, patch, transport) core (the merge vs replace transport policies preserve the same-id custom-definition replacement semantics), and canonical-variant selection delegates to @oh-my-pi/pi-catalog/identity's new resolveCanonicalVariant
Resolver cleanup: five duplicated trailing-:level suffix parses collapsed into splitThinkingSuffix, the matching engine is now the documented matchModel core with the selector grammar and entry points layered on top, and resolveCliModel's hand-rolled decomposed provider/id lookup reuses findExactModelReferenceMatch; runtime discovery tests split out of test/model-registry.test.ts into test/model-discovery.test.ts
TranscriptContainer assembles the transcript incrementally: each block's render is reference-compared and its stripped contribution, separator, and row placement are reused when unchanged, with the persistent row array truncated and re-pushed only from the first divergent block; the leading byte-identical row count is reported to the renderer through pi-tui's new RenderStablePrefix seam so off-screen transcript rows are no longer re-rendered, re-prepared, or re-audited every frame. Block components became reference-stable to make this effective: UserMessageComponent memoizes its OSC 133 zone wrapping, WelcomeComponent and DynamicBorder cache their renders, and dashboards copy before padding (render results are readonly under the new pi-tui contract)
A live block whose trailing row grows in place as a visible prefix (token streaming into the cursor line) is now commit-safe through its full body instead of being held back by the volatile-tail margin — the growing row itself is the block's last and can never commit while it remains last, so a streaming reply's scrolled-off head reaches native scrollback (tmux pane history) mid-stream
Rewrote the bash tool's coreutils guidance (tool prompt and system prompt) around an explicit litmus: pipelines that compute a new fact (wc -l, sort | uniq -c, comm, diff) are legitimate bash, while commands that merely move, page, or trim bytes a dedicated tool can fetch remain banned — output trimming destroys data the artifact:// capture would have saved.
Default API auto-retries now use 10 attempts with a 500ms Anthropic-style exponential backoff capped at 8s with jitter, so transient 502/gateway failures get a longer retry budget without multi-minute local sleeps.

Fixed

Fixed ask question/result renders so option and answer rows are no longer duplicated when the component is re-rendered
Fixed streaming write/diff previews to keep line-number gutter widths stable while content grows, preventing already-rendered preview rows from being reflowed mid-stream
Fixed the welcome screen showing "No LSP servers" when lsp.lazy is enabled: recognized servers are now still discovered at startup and listed with a dim "available" dot (no warmup), and /status reports them as available instead of omitting the section
Fixed edit-tool diffs stacking adjacent ... markers around inserted block-context rows (each row added its own gap markers from a snapshot of the diff, so neighboring insertions doubled them, and a marker could be left stranded between contiguous lines): non-contiguous regions are now separated by a single blank row, normalized after insertion, and rendered as one dim … in the TUI and HTML export
Fixed an uncaught questions.map is not a function TUI crash in the ask tool's call renderer when a model double-encoded the questions array as a JSON string (a bare string passes a truthy .length check but has no .map): the renderer now normalizes untrusted call args — parsing double-encoded questions, dropping malformed entries/options, and falling back to the "No question provided" frame instead of throwing
Fixed direct modelRoles consumers so comma-separated fallback chains are split before model parsing, preserving explicit thinking selectors instead of treating the comma tail as an invalid suffix.
Fixed model-provider detection for append-only mode, authoritative Vertex endpoint checks, and upstream-routing selection by switching from URL substring checks to catalog host-matching helpers
Fixed pasting into the ask tool's "Other (type your own)" text box (and hook input/editor dialogs) on terminals with OSC 5522 enhanced paste (kitty protocol): the enhanced-paste focus routing only targets components exposing a pasteText hook, and the dialog wrappers had none, so the payload was stuffed into the main prompt editor hidden behind the dialog. HookEditorComponent and HookInputComponent now forward pasteText to their inner editor/input (pasting also resets the input dialog's timeout countdown like any keystroke).
Fixed auto-retry giving up after one attempt ("Provider requested Xms wait, exceeds retry.maxDelayMs") on a usage-limit 429 when every sibling account was only momentarily blocked: the retry delay now waits for the earliest sibling unblock when that comes sooner than the provider's multi-hour retry-after, so the next attempt picks up the recovered account instead of failing fast.
Fixed Hindsight per-project-tagged mental-model seeding so each project gets its own conventions/decisions models and session context only injects active-project or untagged models (#2218).
Fixed Windows stdio MCP .cmd commands by wrapping batch shims with cmd.exe /d /s /c using the outer command quotes required by cmd /s, while preserving literal % and quoted JSON arguments for Codegraph MCP (#2220).
Fixed the bundled explore agent's thinking-level: med frontmatter — not a valid effort (minimal/low/medium/high/xhigh), so it silently parsed to undefined and the agent ran without its intended thinking level
Discovery context-file reads (~/.claude, ~/.cursor, project trees, @-imports) now stat-gate to regular files before reading: a FIFO/socket/char device dropped where a context file is expected previously blocked startup forever on a read that can never see EOF.
Fixed the read tool's provider-visible path schema and docs so web URLs and internal URI targets (omp://, issue://, pr://, etc.) are advertised alongside local files (#2215).
Kept IRC cards from being removed after their TTL once everything above them finalized: their rows may already be committed to native scrollback, and removing them was an interior deletion of the committed prefix that the engine could only repair by recommitting everything below the gap (duplicated blocks). Such cards now stay in the transcript as durable history.
Fixed the recommit storm that sprayed stale snapshots of a running task's progress tree into native scrollback. The stable-prefix ratchet promoted any row quiet for one 30-frame window, so slowly ticking rows (per-agent tool/cost counters updating every few seconds) were repeatedly promoted, committed, rewritten, and recommitted by the engine audit for the whole run. The ratchet now floors itself permanently at the first row that mutates after being promoted — settled heads (a task's prompt/context) still reach scrollback, genuine tickers never re-promote.
Fixed the artifact spill dropping the first ~20KB of output: head-retained bytes were never written to the artifact file, so for every bash/eval/ssh command exceeding the 50KB spill threshold, the artifact:// advertised as the "full capture" was permanently missing its head — the agent re-reading it got truncated data presented as lossless.
Fixed streaming-output chunk throttling dropping chunks instead of coalescing them: streaming previews and the auto-background "output so far" text the model reasons over contained output with arbitrary middles silently spliced out.
Fixed vault:// writes bypassing both the approval ladder and plan mode: internal-URL writes were uniformly rated tier read (auto-allowed even in always-ask) and the internal-router branch returned before the plan-mode guard, so the model could silently overwrite real Obsidian notes; writes through schemes with a mutating handler are now tier write and plan-mode-enforced.
Fixed writes into .tar.gz/.tgz archives silently stripping gzip compression (the rewritten archive was a bare tar under the .gz name — masked on re-read because Bun auto-detects, broken for tar xzf/CI consumers), and made archive rewrites atomic via temp-file + rename so a crash mid-write can no longer destroy every other member; symlinked archive paths resolve to their target before the swap so the rename writes through instead of replacing the link with a regular file.
Fixed merge-conflict detection being completely inert on CRLF files: the scanner split on \n and compared === "=======", so =======\r never matched and the agent edited around live conflict markers without warning; CRLF files now detect, splice, and round-trip their line endings correctly.
Fixed cross-line search (\n in the pattern) silently returning zero matches: the native searcher was never switched to multi-line mode (only the regex flag was set), so the advertised feature matched nothing on real files while reporting a confident "No matches found".
Fixed search results lying about completeness: one hot file could consume the entire 2000-match global budget in path order making later files unreachable by any skip (now capped per file with the footer hedging of N+ when truncated), paginating past the last page returned "No matches found" instead of "No more results", directory scans now report how many >4MB files were skipped instead of silently excluding them, adjacent matches in virtual resources no longer emit duplicated backwards-numbered context lines, and patterns are no longer trim()ed (only all-whitespace is rejected — leading/trailing whitespace is meaningful regex).
Fixed the search tool's native grep being uncancellable: neither the abort signal nor any timeout was threaded through, so Esc on a huge-tree search left the native walk burning CPU to completion; both now propagate (30s default timeout).
Fixed archive and sqlite reads that could OOM or hang the process: tar/tgz archives are stat-gated at 256MB before being loaded, zip members reject attacker-declared uncompressed sizes over 64MB before allocation, and binary plain files now return a NUL-sniff notice instead of filling the line budget with mojibake.
Fixed malformed internal-URL selectors (artifact://3:-100) silently dumping the whole resource instead of erroring, selectors directly on an archive root (a.zip:500, a.zip:raw) being misparsed as member names, archive members minting editable hashline tags keyed to the archive path (they are immutable resources), URL selector tokens being case-sensitive (:RAW 404ed), artifact://N resolving into another session's artifacts in multi-session hosts, and not-found paths with archive/sqlite extensions stacking multiple 5s workspace-wide suffix globs (now shared per read, with glob metachars escaped so foo[1].ts can match itself).
Fixed leading cd X && extraction breaking shell-expanded paths — cd "$(git rev-parse --show-toplevel)" && make failed with "Working directory does not exist" because the captured path was resolved literally; extraction now defers to the shell when the path contains $, backticks, or (.
Fixed the echo/printf write-redirect interceptor rule blocking legitimate commands containing > inside quotes (echo "a -> b", printf 'use 2>&1'); the rule is now quote-aware, and also catches >| clobber redirects and $VAR targets it previously missed.
Fixed every completed auto-backgrounded bash invocation leaking its persistent native Shell in the process-global session map, and the running-job cap failing all bash commands outright — at capacity, commands now degrade to direct foreground execution (explicit async: true still errors).
Fixed a duplicate-delivery race where a bash job completing just inside the auto-background threshold could be returned as the tool result and re-injected as a completion notification, and fixed auto-background silently preempting the ACP client-terminal route when an editor advertises terminal capability.
Fixed timed-out/cancelled PTY and client-bridge commands surfacing raw output with no annotation (the model couldn't distinguish timeout from failure and retried identically); the timeout/abort notice is now always appended.
Fixed ask reporting timeout auto-selection as "User selected: X" — fabricated consent for consequential questions; the result now says "(auto-selected after timeout)" with a timedOut detail flag, the transcript card marks the auto-selection distinctly, and a deliberate Esc seconds past the deadline is treated as a cancel instead of being reclassified as a timeout.
Fixed todo accepting duplicate task content/phase names in init (duplicates were permanently unaddressable — every targeting op hit the first match while auto-promotion kept resurrecting the twin) and persisting half-applied batches on error; failed batches no longer mutate state.
Fixed the auto-generated-file guard caching markers by path alone with no invalidation — a file regenerated after first check stayed editable (and vice versa); entries are now validated against mtime+size.
Fixed editor-bridged (ACP) writes skipping the post-write bookkeeping the direct path performs (bumpFileMutationVersion, shebang chmod), so mutation-version consumers saw stale state depending on whether an editor was attached.
Fixed conflict://* resolution failing spuriously when an out-of-band edit shifted a conflict block (stale duplicate registrations are now tolerated as already-resolved — but a DISTINCT conflict block that is merely byte-identical and still present in the file stays addressable), and partial conflict-resolution failures now set isError instead of burying failed files mid-text in a success result.
Fixed patch-mode prefix/substring matches silently truncating line content the model never saw: every non-exact match strategy now emits a warning with strategy + similarity, and prefix/substring matches are rejected unless the discarded fragment survives in the replacement lines.
Fixed ast-edit and file-mention snapshots being recorded under non-canonical paths (invisible to stale-tag recovery under symlinked cwds), and the ast-edit apply step leaving every just-issued preview tag stale — post-apply snapshots are re-recorded and fresh tags surfaced in the result.
Fixed notebook cells containing literal # %% [markdown] marker text being silently split into extra cells on any edit; marker-shaped source lines are now escaped on render and restored on parse.
Fixed the LSP client being published before initialize completed (concurrent callers hit "server not initialized" flakes on first use), reader-loop death leaving a permanent zombie client where every request times out at 30s forever (bad messages are now isolated per-message and a dead reader tears the client down for respawn), framing stalls on header blocks without Content-Length (now resynced past the junk in both LSP and DAP), lsp status hardcoding ready for every client including wedged ones, and shutdown skipping clients still mid-initialize (their server processes outlived exit).
Fixed numeric LSP code-action selectors being shadowed by substring title matches — query: "2" could apply a different quickfix whose title contained "2"; numeric queries now select strictly by index.
Fixed file:// URIs built without percent-encoding: a % in a path threw URIError on round-trip and a # truncated the server-side path, desynchronizing diagnostics and workspace edits; URIs from lax servers carrying a raw #/? now route to the lenient parser instead of parsing "successfully" as fragment/query and misrouting edits.
Fixed multiple LSP inserts at the same position applying in reverse of spec order (transposed import/reference insertions), and applyWorkspaceEdit now overlap-validates every file before writing any, so a conflicting rename no longer leaves the workspace half-renamed.
Fixed lsp reload hanging for the whole tool timeout (didChangeConfiguration was sent as a request; it is a notification), biome failures being silently reported as "no diagnostics", a hung language server adding up to 30s to every edit (writethrough init is now deadline-bounded at 5s with deterministic spawn failures negative-cached for 3 minutes), and DAP pause() burning its full timeout when the stopped event raced the subscription; concurrent DAP breakpoint mutations are also serialized per session (last-writer no longer silently drops the other's breakpoints), queued mutations honor the caller's abort at dequeue, and the DAP output buffer retains a full 128KB tail instead of dropping whole chunks below the cap.
Fixed concurrent isolated background tasks interleaving git stash push/pop + cherry-pick on the shared repository — the merge sequence now runs under the repo lock, eliminating a lost-uncommitted-changes race; a stash-pop failure after successful cherry-picks also no longer reports merged branches as "unmerged" (the duplicate-commit trap) and instead tells the user to pop the stash manually.
Fixed async task batches getting stuck "running" forever (unscheduled/failed-to-register tasks never counted toward completion), error-result jobs being marked completed, semaphore-queued tasks counting against the 15-job global cap (batches >15 dropped the remainder and starved other async work), duplicate task ids skipping validation on the async path, and an abort racing subagent session startup leaking the late-created session's LSP/MCP processes.
Fixed task fail-fast abandoning in-flight siblings uncancelled (the worker signal now propagates), patch-mode merges blocking ALL successful siblings' patches when one task failed, and $@ command expansion interpreting $-replacement patterns in user input.
Fixed eval cells double-writing artifacts (the tool and the per-cell executor each opened a sink on the same artifact path, corrupting >50KB outputs), JS parallel() early-rejecting in violation of its documented barrier (orphaning in-flight agent() thunks with worker-side promises hung forever), Python child subprocesses inheriting the NDJSON frame pipe (their stdout was dropped and could corrupt protocol frames — it is now captured and forwarded), JS cell timeouts silently wiping persistent VM state without annotation, and the JS console bridge throwing on console.dir/time/group/assert/trace.
Fixed pr_push never invalidating the PR/diff cache (the canonical push-then-verify flow read a pre-push diff for up to 5 minutes), current-branch gh pr merge/close with no positional never invalidating at all (exactly the staleness the cache layer claims to eliminate; numeric flag values like --milestone 3 also no longer steal the positional), multi-PR checkouts discarding successful checkouts and racing in-flight git mutations on first failure (allSettled with per-PR reporting), run-watch ending with a failure result and zero logs when an auto-retry raced the grace-period refetch, the per-watch completed-run job cache serving a rerun's FIRST-attempt jobs after the rerun completed (entries are evicted whenever a run is observed non-completed), pagination terminating on post-filter page length, millisecond precision leaking into GitHub search date qualifiers, leading-dash PR identifiers reaching gh as flags, and issue://?state= typos silently coercing to the open list.
Fixed the Exa API key being written to the log file on every failed MCP request (the key rode the query string of logged URLs; key/token/secret/auth params are now redacted), and removed the web-search query rewrite that replaced every 202x substring with the current year — it corrupted CVE identifiers and made historical-year searches silently impossible.
Fixed reopening the sole browser tab with a different dialogs policy disposing Chromium and then using the dead handle, a stale tab release evicting a live replacement browser from the registry (spawning duplicate Chromium processes), and concurrent same-name open calls leaking a worker + refcount via a check-then-set race (acquisitions are now single-flight per name); queued opens honor an abort at dequeue, and an init-payload failure releases the temporary browser hold instead of pinning the refcount forever.
Fixed fetch decoding every response as UTF-8 regardless of declared charset (Shift_JIS/EUC-KR/GBK pages rendered as mojibake through the whole reader pipeline; Content-Type and <meta charset> are now honored via TextDecoder), binary URLs being downloaded twice (body skipped on the first pass for convertible types), >50MB truncation being silent (now flagged in notes), all transport error detail being swallowed into a bare "Failed to fetch URL" (the cause is surfaced and 429s get one Retry-After-honoring, abort-aware retry), MCP SSE keep-alive lines escaping as raw SyntaxErrors, MCP calls having no default timeout (now 60s), and a YouTube fetch budget expiry being misreported as a user abort that also skipped temp-file cleanup.
Fixed archive directory listings silently ignoring the selector offset — a.zip:dir:50 now starts the listing at the 50th entry instead of relisting from the top.
Fixed the model selector dropping an immediate Enter when cached models were available but the selector's offline refresh was still pending.
Fixed dynamic import(...) inside functions passed to the browser tool's tab.evaluate/page.evaluate failing with __omp_import__ is not defined. The eval/browser JS runtime rewrites dynamic-import callees to the worker-injected __omp_import__ helper, but puppeteer serializes evaluate callbacks with Function.prototype.toString() and re-runs them inside the page, where the helper does not exist. The rewriter now substitutes a guarded shim that falls back to native dynamic import when the helper is absent, so serialized code works in the page realm while in-worker imports keep resolving against the session cwd.
Transcript block freezing is now unconditional instead of gated on ED3-risk terminal detection: every finalized block replays its frozen snapshot once it crosses out of the live region, on all terminals including Windows, because the rewritten renderer's committed scrollback is immutable everywhere. Still-mutating blocks (pending tools, streaming messages, async thinking renderers) anchor the live region and keep repainting until they finalize, which structurally fixes stale/duplicated output from late async expansions (#1823).
Fixed the edit tool's post-edit diff preview occasionally echoing a context line twice with out-of-order numbering. Block-boundary context injection classified space-prefixed diff rows as old-file-only, so an unchanged line sitting in a net-offset region (old N / new N+k) was missing from the new file's visibility window; findBlockContextLines then re-surfaced it under its post-edit number and the row was spliced in after the adjacent change run. New-file boundary lines are now translated back to pre-edit numbers (the compact-preview renumbering contract) and merged into a single old-numbered insertion pass — also fixing closers below a net-offset edit being dropped or renumbered incorrectly.
Fixed the Anthropic web-search provider claiming the Claude Code identity on API-key requests: the CC billing header + system instruction were injected whenever the model wasn't Haiku 3.5, regardless of auth mode. Injection is now OAuth-gated like the streaming path, and OAuth search requests patch the billing header's cch attestation (via wrapFetchForCch) instead of shipping the cch=00000 placeholder.
Fixed long streamed content appearing cut off mid-run: scrolled-off rows were erased from the viewport without ever being appended to terminal history. The transcript's commit boundary (deriveLiveCommitState) was all-or-nothing per block — one perpetually rewriting row (a task tool's ticking progress tree, per-agent cost/tool counters, spinner stats) suspended scrollback commits for the entire block, so once the block outgrew the viewport its static head (e.g. a task's prompt/context markdown) was neither committed nor on screen until the tool sealed, and was lost outright if the session ended mid-run. A stable-prefix ratchet now promotes leading rows that stayed visibly identical for a full 30-frame window as commit-safe, so the settled head reaches native scrollback while only the genuinely volatile tail stays deferred; a rewrite above the promoted run retreats the boundary and the engine audit recommits (duplication, never loss).
Fixed local tiny-title worker stdout/stderr leaking raw native model output such as </title> and cache/status lines into the interactive TUI scrollback (#2206).
Fixed task-agent discovery advertising Claude Code custom agents from .claude/agents/*.md as OMP subagents; direct task-agent discovery now only loads OMP-native .omp agent roots, while Claude marketplace plugin agents keep their existing provider path (#2209).

Removed

Removed the clearOnShrink setting and its PI_CLEAR_ON_SHRINK environment variable: the rewritten renderer always clears shrunken rows exactly, so the flicker/perf tradeoff the setting controlled no longer exists. Existing config entries are ignored.
Removed the prompt-submit native-scrollback reconciliation checkpoint and the eager streaming render mode from the interactive controllers — the renderer's append-only contract made both obsolete.

[15.10.10] - 2026-06-09

Added

Added a read-only view op to the todo tool that echoes the current list without mutating state, so the agent can recover exact task text instead of guessing it from memory.

Changed

Rewrote the bash tool's coreutils guidance (tool prompt and system prompt) around an explicit litmus: pipelines that compute a new fact (wc -l, sort | uniq -c, comm, diff) are legitimate bash, while commands that merely move, page, or trim bytes a dedicated tool can fetch remain banned — output trimming destroys data the artifact:// capture would have saved.

Fixed

Fixed the model selector dropping an immediate Enter when cached models were available but the selector's offline refresh was still pending.
Fixed dynamic import(...) inside functions passed to the browser tool's tab.evaluate/page.evaluate failing with __omp_import__ is not defined. The eval/browser JS runtime rewrites dynamic-import callees to the worker-injected __omp_import__ helper, but puppeteer serializes evaluate callbacks with Function.prototype.toString() and re-runs them inside the page, where the helper does not exist. The rewriter now substitutes a guarded shim that falls back to native dynamic import when the helper is absent, so serialized code works in the page realm while in-worker imports keep resolving against the session cwd.
Transcript block freezing is now unconditional instead of gated on ED3-risk terminal detection: every finalized block replays its frozen snapshot once it crosses out of the live region, on all terminals including Windows, because the rewritten renderer's committed scrollback is immutable everywhere. Still-mutating blocks (pending tools, streaming messages, async thinking renderers) anchor the live region and keep repainting until they finalize, which structurally fixes stale/duplicated output from late async expansions (#1823).
Fixed the edit tool's post-edit diff preview occasionally echoing a context line twice with out-of-order numbering. Block-boundary context injection classified space-prefixed diff rows as old-file-only, so an unchanged line sitting in a net-offset region (old N / new N+k) was missing from the new file's visibility window; findBlockContextLines then re-surfaced it under its post-edit number and the row was spliced in after the adjacent change run. New-file boundary lines are now translated back to pre-edit numbers (the compact-preview renumbering contract) and merged into a single old-numbered insertion pass — also fixing closers below a net-offset edit being dropped or renumbered incorrectly.
Fixed the Anthropic web-search provider claiming the Claude Code identity on API-key requests: the CC billing header + system instruction were injected whenever the model wasn't Haiku 3.5, regardless of auth mode. Injection is now OAuth-gated like the streaming path, and OAuth search requests patch the billing header's cch attestation (via wrapFetchForCch) instead of shipping the cch=00000 placeholder.
Fixed long streamed content appearing cut off mid-run: scrolled-off rows were erased from the viewport without ever being appended to terminal history. The transcript's commit boundary (deriveLiveCommitState) was all-or-nothing per block — one perpetually rewriting row (a task tool's ticking progress tree, per-agent cost/tool counters, spinner stats) suspended scrollback commits for the entire block, so once the block outgrew the viewport its static head (e.g. a task's prompt/context markdown) was neither committed nor on screen until the tool sealed, and was lost outright if the session ended mid-run. A stable-prefix ratchet now promotes leading rows that stayed visibly identical for a full 30-frame window as commit-safe, so the settled head reaches native scrollback while only the genuinely volatile tail stays deferred; a rewrite above the promoted run retreats the boundary and the engine audit recommits (duplication, never loss).
Fixed local tiny-title worker stdout/stderr leaking raw native model output such as </title> and cache/status lines into the interactive TUI scrollback (#2206).
Fixed task-agent discovery advertising Claude Code custom agents from .claude/agents/*.md as OMP subagents; direct task-agent discovery now only loads OMP-native .omp agent roots, while Claude marketplace plugin agents keep their existing provider path (#2209).

Removed

Removed the clearOnShrink setting and its PI_CLEAR_ON_SHRINK environment variable: the rewritten renderer always clears shrunken rows exactly, so the flicker/perf tradeoff the setting controlled no longer exists. Existing config entries are ignored.
Removed the prompt-submit native-scrollback reconciliation checkpoint and the eager streaming render mode from the interactive controllers — the renderer's append-only contract made both obsolete.

[15.10.9] - 2026-06-09

Fixed

Fixed streaming thinking (and other styled assistant content) vanishing from native scrollback once it scrolled past the viewport top during a foreground turn. The transcript's append-only commit detector compared raw row bytes, so a styled paragraph wrapping onto a new row (the span-closing SGR and width padding move while the visible cells stay identical) or a streamed token pushing the last word down a line flagged the block as permanently volatile — the commit boundary froze and every later row that crossed the viewport top was committed nowhere. Rows are now compared by visible content, a wrap-shrink of the in-flight bottom line counts as append-only, and a genuine one-off interior rewrite only suspends commits until the block re-earns append-only (30 clean frames), after which the pinned emitter backfills the stalled gap contiguously. Periodically rewriting blocks (spinners, collapsing tool previews) never re-earn and stay deferred.
Fixed bracketed pastes containing multiple image file paths so each image is attached in order instead of treating the whole paste as one unreadable path.
Fixed MCP OAuth fallback prompts so the "Click here to authorize" label emits an auth-safe terminal hyperlink even when hyperlink auto-detection is unavailable, keeping non-browser MCP setup usable (#2196).
Fixed task-spawned subagents repeating filesystem scans the parent had already completed. ExecutorOptions and the createAgentSession() call inside runSubprocess() did not forward rules, the discovered extension paths, or the discovered .omp/tools/ paths, so each subagent re-ran loadCapability<Rule>(), discoverAndLoadExtensions(), and the full .omp/tools/ walk. The toolsession now caches session.rules, session.extensionPaths, and session.customToolPaths; runSubprocess() threads them through; and createAgentSession() accepts new preloadedExtensionPaths and preloadedCustomToolPaths options backed by new exported discoverExtensionPaths() and discoverCustomToolPaths() helpers. Crucially, only path lists are forwarded — never loaded instances. Each session rebuilds its own Extension and LoadedCustomTool objects so the per-session ExtensionAPI/CustomToolAPI (cwd, eventBus, runtime, exec, pushPendingAction, UI) targets the right session; forwarding loaded instances would have routed extension handlers and custom-tool execution back through the parent. The CLI's preloadedExtensions short-circuit is preserved for same-process reuse and now shallow-clones the caller's extensions array so inline-extension augmentation (autoresearch + custom-tools wrapper) cannot bleed back into it (#2190).
Fixed SSH tool cancellation hanging behind OpenSSH ControlMaster streams that stayed open after an Esc/user interrupt (#2180).
Fixed Windows stdio MCP servers launched through PATH shims such as codegraph.cmd so bare commands like codegraph resolve via PATHEXT before spawn (#2174).
Fixed compiled-binary extensions failing to load @oh-my-pi/pi-* packages when bun --compile quietly dropped one of the extra entrypoints (observed on macOS arm64 release builds): the legacy-pi compat shim's package-root override branch returned the bunfs path without checking the target was present, so the rewrite emitted a file:// URL to a missing module and the #1216 fallback (scoped to the throwing getResolvedSpecifier path) never ran. Override targets are now validated against the on-disk filesystem at module init, missing entries are dropped, and resolution falls through to canonical lookup so Bun resolves the import from the extension's own node_modules (#2168).

[15.10.8] - 2026-06-09

Added

Added an optional fetch option to CustomToolContext so custom tools can use a caller-provided HTTP implementation
Added optional fetch overrides to ModelRegistry construction and MCP/web search/tool network calls, enabling callers to inject custom HTTP clients instead of relying on global fetch
Added a bash.enabled setting to disable the model-facing bash tool while leaving user-initiated bang/RPC bash commands available.
Added an @<upstream> model-selector suffix to pin an aggregator model to a single upstream provider per invocation, e.g. --model openrouter/z-ai/glm-4.7@cerebras (sets OpenRouter provider.only; Vercel AI Gateway models map to vercelGatewayRouting.only). Resolved through parseModelPattern, so it works for --model/--smol, model roles, and the SDK, and composes with a trailing thinking level (...@cerebras:high). The base must resolve to an aggregator (openrouter.ai / ai-gateway.vercel.sh); otherwise the @ stays part of the id, so ids that legitimately contain @ (claude-opus-4-8@default, workers-ai/@cf/...) are unaffected.

Fixed

Fixed a turn-ending provider error (e.g. a 502 whose body is the proxy's full HTML page) flooding the transcript: AnthropicApiError folds the entire response body into errorMessage, and the inline transcript render reprinted it verbatim — every embedded blank line included — leaving a tall mostly-empty block ending in </html>. The inline error now drops blank lines, clamps to 8 lines, and width-truncates each line via getPreviewLines, matching the pinned error banner.

[15.10.7] - 2026-06-08

Fixed

Fixed MCP OAuth fallback rendering to show a short terminal hyperlink and keep the raw authorization URL on one unwrapped copy line (#2121).
Fixed omp startup blocking 25–30 s on a single unresponsive MCP server when no cached tools were available for it. MCPManager.connectServers used to fall through to an unbounded Promise.allSettled over every still-pending server without a cached tool list, so one server stuck waiting on the per-request MCP timeout (OMP_MCP_TIMEOUT_MS, default 30 000 ms) gated the entire UI ready signal. Pending-without-cache servers are now left in flight: their tools surface via the existing background #onToolsChanged → refreshMCPTools path the moment the connect completes, and failures continue to log through the background catch handler (#2100).

[15.10.6] - 2026-06-08

Added

Added a /plan-review command that manually (re-)opens the plan-review overlay while plan mode is active. Since there is no fixed plan filename, it reviews the newest local://<slug>-plan.md the agent wrote — useful for pulling the review back up after dismissing it, or reviewing a plan the agent wrote without calling resolve.

Changed

Reverted the task tool's result-header glyph from the ⇶ signature icon back to the quiet status.done bullet (•): white while a subagent is running, accent once it finishes. The reviewer verdict line and per-agent result lines use the same bullet.

Fixed

Fixed /login API-key prompts (OpenCode Zen, Perplexity OTP, GitHub Enterprise URL, manual OAuth redirect URL, …) silently dropping pasted content on kitty/Linux/Wayland — and any other terminal supporting OSC 5522 enhanced paste. InputController enables kitty's enhanced clipboard protocol on TUI start and consumes the resulting OSC 5522 packets in an addInputListener that runs before focus dispatch, so the paste never reached the modal Input's bracketed-paste handler; the routing then stuffed the text into the main CustomEditor unconditionally, even when selector-controller had detached the editor and focused a temporary OAuth input. The pasted API key accumulated in the hidden editor and only resurfaced in the main prompt when the user dismissed the modal with Enter or Esc. The enhanced-paste callback now consults ui.getFocused() and routes the text to the focused component when it exposes a pasteText hook, falling back to the editor only when no modal target is in focus; image pastes refuse with a status message instead of stuffing a binary blob into the hidden editor. (#2127)
Fixed an auto-compaction dead loop when compaction.strategy was shake and the configured threshold was low enough that a single shake pass could not bring the context below it (e.g. a 50K-token threshold on a session well above it). Each pass auto-continued, the next agent turn re-triggered the threshold check, and the second shake had nothing new to drop, so the session spun forever. The shake recovery path now estimates post-shake context and, when it is still above the threshold (or shake reclaimed nothing on overflow recovery), surfaces a one-shot warning and falls back to the summarization-driven context-full compaction so progress actually resumes (#2119).
Fixed /skill: prompts so magic keywords and turn-budget directives in skill args inject the same hidden notices as normal user prompts, matching the editor highlight behavior (#2128).
Fixed MCP OAuth fallback rendering to show a short terminal hyperlink and keep the raw authorization URL on one unwrapped copy line (#2121).
Fixed the task tool rendering a success bullet and a success frame state for detail-less error results (e.g. an argument-validation failure that never executes): the header now shows the error glyph with an error border and error state, and surfaces the dispatched agent name.
Fixed Agent Control Center new-agent creation so Windows Ctrl+Enter sequences submitted as a single LF generate the agent instead of inserting a newline (#2118).
Fixed plan-mode subagents preserving read-only specialty tools such as report_finding while still stripping mutating tools (#1998).
Removed unreachable standalone Exa tool-suite exports and stale tool-count barrel exposure while keeping the live Exa web_search provider helpers (#2093).
Fixed omp commit split plans accepting hunk selectors that resolve to no parsed hunks, which crashed the apply step after the index reset and left the working tree fully unstaged (#2098).

[15.10.5] - 2026-06-08

Added

Added Homebrew and mise package-manager update paths to the self-update command so installations launched from those tools are updated through their native workflows
Added detection of Homebrew and mise install locations so self-update chooses the manager-specific updater when the active omp binary comes from a package-manager-managed path
Added astCondition to TTSR rule frontmatter as a syntax-aware alternative to regex condition, enabling AST-based matching for edit/write tool snapshots
Added a built-in ts-redundant-clear-guard rule that flags redundant guards around clearTimeout, clearInterval, and clearImmediate calls
Added a built-in ts-no-test-timers rule that flags real timers (Bun.sleep, setTimeout, setInterval) in *.test.ts files, steering toward fake timers (vi.useFakeTimers() / vi.advanceTimersByTime())
Added support for paste marker highlighting with accent styling ([Paste #N, +X lines]/[Paste #N, Y chars]) in the prompt editor, matching the visual treatment of image references
Added pixel dimensions to pasted/loaded image placeholders in the prompt — the marker now reads [Image #N, WxH] (falling back to [Image #N] when the header can't be decoded).
The bundled shell now treats nohup as a builtin: nohup … & runs the command without masking SIGHUP or detaching it, so agent-started daemons stay tied to this agent's lifetime instead of leaking as orphans when the agent exits. Updated the bash tool prompt's daemon guidance to match (dropped the nohup … & / setsid … & / disown detach recommendation in favor of a large timeout plus the persistent session).
Added per-tool tool.* theme symbol keys (nerd/unicode/ascii presets) plus a quiet status.done glyph, so each tool's result header can carry a signature icon instead of a generic status mark

Changed

Updated pi-ai OAuth imports to the renamed @oh-my-pi/pi-ai/oauth subpath (was @oh-my-pi/pi-ai/utils/oauth) across the login UI, MCP OAuth flow, model registry, setup wizard, and web-search Codex auth. The legacy-plugin specifier shim drops its pi-ai/oauth → pi-ai/utils/oauth subpath rewrite, since the canonical @oh-my-pi/pi-ai/oauth export now resolves directly.
Changed forced self-updates for Homebrew installs to run brew reinstall and for mise installs to run mise install --force after mise upgrade when --force is requested
Changed TTSR rule bucketing and matching so rules with only astCondition are treated as TTSR rules and evaluated in the interrupt flow using reconstructed edit/write source snapshots
Normalized image content before it enters model context so attached images are downscaled and preprocessed for prompts, steering messages, follow-ups, and custom agent messages
Changed image marker format to include pixel dimensions when available ([Image #N, WxH]), falling back to bare [Image #N] when header cannot be decoded
Changed the prompt editor to highlight large-paste placeholders ([Paste #N, +X lines]/[Paste #N, Y chars]) with the same accent styling as image references (bold, no hyperlink), and to delete image/paste markers atomically: a single backspace or forward-delete removes the whole marker instead of leaving a broken [Paste #N, +X lines behind.
Browser tool helpers (tab.*) are now individually tracked and time-bounded: when a run cell hits its budget, the timeout error names the still-running helper(s) and how long each has been stalled (e.g. ... (stalled on tab.screenshot({ selector: ".x" }) (29.9s))) instead of the opaque Browser code execution timed out after 30000ms. Page-coupled helpers that should resolve quickly (observe, screenshot, extract) also fail fast with a named per-op error at min(cellBudget, 20s), leaving budget for the rest of the cell, rather than silently consuming the whole budget.
Derived the auth-broker OAuth callback ports (CALLBACK_PORTS) and the paste-code login-provider set from the @oh-my-pi/pi-ai provider registry, removing the duplicated CALLBACK_SERVER_PROVIDERS tables in the model selector and the setup-wizard sign-in scene.
Raised the eval tool's per-cell timeout ceiling from 600s to 3600s (matching bash), in both the Zod schema and the TOOL_TIMEOUTS.eval runtime clamp, so heavy local-compute cells can request budgets above 10 minutes.
Derived the auth-broker OAuth callback ports (CALLBACK_PORTS) and the paste-code login-provider set from the @oh-my-pi/pi-ai provider registry, removing the duplicated CALLBACK_SERVER_PROVIDERS tables in the model selector and the setup-wizard sign-in scene.
Reworked tool result-header glyphs to cut the overused success checkmark/dot: each tool now shows its own signature icon on success (terminal for bash, pencil for edit, magnifier/globe for search, plug for MCP, etc.; read keeps the read-group status dot), tools without a custom renderer fall back to a quiet status.done dot, and error/warning/pending states keep the universal cross/warning/spinner
Changed steady-state health indicators (LSP server ready, OAuth logged-in, plugin-doctor checks) from a success checkmark to a colored status.enabled dot, so failures stand out instead of every line reading as a check
Changed one-shot MCP/SSH/debug confirmation messages from a generic checkmark to contextual action glyphs (add/remove, connect/enable/disable toggles, reload, job-completed), reflecting what happened rather than just "success"
Derived the auth-broker OAuth callback ports (CALLBACK_PORTS) and the paste-code login-provider set from the @oh-my-pi/pi-ai provider registry, removing the duplicated CALLBACK_SERVER_PROVIDERS tables in the model selector and the setup-wizard sign-in scene.

Fixed

Fixed package subpath exports for status-line, setup-wizard, tool-discovery, and gallery fixture modules so rewritten test imports resolve through @oh-my-pi/pi-coding-agent.
Fixed runtime model provider discovery so extension-registered providers are now refreshed after extension load and extension-supplied models appear without restarting
Fixed task-row shimmer timing so every running description starts its highlight on the first character together and reaches the last character together, regardless of text length.
Fixed the eval tool's read/write/append helpers (both Python and JS backends) treating local:// (and other internal-URL) paths as plain filesystem paths. pathlib.Path/path.resolve collapse local://x.md to local:/x.md, so write("local://x.md", …) created a junk local: directory under the cwd instead of writing where read local://x.md resolves. The helpers now substitute injected on-disk roots for known schemes (currently local://, pinned to the session's own local:// root), reject path traversal and unknown scheme:// paths, and leave plain paths resolving against the cwd.
Fixed read and edit previews to surface the enclosing syntactic block's off-window boundary line (behind an ellipsis) when a shown line opens or closes a block whose other end falls outside the displayed range. Powered by a new tree-sitter enclosingBlockBoundaries native, so it covers brace languages and indentation languages (Python) using real syntactic spans, with a lexical bracket scan as fallback for unparseable sources.
Fixed tab.screenshot({ selector }) hanging for the entire cell budget on continuously-animating pages (WebGL / backdrop-filter "glass" effects). The element-screenshot path no longer routes through puppeteer's scrollIntoViewIfNeeded(), whose IntersectionObserver promise can stall indefinitely under heavy rendering; it now does a single instant scrollIntoView and captures with scrollIntoView: false (relying on captureBeyondViewport), so off-screen elements are still captured without the stall.
Fixed follow-up message submissions to forward pending clipboard-pasted images to session.prompt in both streaming and non-streaming flows
Fixed follow-up handling to clear consumed clipboard image state after submission so pasted images are not silently carried into later messages
Fixed clipboard-pasted images being rejected when steering or following up during compaction. Instead of bailing with "Retry after it completes to send images", the message and its images are now queued via queueCompactionMessage and forwarded to the session (steer/follow-up/prompt) when the compaction queue flushes.
Fixed edit tool result previews to show only current-file lines and collapse long inserted blocks instead of echoing removed content.
Fixed generateDiffString to omit the mid-skip ... placeholder between two nearby edits, conveying the elided gap via the jump in line numbers instead (consistent with how leading/trailing context skips already render). The placeholder row was indistinguishable from a genuine ... context line and wasted a row in compact previews.
Fixed concurrent interactive dialogs clobbering each other on the shared editor surface. ExtensionUiController presents the selector / input / editor modals by swapping a component into the single editorContainer and stealing focus, with no serialization — so a second select/input/editor request (from a hook, extension, the ask tool, or an internal flow) opened while one was already up would clear the container and re-focus, orphaning the first dialog. Its promise then hung until the caller's signal aborted (surfacing a stray Ask input was cancelled on top of the answered call). These modals are now serialized through #presentDialog: at most one shows at a time and the rest queue (FIFO); a queued request whose signal aborts before its turn resolves undefined and is never shown. The first dialog is still presented synchronously, so single-dialog timing is unchanged.
Fixed the ask tool potentially hanging when the model emitted two ask calls in one tool batch. ask now declares concurrency: "exclusive", so the agent loop serializes the batch and each question's selector runs to completion before the next starts, instead of racing for the shared selector surface.
Expanded @path/to/file import references in CLAUDE.md / AGENTS.md / GEMINI.md (and the other discovered context-file flavors) when loading them into the system prompt, matching the convention used by Claude Code, Goose, and other agents. Imports resolve relative to the importing file's directory, support ~/..., recurse up to 5 hops, and are skipped inside fenced code blocks and inline code spans so technical examples like npm install @types/node survive intact (#2111).

Removed

Removed the special Anthropic claude-opus-4-8 tool-call batch cap; sessions no longer abort an in-flight provider stream after a fixed number of completed tool calls.

[15.10.4] - 2026-06-08

Added

macOS release binaries are now signed with a Developer ID Application identity (hardened runtime + secure timestamp + JIT/library-validation entitlements) and notarized in CI when the APPLE_* signing secrets are configured; releases auto-fall back to ad-hoc signing until then. This makes the shipped binaries Gatekeeper-acceptable, unblocking an official Homebrew submission (#776). See docs/macos-signing-notarization.md.
Added a Homebrew install path: brew install can1357/tap/omp. The can1357/homebrew-tap formula installs the prebuilt release binary, and a release_brew CI job regenerates it (version + per-asset sha256) from each published release via scripts/ci-update-brew-formula.ts (#776).

Changed

Adjusted completion() model resolution so the default tier now prefers the session’s active model and falls back to the configured default role when needed
Rewrote the session auto-title prompt (prompts/system/title-system.md) and the set_title tool description to ask for a concise, sentence-case title (3-7 words) that captures the session's topic/goal, with good/bad examples and explicit guidance to treat the first message as data (no following embedded links/instructions, no refusals, describe URL/reference asks). The local on-device title prompt (tiny-title-system.md) was aligned to the same 3-7 word, sentence-case convention. The deterministic greeting/low-signal filter and the none deferral sentinel are unchanged.
Renamed the eval oneshot helper from llm() to completion() in both JavaScript and Python preludes, including status events, prompt docs, and runtime tests.

Fixed

Fixed completion() to always send a non-empty default system prompt when system is omitted so providers that require instructions no longer reject requests
Fixed structured completion() mode to return parsed JSON from plain text output when the model skips the forced respond tool call
Fixed slow-tier completion() reasoning requests to avoid unsupported effort settings by only enabling reasoning on reasoning-capable models and capping effort to supported levels
Fixed JS eval worker reset/dispose to close workers gracefully before forced termination, avoiding Bun 1.3.14 N-API teardown crashes with native modules such as canvas.

[15.10.3] - 2026-06-08

Added

Added clickable file path hyperlinks to read tool outputs (read-call rows, grouped summaries, and inline previews) using resolved or absolute file targets with selector-based line anchors for quick navigation
Added a resolved-span echo to replace block/delete block edits: a successful block op now prints replace block N → resolved lines A-B (K lines) between the section header and the diff preview, so the model can confirm tree-sitter matched the construct it intended (e.g. catch a decorator left outside the block) instead of inferring the span from the diff after the fact.

Changed

Changed the find tool to process each explicit multi-path target separately before merging results so searches stay scoped to the requested paths
Changed multi-path find handling so invalid extra targets no longer fail the whole query and now return matches from valid targets only
Changed background-job completion and late LSP diagnostic delivery to inject at the next agent step boundary (mid-run), via the new non-interrupting "aside" channel, instead of only when the agent reaches a yield/follow-up point. The model now sees these notifications between its own requests without the turn having to end first, and in-flight tools are never interrupted; job-poll acknowledgement still suppresses results the agent already saw.
Changed late LSP diagnostics after edit or write to surface in the chat transcript as Late diagnostics entries rendered through the same grouped tree renderer the edit/write tools use (per-file nodes, severity icons, :line:col locations), and to honor the global tool-output expand toggle (collapsed entries cap at 5 diagnostics with a … N more hint)
Changed delayed diagnostics delivery to batch late results in one message per flush instead of a raw hidden custom payload
Changed hidden custom messages and file-mention context to reach providers as developer messages instead of user-authored turns, so system reminders no longer pollute compacted user history.
Rewrote the plan-mode active prompt (prompts/system/plan-mode-active.md) from scratch to stop producing shallow plans. Reframed the artifact as an execution spec a fresh agent runs after the planning conversation is cleared/compacted (zero design decisions for the implementer) rather than a brevity-capped summary. Folded high-consensus requirements into the existing sections as inline, conditional rules — no new boilerplate sections: ordered Approach steps that keep the build/tests green after each step (sequencing); exact signatures/literals for new or load-bearing symbols (contracts); full callsite list + clean cutover for renames/signature-changes/removals; Verification that must exercise the new behavior (input → observable output) with run preconditions, not just build/typecheck; Assumptions restricted to user-overridable choices plus pre-decided fallbacks for load-bearing assumptions; a provenance rule (plan facts must come from a read this session; unverified claims flagged inline); and bans on conversation back-references and decision-free sections (Non-Goals/Alternatives/Risks/Future Work). Kept the decision-complete self-check and the brevity-vs-completeness tiebreak (completeness wins). Render contract (Handlebars vars/conditionals) unchanged; verified across all planExists/reentry/iterative branch combinations.

Fixed

Fixed duplicate find matches in multi-target queries by deduplicating overlapping paths in merged results
Fixed find partial updates to avoid repeated streamed rows while scans are still running
Fixed stale late diagnostics from older edits being shown after a file was edited again
Fixed read output paths so selector suffixes are preserved when corrected paths were returned without selectors
Fixed read surfacing a misleading red "Operation aborted" on a plain-file or directory read when a turn was interrupted mid-read. Those reads are deterministic and fast, so execute now runs them to completion instead of cancelling them; slower/non-deterministic reads (archive, sqlite, document, image, summary, conflict scan, URL) stay cancellable.
Fixed edit tool headers to hide first-change line suffixes, middle-elide long paths only when the header width needs it, show compact change stats, and target encoded file:// hyperlinks.
Fixed Esc interrupts rendering a redundant Interrupted by user assistant transcript line while preserving the interrupt reason for tool-result placeholders and continuation logic.
LSP writethrough no longer burns the full diagnostics poll on every edit/write. typescript-language-server never echoes the document version in publishDiagnostics (upstream #983), so the exact-version gate never passed; waitForDiagnostics now accepts an exact version match instantly and otherwise settles on the latest publish after a short quiescence window, dropping superseded in-flight diagnostics.
LSP writethrough no longer blocks the whole edit/write on slow diagnostics: it now waits only a short inline window (~500ms) for a settled result, then hands the in-flight fetch to the deferred channel so a slow or cold language server (e.g. a large-project tsserver) delivers its diagnostics as a follow-up message instead of stalling the tool 3–5s on every edit. The background fetch also gets a longer budget so slow servers still surface late rather than being dropped.
Fixed the c/. continue shortcut making the agent second-guess itself after an Esc interrupt. Continuing used to submit an empty user turn, which left the model with only the aborted-turn context — so it tended to restate the halted state and ask whether to proceed rather than just continuing. The shortcut now resumes with a hidden agent-authored developer directive ("keep going — don't stop to summarize or re-confirm the plan") instead of an empty turn. It still produces no visible transcript entry, same as before.
Fixed native scrollback commit boundaries to be computed generically from finalized transcript blocks and observed append-only live growth, so tall final tool results and streaming previews keep their scrolled-off heads on ED3-risk terminals without per-tool append-only predicates; live blocks that re-layout remain deferred until finalization or the next checkpoint.
Fixed read-group summaries for multi-path read results to use result-provided display targets so each resolved path is shown as its own row
Fixed read-group range summaries to abbreviate long merged selectors with ellipsis to keep repeated-file range rows readable
Fixed read-group TUI summaries so a single delimited read call renders as separate read rows, and repeated reads of the same file collapse under one file with full-file/range children.
Fixed grouped read rows freezing on their pending "⏳ Read " preview on ED3-risk terminals (ghostty/kitty/iTerm2/…) when a parallel sibling tool closed the read run and appended a block below the group before the read's result arrived. The read-group block now stays in the repaintable live region until its entries settle, so the late success result repaints instead of being stranded; a seal() escape hatch (turn end / transcript rebuild) still lets a never-delivered read freeze rather than pinning the live region.
Fixed session search to return all sessions unchanged when the query is blank
Fixed duplicate session suggestions by deduplicating history matches by session path when merging metadata and prompt-history results
Fixed /resume search ranking so sessions whose prompts or metadata match the query now prefer prompt recency and recent literal matches instead of letting older earlier-title fuzzy matches outrank a just-used session.
Fixed omp --resume <id> / --fork <id> crashing with [Uncaught Exception] when the id did not match a known session. createSessionManager now throws a dedicated SessionResolutionError, which runRootCommand catches to print Error: Session "..." not found. plus a hint to stderr and exit with code 1. The same path covers --fork combined with --no-session and the non-interactive cross-project / moved-cwd prompts that previously surfaced raw stack traces (#2084).

Removed

Removed the animated pending border ("shimmer") on running bash, eval, and ssh execution blocks. While pending, a block now shows a static accent border instead of sweeping a dark segment around its bottom edge; display.shimmer still governs the working-status line and task row animations.
Removed the tool-level nonAbortable bypass so write and edit honor the active turn AbortSignal. read is abortable for everything that is slow or non-deterministic (URL/internal-URL reads, archive, sqlite, document conversion, image decode, structural summary, conflict scan, suffix glob); only the deterministic plain-file line/range reads and directory listings run to completion.

[15.10.2] - 2026-06-08

Added

Added raw-sse.txt to debug report bundles, exporting recent raw provider SSE diagnostics when captured
Added /model visibility for auto-selected role defaults: inferred pi/smol/pi/slow/designer choices now show as compact [ROLE auto] badges, while explicitly configured roles keep the existing solid badges and thinking labels.
Added credential provenance to the /login and /logout provider picker: each authenticated provider now shows where its credential comes from — (login), (api key), (env: VAR_NAME), (config), (--api-key), or (custom provider) — so a real OAuth login is distinguishable from an env var that merely aliases the provider (e.g. COPILOT_GITHUB_TOKEN). The origin is also matched by the picker's type-to-search filter.

Changed

Changed raw SSE debug export output to prepend dropped-record metadata so truncated sessions in debug bundles now report dropped record and character counts
Changed settings reads to cache pre-split schema paths and resolved values, with coarse invalidation on source/cwd changes.
Changed status-line rendering to cache merged effective settings until updateSettings() changes the configuration.
Changed CustomEditor app shortcut dispatch to parse each input packet once and match against precomputed canonical key sets, preserving the existing shortcut precedence while avoiding repeated key reparses.
Changed lsp references to retry only when no references or only the queried declaration are returned, using two fixed 250ms retries for project-aware servers
Changed read handling of https://github.com/<owner>/<repo>:raw to use raw page rendering only, removing the GitHub API README fallback
Changed model resolution to apply provider-priority ordering when selecting models for roles and ambiguous patterns, using modelProviderOrder settings and built-in provider priority so first-party providers are preferred over relays in tie cases
Changed model canonical variant selection to use the same provider-priority ordering instead of candidate order when deduplicating equivalent upstream models
Changed the working-message shimmer to sweep at a fixed velocity (cells/second) instead of a fixed sweep duration divided by the message length. The band now advances ≤1 cell per 30fps redraw frame and stays equally smooth on short and long messages — previously a longer message swept proportionally faster and stepped visibly because it outran the redraw cadence. Sweep/round-trip duration now scales with length. Additionally, when display.shimmer = disabled the working line is static, so the loader no longer schedules 30fps redraws for it and falls back to the spinner-only ~12.5fps cadence.
Changed the eval fan-out trigger keyword from workflow/workflows to workflowz.

Fixed

Fixed working-message loader session accents so spinner/message color math is cached per session name, session-accent setting, and theme luminance while still updating immediately on renames, setting toggles, and theme changes.
Fixed startup model fallback selection so sessions now prefer each provider’s configured default model before choosing the first available authenticated model
Fixed implicit model selection path for tools and sessions by honoring persisted model-provider order when no explicit pattern is provided
Fixed the working spinner appearing to ignore Esc for 2-3 seconds when an interrupt lands mid-tool. Esc fires the abort synchronously, but the agent loop only stops the loader at agent_end, which it cannot reach until every in-flight tool settles in executeToolCalls' await Promise.allSettled(...) — and process/subagent/kernel-owning tools tear down gracefully (SIGTERM, 2-3s grace, SIGKILL), so the loader kept showing the unchanged "Working…/" line and read as a dropped keypress. The loader now switches to "Interrupting…" the instant Esc requests the abort and freezes intent-driven label updates until the turn unwinds (EventController.notifyInterrupting), so the interrupt is acknowledged immediately even while teardown completes.
Fixed a flaky JS eval worker startup that intermittently failed unrelated CI runs. The worker-ready wait reused Bun's 5s default per-test timeout as its floor, so a slow cold-start under --isolate + high concurrency was aborted mid-init; terminating a still-initializing Bun worker is the documented SIGILL/SIGTRAP crash trigger, which took down the whole test file. Worker init now floors at a fixed 15s infrastructure budget (independent of, and still dominated by, a larger per-cell timeout), and the JS eval test suites set a 20s file-local timeout so cold starts complete instead of being torn down.
Fixed reviewer-style subagent yields crashing the calling eval cell when a caller-supplied output schema declares additionalProperties: false without a findings property. normalizeCompleteData now consults the active validator before splicing collected report_finding entries onto the yielded payload, so injection is suppressed when the schema would reject it — keeping the executor's post-mortem validation in lockstep with the in-tool yield validation that already accepted the same raw payload (#2070)
Fixed Anthropic empty toolUse stops without tool calls corrupting session history by retrying them and removing orphaned turns even at the retry cap.
Fixed MCP tools hanging in non-yolo modes by declaring approval = "write" on MCPTool and DeferredMCPTool, and propagating the approval property through customToolToDefinition() in sdk.ts
Fixed session resumption after a working directory is moved/renamed (e.g. git worktree move): --continue now re-roots the terminal's last session into the new directory when its original directory no longer exists, explicit --resume <id> --session-dir <dir> local matches re-root instead of reopening with the stale cwd, and cross-project --resume <id> offers to move (re-root) the session rather than only forking a duplicate copy when the source directory is gone
Fixed Kitty OSC 5522 paste rejecting plain text as "no supported text or image data": the listing parser now decodes the mime="." DATA payload (whitespace-separated MIME list) Kitty actually sends, in addition to the per-type DATA packets described by the ancillary 5522-mode spec, and per-type spec listings now request the selected payload with type=read:mime=... instead of Kitty's dot-payload request shape (#2051)
Fixed follow-up shortcut submission of builtin slash commands so /goal set ... applies goal mode instead of queueing as plain text.
Fixed Ctrl+Z crashing the agent on Windows with TypeError: Unknown signal: SIGTSTP. InputController.handleCtrlZ called process.kill(0, "SIGTSTP") unconditionally, but SIGTSTP is POSIX job-control and Bun/Node on Windows rejects the signal name from the JS side; the throw propagated out of the TUI input dispatcher as an uncaught exception. The handler now no-ops with a "Suspend (Ctrl+Z) is not supported on this platform" status on Windows, and on POSIX wraps process.kill in a try/catch that detaches the registered SIGCONT resume hook and re-start()s the TUI on failure so a rejected signal can never leave the UI stranded with a leaked listener (#2036).
Fixed a relative --cwd target (e.g. omp --cwd repo launched from /tmp) leaking the raw relative string into the session config. applyStartupCwd chdired into the resolved directory via setProjectDir but left parsed.cwd as "repo", so buildSessionOptions (which prefers parsed.cwd over getProjectDir()) handed downstream settings/discovery/session creation a value that re-resolved against the new process cwd (/tmp/repo/repo) or persisted a relative session cwd. parsed.cwd is now re-synced to the resolved absolute project dir after the chdir.
Fixed the --cwd launch flag so it is parsed and can override the startup directory instead of always falling back to the current process directory or home auto-switch target.
Fixed session auto-retry for generic upstream_error: Upstream request failed gateway failures.
Stripped read-output line-number prefixes (N:) from auto-piped bare body rows in the hashline edit parser, so pasting 3:text without a + prefix no longer injects 3: as literal content. Uses single-pass stripping to avoid corrupting content whose own text starts with digits: (#1492).
Fixed eval llm() returning HTTP 400 "Instructions are required" when called without a system prompt against providers (notably openai-codex) whose Responses transformer drops the instructions field on an empty system prompt. runEvalLlm now sends a minimal default system prompt ("You are a helpful assistant.") when no system is supplied, so llm("question") works against every provider; an explicit system= still wins.
Fixed the Python read(path, offset, limit) prelude helper rejecting documented positional arguments with TypeError: read() takes 1 positional argument but 3 were given. The signature was keyword-only (def read(path, *, offset=1, limit=None)) while the eval helper table advertises positional optional args; agents that called read("file.py", 10, 20) literally crashed. The * is removed so both read("f", 10, 20) and read("f", offset=10, limit=20) work.
Fixed eval reset cells failing with "Python kernel reset already in progress" / "JS context reset already in progress" when two cells happened to overlap on the same session (e.g. a rapid resubmit, or a parallel-cell race). The executor now coalesces concurrent resets — additional callers wait for the in-flight reset to finish and then run on the freshly restarted kernel — instead of throwing a user-visible error for what is purely an internal coordination state.
Fixed the eval tool description advertising the agent() helper unconditionally even in subagent sessions whose parent forbids spawning. When getSessionSpawns() returns "", the prelude doc now omits agent() so the model is not promised a helper that can only ever throw "Cannot spawn 'task'. Allowed: none (spawns disabled for this agent)".
Fixed plan mode rejecting local:// plan-artifact edits when addressed via the absolute path the read tool echoes back in the [path#tag] header. enforcePlanModeWrite previously only matched the literal local:// scheme; it now also accepts any absolute path whose realpath resolves inside the session's local sandbox root, so the absolute spelling and the local:// spelling are interchangeable in plan mode.
Fixed snapshot tags freshly minted by read being rejected as stale by a subsequent edit against the same file when the two sides reached the file via symlink-equivalent spellings (e.g. macOS /tmp/… vs /private/tmp/…, or read local://foo.md recording under the file's fs.realpath while edit local://foo.md looked up under the raw path.resolve(localRoot, …) form). The file snapshot store now keys every record/lookup through a realpath-canonicalized key (canonicalSnapshotKey), fusing all spellings of the same on-disk file onto one snapshot entry.
Fixed read of a github.com/<owner>/<repo> URL with :raw returning the full JS-rendered HTML shell. Repo roots now resolve to the decoded README via the GitHub API (/repos/<owner>/<repo>/readme), falling back to the raw HTML only when the API returns no usable payload.
Fixed issue:// and pr:// reads returning stale OPEN/CLOSED state after a successful gh issue close / gh pr merge (or any other state-changing gh invocation) in the same session. The bash tool now invalidates the matching github-cache rows before executing any gh (issue|pr) <close|reopen|merge|delete|edit|comment|lock|unlock|pin|unpin|transfer|develop|ready|review> command.
Fixed line-range selectors on PDF/DOCX/PPTX/XLSX/RTF/EPUB reads being ignored. The markit-converted markdown body now flows through the same in-memory range slicer used for plain text, so file.pdf:50-100 and file.pdf:5-16,40-80 slice the converted body instead of returning the whole document.
Fixed the read selector cheatsheet incorrectly promising "exactly one line" for :N+1 while the implementation pads single-line reads with ≤1 leading and ≤3 trailing context; documented that multi-range selectors do not pad, giving callers a way to request exact bounds.
Documented the bash.autoBackground.enabled behavior in the bash tool prompt so the Background job <id> started: … notice for foreground commands that exceed autoBackgroundThresholdSeconds no longer reads as a tool malfunction.
Fixed task subagents whose in-tool yield validator had already accepted a payload after exhausting MAX_SCHEMA_RETRIES being rejected a second time by the post-mortem executor validator. The override now propagates through the yield tool's details.schemaOverridden flag, and the executor surfaces a SUBAGENT_WARNING_SCHEMA_OVERRIDDEN stderr line instead of re-emitting schema_violation for data the subagent already had to ship. Finalize also degrades to no validation (matching the yield tool's looseRecordSchema fallback) when the caller-supplied output schema fails to normalize.
Fixed the web_search codex provider returning (see attached image) / [Attached image] / See image above and similar non-informational image-placeholder strings as the answer. Detection broadened to a small regex set, and when annotations did produce sources we now drop the placeholder prose from answer (returning sources only); when neither annotations nor a real answer materialize, we throw 502 to advance the provider chain.
Fixed search, find, ast_grep, and ast_edit rejecting bracket-containing file paths (Next.js routes like apps/[id]/page.tsx) as glob patterns when the literal path exists on disk. parseSearchPathPreferringLiteral now prefers the literal interpretation for paths that resolve on disk and only falls back to glob expansion when the literal does not exist.
Fixed search with an external http(s)/ftp/ws/file:// URL in paths surfacing a misleading "Path not found" error. The tool now rejects external URLs with a clear "use read for URLs" message.
Fixed search rejecting skip: null at the schema layer; null now normalizes to 0 alongside the omitted case, matching how callers serialize default pagination state.
Fixed search returning zero matches with no explanation when explicit file targets exceed the native grep cap (4 MB). The tool now surfaces a Skipped oversized files (>4MB grep limit; …) notice listing the truncated paths.
Fixed the archive-extraction error message in search recommending grep — which the system prompt forbids — instead of pointing to read <archive>:<member>.
Fixed browser tab.open(name, { viewport }) on an existing tab not applying the new viewport: acquireTab's reuse path now resizes the page in addition to navigating.
Fixed browser tab metadata leaking user:pass@ basic-auth credentials in the URL surfaced to transcripts and observe snapshots; URLs are now redacted via redactUrlCredentials().
Fixed browser tab.extract(format) returning a ReadableResult | null shape that the tool prompt advertised as plain content. The helper now returns the markdown/text string directly (or throws a clear ToolError when extraction is empty), and the prompt matches.
Fixed lsp requests timing out at a hard-coded 30 s ceiling when the caller supplied an explicit abort signal (e.g. the tool wall-clock). The signal is now the deadline; the 30 s default still applies when neither a signal nor an explicit timeoutMs is provided.
Fixed lsp status reporting servers as Active language servers: … when the binary resolves on PATH but never spawns (rustup wrapper, missing toolchain component, etc.). Status now labels each entry as (ready) or (configured, not started).
Fixed lsp rename_file fanning willRenameFiles requests across every configured server (including ones with no jurisdiction over the file type) and burning the wall-clock timeout. The action now pre-filters configured LSPs to those whose fileTypes cover the source or destination path, falling back to a plain filesystem rename when no server claims the type.
Fixed lsp references returning only the queried declaration (or only in-file results) on project-aware servers that had not finished indexing. The retry budget is raised from 2 → 3 with 250 / 500 / 1000 ms backoff, and the retry trigger now also fires when all results live in the queried file.
Fixed lsp config accepting fileTypes entries with or without a leading dot inconsistently across actions; both .ts and ts are now normalized so a missing-dot entry no longer silently excludes a server from extension-based routing.
Fixed lsp request error path swallowing the params that were sent, making shape/coercion bugs on raw LSP calls impossible to diagnose in one round-trip; the error now echoes a truncated copy of the request params.
Fixed find with a single-star segment like dir/* recursing into subdirectories and returning nested matches. parseFindPattern already prepends **/ for top-level globs (*.ts → **/*.ts), so anything reaching native without **/ was deliberately scoped by the user; recursive: false is now passed to natives.glob to honor that scope.

[15.10.1] - 2026-06-07

Added

Added display.smoothStreaming setting (default true) to let users enable or disable smooth assistant-stream text reveal
Added /tan <work> slash command to fork the current conversation into a background agent so tangential work can continue asynchronously while your main session stays active
Added a background /tan dispatch message that records the handoff in the transcript and marks the delegated work as non-blocking
Added providerPromptCacheKey support to CreateAgentSessionOptions so /tan background sessions can reuse the parent session’s prompt-cache lineage
Added session cloning for /tan runs with copied artifacts and shared MCP proxy tools
Added SessionManager.forkFrom’s optional suppressBreadcrumb mode to avoid breadcrumb updates when forking background /tan sessions
Added OSC 5522 enhanced paste handling in InputController, so terminal clipboard events are decoded as image or text payloads and inserted without passing raw paste sequences to the editor
Added bracketed image-path paste support in CustomEditor so a single pasted image file path (PNG/JPEG/GIF/WEBP) is loaded from disk and inserted as an image candidate
Added direct support for Image #N insertion from pasted local image paths by routing successful image-path pastes through the same image normalization and resize flow as clipboard image pastes
Added /fresh to rotate the provider-facing session id and clear in-memory provider stream/cache state without changing the local session file.
Added a ChatBlock transcript primitive (modes/components/chat-block.ts) and a single ctx.present(...) sink (with ctx.resetTranscript()) so chat output is mounted in one place instead of the repeated chatContainer.addChild(...) + ui.requestRender() pattern scattered across controllers. ChatBlock carries a React/Svelte-style lifecycle — onMount starts effects, onCleanup registers teardown, finish() self-completes (stops timers and freezes the block at its final content), and dispose()/resetTranscript() tears everything down — so animated blocks own their own resources instead of leaking setInterval/requestRender bookkeeping into callers. The MCP "Connecting…" spinner is now such a block.
Added a framedBlock output-block helper (tui/output-block.ts) plus a borderColor override and applyBg: false (no background fill) on output blocks, a renderStatusLine iconOverride, and an icon.search (magnifier) theme symbol — so tool renderers can draw self-contained muted-outline frames and search-family tools can show a magnifier instead of a checkmark.

Changed

Changed the bash tool frame to use a plain top rule instead of repeating "Bash" in the title bar, and folded minimizer raw-output artifact links into the status footer as Artifact: <id>.
Changed grouped read output to use a white filled-circle mark for the group/single-read success state and omit duplicate per-file success marks inside multi-read groups.
Changed assistant streaming output to reveal text incrementally at 30 FPS with grapheme-safe adaptive catch-up, instead of replacing the whole message chunk-by-chunk
Changed shimmer-driven TUI animations (working text, pending bash/eval borders, and theme activity-spinner documentation) to render at 30fps instead of 60fps.
Changed running task tool agent rows to use a static • marker and shimmer only the subagent name, leaving descriptions, stats, and nested tool detail text solid while removing the rotating status glyph from those rows.
Changed settings singleton method access to reuse bound methods for the active instance instead of allocating a new bound function on every settings.get lookup.
Changed plan-mode approval to keep the drafted local://<slug>-plan.md file at its original name as the canonical plan path, so approved plans are no longer renamed when leaving plan mode
Changed plan-mode write enforcement so only local:// artifact files are writable during planning, blocking working-tree edits and allowing scratch or draft plan files in the local artifact area
Changed the todo tool result renderer to stop redrawing every phase's full task list on each update: when a multi-phase list is rendered collapsed (the default, not manually expanded), only phases the latest update touched — the phase holding the in_progress task, any phase with a just-completed task, and phases named by the ops that ran (init counts as touching all) — render their tasks; untouched phases collapse to a one-line N. Name done/total summary. When call args are unavailable (e.g. transcript rebuilds) it falls back to the in_progress/completed-transition signals, and the manual expand toggle still shows every task. Also dropped the blank separator line previously inserted between phases.
Changed non-agent API operations (title and commit-message generation, image generation, web search, eval llm(), auto-thinking classifier, memory consolidation) to use session-aware API key resolution with auth retries via registry.resolver() / authStorage.resolver(), refreshing the active credential before rotating to another account
Changed image generation to wrap every provider fetch branch in withAuth, so 401 / usage-limit errors trigger credential force-refresh and rotation for authStorage-backed providers (OpenAI-hosted, antigravity, xai-oauth) while env-only providers (openrouter, gemini) stay single-attempt
Changed web-search providers using authStorage.getApiKey (anthropic, exa, tavily, parallel, synthetic, zai, kimi) to wrap HTTP calls in withAuth for automatic credential rotation on 401 / usage-limit errors
Changed the directory grouping for find, search, ast_grep, ast_edit, and lsp diagnostics from a single flat # dir/ heading per immediate directory to a multi-level tree that folds the common path prefix into one heading. Previously every group repeated the full directory path — so results rooted outside cwd printed the absolute prefix (e.g. /Users/me/proj/) on every heading and nested directories were never collapsed. Now a single-child directory chain folds into one heading (# packages/pkg/src/, including an absolute root for out-of-cwd results), subdirectories nest one # deeper (## nested/ → ### child.ts), and each directory's own files are listed before its subdirectories. TUI hyperlink reconstruction tracks the nested directory stack across the whole output so file and code-frame links keep resolving to the correct absolute paths.
Changed the plan-mode approval surface from an inline transcript block plus a separate bottom selector into a single fullscreen overlay (like /copy) and overhauled its navigation. The overlay now renders the plan per-section through ScrollView (line-level ↑/↓ scroll, Shift+↑/↓ to scroll faster, PgUp/PgDn, g/G) with no stray per-line …, and — when the terminal is wide enough and the plan has ≥2 headings — shows a compact VS Code-style section sidebar (the redundant plan-title heading and any "Contents" label are omitted). Focus moves between regions with Tab/Shift+Tab (and flows at the edges: Down past the last section or the bottom of the body drops into the approval options; Up steps back), while the sidebar glows to track the scrolled section. The sidebar can fast-jump between sections, delete a section (with u undo), and annotate sections with feedback (a); deletions and annotations are collected into refinement feedback that is submitted back to the model when the operator picks "Refine plan". Mouse works too: clicking an approval option activates it, clicking a sidebar section jumps to it, and the wheel scrolls the plan. ←/→ always drive the model-tier slider, Enter confirms, the external-editor key opens the plan, and Esc cancels. The overlay borrows the terminal's alternate screen buffer for its lifetime (fullscreen overlay), so the transcript stays put on the normal screen instead of bleeding through scrollback behind the modal.
Changed the interactive controllers (command, MCP, selector, extension-UI, event), debug panels, and the status/error/warning helpers to render chat output through ctx.present(...) instead of appending to chatContainer and calling ui.requestRender() directly; transcript rebuilds dispose live blocks via ctx.resetTranscript() so animated blocks' timers stop on reset.
Changed tool-execution block rendering so the container (ToolExecutionComponent) is a transparent passthrough — it no longer inserts a top/bottom blank line, adds left/right padding, or paints a state-colored background behind tool output. Tools with substantial body now self-frame with a muted outline and the tool title in the frame's top bar (edit/apply_patch, write, ask, todo, github, goal, inspect_image, search_tool_bm25, task), matching the already-framed bash/read/eval/debug/web_search/lsp blocks, while streaming/in-progress and trivial results collapse to a clean status line. The search-family list tools (find, search, ast_grep) and job render frameless/minimal; find/search/ast_grep show a magnifier on success instead of a checkmark, and job drops its Job: label prefix (the per-job rows are self-describing). The search_tool_bm25, github, and inspect_image frames draw with no background fill, and inspect_image's label was shortened to Inspect.
Changed the plan-mode active prompt (prompts/system/plan-mode-active.md) to make plans decision-complete and cut filler. Added an Objective framing ("another engineer can execute end-to-end without making a single design decision"), a shared "Resolving Unknowns" section (explore discoverable facts before asking; reserve ask for non-derivable preferences/tradeoffs with 2–4 options + a recommended default), and a single shared "The Plan" structure (Context / Approach grouped by behavior not file-by-file / ≤5 Critical files / Verification / Assumptions) that replaces the per-branch structure guidance previously duplicated across the iterative and parallel workflows. Added explicit prohibitions on sections that decide nothing (Non-Goals, Out of Scope, Alternatives Considered, Risks/Mitigations boilerplate, Future Work), on enumerating every file/line, and on inventing schema/validation/precedence policy the request never established.
Changed completion notifications (completion.notify) to fire whenever the agent yields its turn, including in the foreground. The agent_end notification was previously gated behind background mode (isBackgrounded), so an ordinary foreground turn never emitted one; the gate is gone and the desktop toast now fires on every normal turn completion (still skipped for aborted/error turns and when completion.notify is off).
Changed the in-progress task tool block to keep the shared context brief (# Goal / # Constraints background) visible after the first progress snapshot arrives, instead of dropping it the moment the streaming call view was replaced by the result frame, and to stop animating a spinner/clock next to the Task frame header while running — the per-agent body lines already carry their own running spinner, so the header now shows a static state icon (matching the completed/failed header icons). The context is rendered through a shared buildContextSection helper that also undoes per-field double-encoding, so the brief reads cleanly in the result frame even though renderResult receives the raw (un-repaired) tool args.
Changed the messaging shown when you press Esc to interrupt a streaming turn from the ambiguous Operation aborted / Tool execution was aborted: Request was aborted to Interrupted by user, so a deliberate user interrupt no longer reads like an internal failure. Every Esc/flush interrupt path (onEscape while streaming, the queued-message restore-and-abort path, and the empty-submit queue flush) threads the reason through AgentSession.abort({ reason }) → Agent.abort(reason) so it rides the AbortController onto the aborted assistant message's errorMessage; the turn label renders it verbatim on both the live and replay paths, and the synthetic placeholder results paired with in-flight tool calls now read Tool execution was aborted: Interrupted by user. Aborts that carry no reason still fall back to the retry-aware Operation aborted generic. Transcript label resolution is centralized in resolveAbortLabel (session/messages.ts).

Removed

Removed the /background (and /bg) slash command and the background-mode subsystem it was the sole entry point for — InteractiveMode.isBackgrounded, createBackgroundUiContext, handleBackgroundEvent, and every isBackgrounded guard across the input/event/extension-UI controllers and UI helpers. The command suspended the whole process group via SIGTSTP (a leftover testing shortcut) instead of detaching the running agent, which is not the expected workflow — use terminal panes or a multiplexer instead.

Fixed

Fixed inline find and search result blocks to align with grouped read output and render their success headers with the normal tool-title color instead of accent blue.
Fixed the working-status shimmer to opt into the loader's 30fps animated-message repaint path while keeping both the status spinner and pending bash/eval tool spinners on their normal 80 ms glyph cadence.
Fixed consecutive read tool calls failing to collapse into a single grouped block when a reasoning model emits one read per completion ([thinking, read]). The read group was reset on every assistant message_start, so each read rendered as its own one-entry Read … line; now a read run accretes across completions and is broken only by a rendered non-empty text/thinking block, a non-read tool, or a user/IRC message — matching the transcript-rebuild path. ReadToolGroupComponent now reports its live/finalized state so the growing Read (N) header repaints correctly on native-scrollback (risk) terminals.
Fixed the task tool shared-context brief rendering raw Markdown headings (# Goal, # Constraints) inside framed call/result blocks instead of using the normal Markdown renderer.
Fixed the animated pending border on bash/eval blocks leaving a frozen dark "bar" segment behind after a backgrounded command finalized through the async update path. Once a command is auto-backgrounded (details.async.state === "running") the block stays "partial" in the TUI until the async job-manager delivers the final result, but it also gets committed to native scrollback — so a mid-sweep shimmer frame baked a stray darkened border segment into the committed copy. The border now stops animating (and the 60fps redraw loop stops) the moment a block enters the backgrounded state, so the committed frame is a clean static border.
Fixed cold omp launch to clear native terminal history on the first paint, avoiding a once-per-launch duplicate welcome/transcript copy before the normal session replay.
Fixed plan approval resolution so resolve with action: "apply" can still find the plan file when extra.title is missing or stale by falling back to the current plan path and most-recent local plan artifacts
Fixed the search-family tool magnifier glyph (find, search, ast_grep, search_tool_bm25) to use the accent title color instead of success green, so the icon matches the tool title in the status header instead of standing out
Fixed TTSR stream interrupts to pass the matched rule name through the abort reason, so aborted in-flight tool placeholders say why they were stopped instead of Request was aborted.
Fixed URL reads for binary/special payloads to reuse local readers: remote archives list their root entries, SQLite databases show their table overview, notebooks render as editable cells, and unrenderable binary returns a metadata notice instead of decoded byte garbage.
Fixed pasted image-file paths that cannot be loaded to fall back to normal text paste with status feedback instead of disappearing.
Fixed tool-output file paths not being clickable OSC 8 file:// hyperlinks in several renderers. read titles for plain text and image files (the common case) emitted no link at all because the renderer only linked when a resolvedPath was recorded — which the ordinary file/image read paths never set, keeping the absolute path only in meta.source; the renderer now falls back to that source path. write headers were never wrapped in a hyperlink and now link to the absolute path written (file, archive entry, SQLite, and conflict resolutions). edit/apply_patch headers wrapped the model-supplied (often cwd-relative) argument path, producing a root-anchored file:///rel/path URI; they now link the absolute details.path instead. Finally, search, ast_grep, and ast_edit produced doubled link targets (/proj/src/src/file.ts) for searches scoped to a subdirectory, because the renderer resolved the cwd-relative display paths against the scope directory rather than cwd — the scoped-search base is now the session cwd (with the scoped file's absolute path still seeding single-file body lines).
Fixed omp dry-balance --bench to recover from 401 token failures by re-minting the failing OAuth credential in place before switching accounts
Fixed the bash tool corrupting commands that embed multi-byte UTF-8 (e.g. ✓/× inside a grep -E pattern) ahead of a trailing | head/| tail. The bash.stripTrailingHeadTail rewrite cut at char-offset positions reported by brush-parser while slicing the command by byte offset, so the trailing-pipe strip landed mid-pattern and dropped the closing quote — turning … |✓|×|XCTAssert" | tail -80 into … |✓|×-80 and making execution fail with pi-natives:command: unterminated double quote. Fixed in pi_shell::fixup (@oh-my-pi/pi-natives).
Fixed omp dry-balance --bench to recover from 401 token failures by re-minting the failing OAuth credential in place before switching accounts
Fixed duplicate file entries in grouped outputs for find, search, ast_grep, ast_edit, and lsp diagnostics when the same path appeared multiple times
Fixed search, grep, and edit output rendering so repeated directory group blank-line boundaries no longer break nested path/link reconstruction
Fixed omp dry-balance --bench flooding the terminal with staircased, duplicated spinner/status lines (and an indented summary) when the tty has ONLCR/OPOST disabled (raw mode). The interactive progress region separated rows with a bare LF and repositioned with a column-preserving \x1b[<n>A cursor-up, both of which only land at column 0 when the terminal translates LF→CRLF; with that translation off, every 80 ms redraw cascaded down and to the right into scrollback. The live region now carriage-returns before every cleared row, terminates each row with CRLF, and caps each row to the terminal width so a wrapped line cannot desync the cursor-up from the logical line count.
Fixed inconsistent vertical spacing between transcript blocks: some blocks (tool results from search/find and other renderer-backed tools) rendered with a doubled gap (a leading Spacer plus the content box's own paddingY), while others (the grouped read card, file-mention lists, IRC cards) rendered with no gap at all. Vertical spacing is now owned entirely by the chat renderer: TranscriptContainer strips each block's plain-blank top/bottom edges and inserts exactly one blank line between consecutive blocks, so every block is separated by a single consistent gap regardless of which component produced it. Individual components (assistant/user/tool/read-group/bash/eval/skill/custom/hook/compaction/branch/todo-reminder/plan-review messages) no longer emit their own leading Spacer/paddingY for separation, and multi-row groups (IRC cards, file-mention lists, completed-job batches, and the bordered command//changelog//context/version/OAuth/debug panels) are wrapped as single TranscriptBlock children so the renderer spaces them as one unit. Background-colored box padding is preserved as block-internal design.
Fixed resolve with action: "discard" surfacing a hard isError "No pending action to resolve" failure to the model when the agent asked to cancel a staged action (e.g. an ast_edit preview) but nothing was pending. A discard is a request to reach the "no staged change" end-state, which already holds in that case, so it is now honored as a successful cancellation ("Nothing to discard; no pending action remains." with details.action: "discard") instead of an error. action: "apply" with no pending action still errors.
Fixed the collapsed tool-output expand hint rendering double brackets (e.g. ((Ctrl+O for more))) — the EXPAND_HINT text already carried its own parentheses and then formatExpandHint wrapped it again with the theme's bracket glyphs. The hint now resolves the key actually bound to app.tools.expand at render time and reads ⟨<key>: Expand⟩ (e.g. ⟨Ctrl+O: Expand⟩), so a single bracket pair surrounds it and a user remap of the expand keybinding is reflected instead of a hard-coded Ctrl+O.
Fixed the edit/apply_patch tool dropping its outlined frame while streaming/in-progress (only the final result was framed); the in-progress diff preview now renders inside the same muted frame as the completed result.
Fixed the todo and job tools rendering a success icon and success styling on a failed/error result; error results now show the error icon and a red frame border.
Fixed debug tool refusing every dlv launch on Go modules. The launch handler ran validateLaunchProgram before adapter selection and rejected any directory program with launch program resolves to a directory, while dlv's default mode=debug requires a Go package path (a directory or .go source file). Adapter resolution now precedes validation, directory programs prefer adapters that advertise acceptsDirectoryProgram before falling back to native extensionless debuggers, the rejection only fires when the resolved adapter does not advertise that flag (set on dlv in dap/defaults.json), and dlv's mode is derived from the program shape — directories and .go files launch as mode=debug, other files as mode=exec — so omp can debug both Go packages and pre-built binaries (#2020).

[15.10.0] - 2026-06-06

Breaking Changes

Replaced the providers.parallelFetch boolean setting with the providers.fetch enum (auto / native / trafilatura / lynx / parallel / jina) that selects the URL reader-backend priority for the read/fetch tool, mirroring providers.image/providers.webSearch. Existing configs are migrated automatically: the legacy key is dropped and the new auto default applies.

Added

Added a GitHub Actions read handler to the read/web-fetch GitHub scraper. Fetching github.com/{owner}/{repo}/actions/runs/{id} renders the run metadata plus a per-job breakdown (steps listed for any job that did not succeed), and …/actions/runs/{id}/job/{id} (also the API-style …/jobs/{id}) renders a single job's metadata, step table, and full plain-text logs. Logs are fetched via the actions/jobs/{id}/logs redirect using GITHUB_TOKEN/GH_TOKEN when present, with the per-line ISO timestamp prefix and leading BOM stripped; the section degrades to an explicit notice when logs are unavailable (no token, private repo, or expired/unfinalized run).

Changed

Changed eval agent() subagents so they are never subject to the task.maxRuntimeMs wall-clock cap. The parent cell's idle watchdog is already suspended for the entire bridge call (withBridgeTimeoutPause), so a long-running fan-out/recovery workflow must not be killed by a per-subagent runtime limit. runEvalAgent now passes maxRuntimeMs: 0 to runSubprocess, which honors an explicit ExecutorOptions.maxRuntimeMs override over the inherited setting.
Changed interactive timing behavior so PI_TIMING=x pi preloads the module timer before the CLI graph loads and includes the (modules) report. PI_TIMING=full now also exits after printing, matching PI_TIMING=x, so full module reports are usable for cold-start measurement without launching the TUI. Added the root dev:timing script for the same profiled startup path.
Changed coding-agent startup imports so normal TUI launch imports InteractiveMode directly, keeps print/RPC/ACP runners on their branch-only paths, and moves marketplace auto-update work behind a lightweight deferred starter.
Changed cold-launch setup gating so the full setup wizard (every scene plus the overlay and their TUI/OAuth/web-search/theme dependencies) is no longer statically imported by main.ts. The current setup version now lives in a tiny dependency-free modes/setup-version module, and the wizard barrel is lazy-loaded only when the stored setup version is stale or the wizard is forced — the common up-to-date launch skips loading it entirely.
Changed cold-launch startup imports so the hot-path CLI files no longer pull the full @oh-my-pi/pi-ai barrel: commands/launch.ts and cli/args.ts import THINKING_EFFORTS/Effort from the tiny @oh-my-pi/pi-ai/effort module, and config/model-registry.ts now imports its ~20 symbols from narrow subpaths (api-registry, model-cache, model-manager, model-thinking, models, provider-models, types, utils/event-stream) instead of the barrel — so launching no longer eagerly loads every provider, auth, OAuth, and usage module re-exported by the barrel.
Changed the read/fetch HTML reader-backend priority to native > trafilatura > lynx > parallel > jina (was parallel > jina > trafilatura > lynx > native). The in-process native htmlToMarkdown runs first — instant, no network, full-fidelity — so the common case no longer depends on a remote service, and a stalled remote backend can no longer mask it. Selecting a specific backend via providers.fetch tries it first, then the rest fall back. The low-quality gate (>100 chars and not isLowQualityOutput) now applies uniformly to every backend; when none clears it, the highest-priority substantial-but-low-quality output is still surfaced so the llms.txt / document-extraction fallbacks keep running.

Fixed

Fixed eval agent() failures surfacing as an opaque RuntimeError: bridge call '__agent__' failed with no reason. When a subagent aborted, runEvalAgent built its failure message with result.error ?? result.stderr ?? result.abortReason ?? …, but result.stderr is the empty string on a clean abort (and result.error is gated on a non-empty stderr), so the nullish chain stopped at "" and never reached abortReason. The empty string propagated through the loopback bridge and the Python prelude's RuntimeError(msg or "bridge call … failed"), discarding the real reason. The chain now uses || so an empty stderr falls through to abortReason.
Fixed subagent aborts being mislabeled as the generic "Cancelled by caller" when the abort originated inside the subagent's own turn (stopReason: "aborted" with no caller signal and no runtime-limit timer). runSubprocess now prefers the aborted assistant message's errorMessage (e.g. "Request was aborted" or a specific stream error) for that case, while a real caller signal or wall-clock abort still reports its precise reason.
Fixed a long streaming tool preview that alone overflows the viewport dropping its scrolled-off head on ED3-risk terminals (ghostty/kitty/iTerm2/…). When expanded with Ctrl+O, a streaming write (content streaming in) and a streaming eval (stdout streaming below its fixed code cell) render top-anchored and grow append-only, but the tool block never reported itself append-only to the transcript, so the renderer's commit-as-you-go boundary stopped at the block start and the earlier rows that scrolled above the viewport were committed nowhere — they vanished, leaving the preview looking like a viewport-tall circular buffer. ToolExecutionComponent now implements isTranscriptBlockAppendOnly() (gated on isTranscriptBlockFinalized(), so it also covers partial-result streams like eval), delegating to a renderer-declared isStreamingPreviewAppendOnly predicate so the expanded stream commits its head exactly like a streamed assistant reply. Collapsed previews (bounded sliding tail windows) and finalized/result previews (which can collapse to a capped view) stay deferred.
Fixed read/fetch silently dropping whole list sections on pages with malformed list markup — stray <img>, text, or <video> nodes as direct children of <ul> (e.g. the Alacritty changelog). The Jina reader (previously tried before the local renderers) mis-extracts such lists, returning empty Added/Changed sections; the native in-process renderer now runs first and preserves the full content.
Fixed /usage aggregate amount fallback using raw limits.length as account count — now counts unique accountId values from limit scopes, so N limits from a single account no longer display as "N accts".
Fixed /usage account labeling falling back to "account N" for providers that use projectId as their primary identity (e.g. Google Antigravity, Gemini CLI) — projectId from report metadata is now considered before the generic fallback.
Fixed the todo tool's TUI renderer crashing with TypeError: args?.ops?.map is not a function when a streaming tool-call delta surfaced a non-array ops field (mid-stream parseStreamingJson shapes like { ops: "[{" }, or [null] entries before fields arrive). The renderer now treats non-array ops, non-object entries, and non-array items as missing structure instead of crashing, which also stops the spam-warn cascade that followed each malformed delta. Paired with the Anthropic-side reasoning-replay fix in packages/ai (#2005).
Fixed Python eval agent() collapsing subagent runtime-limit aborts (and other empty-stderr aborts) into a generic RuntimeError: bridge call '__agent__' failed. runEvalAgent coalesced the failure message with ??, which stopped at the empty stderr and never reached abortReason, shipping an empty error through the loopback bridge. The bridge now prefers abortReason for aborts and trims empty stderr/error out of the fallback chain, so Python surfaces the actionable reason (e.g. Subagent runtime limit exceeded (task.maxRuntimeMs=900000)) (#2006).

[15.9.69] - 2026-06-06

Added

Added anonymous fallback for Perplexity web search, allowing web_search and explicit Perplexity provider usage when no Perplexity credentials are configured
Added gallery CLI command to render built-in tool renderer output across streaming, in-progress, success, and failure states
Added omp gallery filtering and rendering options (--tool, --state, --width, --expanded, and --plain) for focused renderer previews and plain-text output
Added omp gallery fidelity for tools whose renderers are attached on the tool instance (lsp, task): the gallery now drives them through the same custom-tool render branch production uses, so regressions in that path surface in the gallery rather than only in a live session.
Added omp gallery --screenshot, which renders the gallery through a real virtual terminal (VHS) and writes PNG screenshot(s) instead of ANSI, so agents (and anything that can only read raw bytes) can actually see the rendered output. The capture forces truecolor and matches the active theme/symbol preset; tall galleries split across multiple images (whole renderers are never cut). Tune with --out, --font, and --font-size; requires vhs on PATH and fails with install guidance when absent.
Added app.display.reset, bound to Ctrl+L by default, to force an immediate terminal display reset/redraw without resizing the window.

Changed

Changed authenticated Perplexity ask requests to default to the experimental model preference, matching the anonymous fallback.
Changed Perplexity explicit provider availability checks so the setup wizard can mark perplexity as available for manual selection without credentials, while auto provider discovery still requires auth
Changed the default persistent model selector shortcut from Ctrl+L to Alt+M, leaving Ctrl+L for display reset. Existing user remaps in keybindings.yml are preserved.
Changed the edit tool result header to carry the diff change stats (+N / -M / K hunks) inline next to the file path, and removed the redundant lone language-icon metadata row and the blank line between the header and the diff body, so a single-hunk edit renders as ✔ Edit: <icon> path:LINE ⟨+3 / 1 hunk⟩ immediately followed by the diff.
Changed the web_search tool result rendering: the answer text now shows in full instead of being truncated to a "… N more lines" preview (the omp q CLI still caps its compact output), each source renders as a single title (domain) · age line with the URL linked on the title (dropping the snippet and bare-URL rows), and the metadata block collapses to one Provider: <model> @ <provider> (<auth>) line plus Usage: (removing the redundant Sources/Citations/Request/Queries rows).

Fixed

Fixed Perplexity perplexity_ask response parsing so OAuth, cookie, and anonymous searches correctly extract answer text and sources from JSON payloads
Fixed framed read results rendering with an extra blank row above and below the output block.
Fixed collapsed search result previews that could show only "… N more matches" when the first grouped section exceeded the preview budget. Collapsed search output now compacts to match rows, fills the budget with visible hits before the summary, and keeps truncation details out of the bottom user-visible notice.
Fixed the expanded search result view dumping every match when all hits live in one file. A single file's matches collapse into one blank-line group, and the expanded tree list ignored the line budget, so a hot file whose matches span its whole length rendered every row (e.g. lines 374–2858). Expanded search output is now bounded by a larger-than-collapsed budget (EXPANDED_LINES × 2), keeps surrounding context rows, and appends a … N more matches summary when truncated.
Fixed boolean environment flag overrides that were ORed with settings, so PI_INTENT_TRACING=0, PI_AUTO_QA=0, and per-backend eval flags now take precedence when present while falling back to config when unset.
Fixed custom-rendered tools that set mergeCallAndResult (e.g. lsp) rendering a redundant tool-name line above the framed result once a result arrived. ToolExecutionComponent's custom-tool branch now emits the fallback label only when the tool has no renderCall and the call is not suppressed by an existing result, matching the built-in renderer branch.
Fixed the write tool result rendering with a green success checkmark even when the write failed. writeToolRenderer.renderResult now branches on result.isError, rendering the error status icon plus the failure message instead of the success header and content preview.
Fixed Perplexity OAuth/cookie web search returning a refusal answer ("I don't currently have access to the web-search tools in this turn") despite returning real sources. callPerplexityOAuth was prepending the API-style web-search system prompt to the query (query_str = systemPrompt + "\n\n" + query), but the consumer www.perplexity.ai/rest/sse/perplexity_ask endpoint has no system-message slot and reads the prepended instruction as a meta-prompt, making the model decline. The OAuth/cookie path now sends the bare query; the API-key path still passes the system prompt as a proper system message.
Fixed Perplexity OAuth web search always returning the free turbo model instead of the account's Pro model (e.g. pplx_pro_upgraded/Sonar). The www.perplexity.ai/rest/sse/perplexity_ask endpoint authenticates via the __Secure-next-auth.session-token cookie and ignores the Authorization: Bearer header entirely — so sending the OAuth session token as a bearer was treated as an anonymous request, which silently downgrades to turbo regardless of model_preference. The stored Perplexity OAuth token is itself the next-auth session JWT (the macOS app injects the same value as that cookie), so callPerplexityAsk now sends it as the __Secure-next-auth.session-token cookie, unlocking Pro model selection.

[15.9.67] - 2026-06-06

Added

Added timeout-pause and timeout-resume eval bridge status events emitted around agent()/llm() operations
Added a /copy picker: /copy now opens a fullscreen, outlined tree of recent assistant messages with their code blocks nested beneath (like /tree). Navigate with ↑↓, and Enter copies the highlighted node — a whole message, an individual code block, "All N blocks", or a bash/eval command interleaved with the assistant turn that issued it. A live preview pane shows the selected target, wrapping prose and syntax-highlighting code/commands.

Changed

Changed eval timeout accounting so delegated bridge calls now suspend the cell watchdog and start a fresh timeout window when runtime control returns
Changed IdleTimeout to support reference-counted pauses so overlapping delegated bridge calls keep timeout paused until all calls complete
Changed the default app.message.followUp binding from Ctrl+Enter alone to [Ctrl+Q, Ctrl+Enter] so the follow-up shortcut works in Windows Terminal, which does not deliver a distinct Ctrl+Enter event to console apps. Ctrl+Q mirrors the GitHub Copilot CLI default for the same action; existing remaps in ~/.omp/agent/keybindings.yml are untouched, and if another user-remapped action already claims Ctrl+Q, that user binding wins while follow-up keeps Ctrl+Enter. Ctrl+Q is also reserved by ExtensionRunner so an extension cannot register that chord and be silently overwritten by the built-in follow-up handler (#1903).
Changed all scrollable TUI pickers and viewports to render through the shared ScrollView right-edge scrollbar for a uniform look, replacing their ad-hoc (N/M) / [a-b/total] text indicators (search hints and the tree filter-mode label are preserved). Covers the session/resume picker, model selector, OAuth provider selector, history search, session tree selector, agent dashboard list, extension list, user-message selector, the raw SSE debug viewer, the autoresearch dashboard overlay, and the session observer overlay.
Changed the /model and /switch selectors to dim and skip models whose context windows are smaller than the current chat context.
Changed /copy command targets to appear inline with recent assistant messages instead of as a separate "Last bash command" row at the end of the picker.

Fixed

Fixed the idle Working... loader freezing on ED3-risk terminals with unobservable native scrollback by keeping foreground live-region rendering enabled from agent_start until agent_end, before the first assistant or tool event arrives.
Fixed framed tool output blocks rendering one column inset inside tool boxes; modern bordered blocks now span the same width as legacy background-filled tool boxes.
Fixed potential TimeoutError aborts for short timeout eval cells during long bridged agent()/llm() work where no progress events are emitted until completion
Fixed retry recovery to allow automatic retries without switching models when retry.modelFallback is disabled.
Fixed ttsr.enabled: false being ignored at runtime. TTSR rules were still being registered with TtsrManager.addRule and matched against stream deltas even when the global toggle was off, so disabling TTSR did not suppress rule injection or stream abort. The manager now gates addRule, hasRules, and #matchBuffer on the enabled flag, so disabling fully short-circuits the TTSR path. Condition rules fall through to the rulebook bucket instead of being silently swallowed. (#1767)
Fixed the Python eval kernel hanging on Windows during import pandas / import numpy, with SIGINT unable to recover the cell. PythonKernel.start() spawned the runner with windowsHide: true, which in Bun maps to the Win32 CREATE_NO_WINDOW flag and detaches the long-lived child from any inherited console — so native extensions like numpy/_core/_multiarray_umath.pyd (and its bundled OpenBLAS/SLEEF thread-pool init) could deadlock inside LoadLibraryExW, and GenerateConsoleCtrlEvent-based SIGINT delivery silently became a no-op. The kernel now hides its window only when the host itself has no console to share (service / piped launch); an interactive TUI launch lets the kernel inherit the parent's console, matching the behavior of python.exe invoked from cmd.exe (#1960).
Fixed task renderer crashing the TUI with TypeError: completeData?.map is not a function when a subagent's extractedToolData.yield slot held a non-array value. renderAgentResult (and the live-progress sibling) cast the slot to Array<{ data }> and called ?.map, but optional chaining short-circuits only on null/undefined, so a plain object made .map undefined and threw — taking down every review task render. Both sites now go through normalizeYieldData, which wraps a single object as a 1-element array and drops primitives (#1987)
Fixed sdk-async-job-manager-singleton tests flaking under the full parallel suite. The four createAgentSession-based cases ran on the default 5000ms per-test timeout, which two real session startups can exceed when test:ts saturates the machine across packages; on timeout the still-running test body and afterEach reset raced, surfacing a spurious "Unhandled error between tests" on the AsyncJobManager.instance() assertion. They now carry an explicit 60000ms timeout, matching the convention used by the other session-creating tests in this suite.
Fixed streaming eval, bash, ssh, and task call previews overflowing the live transcript viewport and cutting off their top while pending. A volatile tool block taller than the viewport could strand its scrolled-off head out of native scrollback on ED3-risk terminals (committed nowhere, repainted nowhere) until the result landed. The pending eval source preview now follows the streaming edge in a bounded 12-line tail window (newest lines pinned to the bottom, "… N earlier lines" on top) so you can watch the code being written without the box overflowing; bash/ssh commands and task context use a bounded head+tail window. Ctrl+O still lifts the cap for a full view.
Fixed the streaming write call preview ignoring Ctrl+O so the expand toggle was a no-op while a file was being written. Unlike the eval/bash/ssh/task streaming previews, formatStreamingContent never received the expanded flag, leaving the preview pinned to a bounded 12-line tail window even after pressing Ctrl+O — so on a large write you could not widen past the streaming edge until the tool result landed. The preview now lifts the cap to the full file (head through tail) when expanded, matching the documented streaming-preview behavior of the other tools.
Fixed turn-ending provider errors rendering twice — once as the transcript's inline Error: … line and again in the pinned banner above the editor (added in 15.9.5). The inline line is now suppressed while the same error is mirrored in the banner and restored to the transcript when the banner clears at the next turn, so the error stays in history without the duplicate render at the error moment.

Removed

Removed the /copy last|code|all|cmd subcommands; every copy target is now reachable by picking it in the /copy tree.

[15.9.5] - 2026-06-05

Added

Added a persistent error banner pinned above the editor when an assistant turn ends on a provider error (e.g. Anthropic's "Output blocked by content filtering policy"). The transcript Error: … line scrolls away as the conversation grows, so terminal turns that ended on a stream error could pass unnoticed; the banner stays in the fixed region above the input and is cleared when the next turn starts.
Added bold, underlined, clickable [Image #N] placeholders in the draft editor and sent user-message bubbles, backed by extension-bearing blob-store sidecar files so terminal file:// links open in image viewers.
Added the active model identifier (provider/id) to the system prompt's <workstation> block so the agent knows which model it is running as. Gated by the new includeModelInPrompt setting (default on); the base prompt is rebuilt on a mid-session model switch so the surfaced identifier stays current.
Added OLLAMA_HOST support for implicit local Ollama discovery when OLLAMA_BASE_URL is unset, so OMP picks up the same host setting used by Ollama.
Added OLLAMA_CONTEXT_LENGTH as a positive-integer context-window override for implicit local Ollama discovery, so users can correct OMP context budgeting without writing per-model overrides.

Changed

Changed tools.discoveryMode to default to auto, which keeps discovery off for small tool sets and automatically switches to MCP-only tool discovery when more than 40 tools are registered.

Fixed

Fixed user-message rendering to materialize image links from embedded image blocks when rebuilding chat output, so image placeholders remain clickable after replayed or restored messages
Fixed queued/steering user messages carrying a pasted image rendering out of order — sometimes dropping the user bubble below the very tool output it was sent to steer. EventController.#handleMessageStart awaited async image-link materialization between the user message_start and addMessageToChat; since AgentSession.#emit dispatches TUI listeners fire-and-forget, that mid-handler yield let the next synchronously-handled events (assistant message_start, tool execution start/end) append their components first, scrambling transcript order and live-region block boundaries. The bubble is now appended synchronously, with clickable image links still materialized via the synchronous blob-store fallback.
Fixed tool execution cards to finalize promptly when a turn is abandoned or completed so stale streaming previews and frozen spinner frames no longer keep transcript rows in the live region
Fixed read and search TUI rendering to emit OSC 8 hyperlinks for HTTP URLs, local:// resources backed by files, and filesystem search targets, including line-specific links for search match rows.
Fixed aborted streaming assistant messages staying frozen before their red "Operation aborted" label when status rows were appended underneath on ED3-risk terminals.
Fixed omp / omp -c stacking a fresh welcome screen and transcript on top of the previous run's leftover terminal scrollback. The cold-launch transcript render was the only session-load path that did not pass clearTerminalHistory, so the TUI's scrollback-preserving initial paint left the prior run's welcome + conversation above the new one; the cold launch now clears native scrollback before painting, matching every in-process session switch.
Fixed a long streamed assistant reply dropping its earlier lines on ED3-risk terminals (Ghostty/kitty/iTerm2) once it grew past the viewport — the head scrolled off the top and never reached scrollback, so the reply rendered as a ~viewport-tall circular buffer of only its latest lines. AssistantMessageComponent now reports itself as an append-only transcript block and TranscriptContainer surfaces the resulting commit-safe boundary, so the renderer commits the scrolled-off head to native scrollback instead of discarding it (volatile tool previews stay deferred as before).

Security

Blocked OSC 8 hyperlink wrapping for URI targets containing terminal control bytes to avoid rendering malformed control-sequence links

[15.9.4] - 2026-06-05

Fixed

Fixed chat transcript updates after submitting input so frozen scrollback is only thawed when native scrollback replay succeeds, preventing misplaced or duplicated rows when the viewport is not at the tail
Fixed read of .zip archives to list the central directory without inflating every member, so large or corrupt zip payloads no longer freeze directory reads; member contents are inflated only when a specific entry is read.
Fixed the Python eval kernel being hard-killed (and its persistent session state lost) when a cell blocked in parallel() / agent() was interrupted. Each agent()/tool.* call blocks a kernel worker thread in a synchronous urllib request to the host bridge, and parallel()'s ThreadPoolExecutor exit joins those threads — so the kernel cannot unwind a KeyboardInterrupt until every in-flight bridge call returns. A wide subagent fan-out's teardown routinely outlasted the kernel's 5s SIGINT-escalation window, so the kernel was force-killed (surfacing [kernel] Python kernel shutdown) while the subagents were still winding down. The host bridge now resolves an in-flight call the instant the cell's signal aborts, so the kernel unwinds cleanly and keeps its state; the already-signaled subagent continues tearing down in the background.
Fixed github tool run_watch op ignoring the explicit repo argument and silently watching the cwd-inferred repository in nested/umbrella workspaces. executeRunWatch passed undefined for the user-supplied repo to resolveGitHubRepo, so a call like {op: "run_watch", repo: "owner/cxf", branch: "main"} fell back to gh repo view in cwd and streamed watching <sha> on <cwd-repo> against the wrong repository. The explicit repo now takes precedence over the cwd inference, and the no-branch/no-run path refuses to derive the watched commit from git HEAD unless the cwd actually points at the resolved repo — otherwise it raises a ToolError telling the caller to pass branch or run instead of silently rebinding to an unrelated commit (#1949).
Fixed omp plugin install <local-path> failing with Invalid package name: . (and similar) for cwd-relative (., ./pkg), absolute (/abs, C:\…, \\unc), and tilde-prefixed (~/pkg) specs. classifyInstallTarget now returns a local arm in addition to marketplace/npm, and plugin install routes those specs to PluginManager.link() — the same code path as omp plugin link. (#1945)
Fixed the web_search result renderer capping the synthesized answer at 12 lines even when expanded, while the Sources list expanded in full — so a long answer stayed truncated ("… N more lines") after Ctrl+O, making the expand toggle look like a no-op and dwarfing the answer next to its sources. The answer now renders the full text (markdown-formatted) when expanded and a short markdown preview when collapsed, matching the sources' collapse/expand behavior. The answer is also rendered through the Markdown component instead of dimmed raw lines, so headings, bold, lists, and code in the answer display formatted.

[15.9.3] - 2026-06-05

Fixed

Fixed @-mention auto-read injecting an unrelated, same-named file when a mention did not point at a real path — e.g. an npm scope like @scope/, a partial path, or a bare token. generateFileMentionMessages resolution previously fell back to prefix and repo-wide fuzzy matching (globbing the whole project on every such mention) and auto-read the single "best" guess. Resolution is now exact-only: a mention is auto-read only when it resolves to an existing file or directory; otherwise it is left as prose. The TUI @-selector already inserts the real, complete path before send, so post-send guessing was both unnecessary and the source of the wrong-file reads. Directories still resolve and are listed. Removes the per-mention **/* project scan.

[15.9.2] - 2026-06-05

Added

Added an encrypted local auth-broker snapshot cache for discoverAuthStorage, with OMP_AUTH_BROKER_SNAPSHOT_TTL_MS and OMP_AUTH_BROKER_SNAPSHOT_CACHE, so fresh cached broker credentials can boot without a blocking /v1/snapshot fetch and survive broker-down startup windows.
Added dry-balance CLI command to perform a dry-run OAuth account balancing check across configurable random session IDs, with sample and concurrency options, JSON output, and success/failure summary reporting
Added --json output mode and machine-readable result format to omp dry-balance for automated use
Added omitMaxOutputTokens to models.yml model definitions and modelOverrides, so users can opt a model out of the on-the-wire max_output_tokens / max_tokens cap while keeping the catalog maxTokens for local budgeting. Intended for Ollama-style proxies whose upstream output limit OMP cannot discover. (#1881)

Fixed

Fixed TTSR rule-violation injections leaking the absolute home directory to the model: the ttsr-interrupt / ttsr-tool-reminder blocks rendered the matched rule's path as its absolute on-disk path (e.g. /Users/me/Projects/app/.omp/rules/no-any.md). The path is now relativized to the session cwd when the rule lives in the project (.omp/rules/no-any.md), or ~-relative when it lives under home, so no absolute path is fed into the agent's context outside the system prompt.
Fixed AsyncJobManager.instance() being cleared while the owning top-level session was still live, which broke the task async path with "Async execution is enabled but no async job manager is available" until process restart. Any in-process secondary top-level createAgentSession() call (e.g. the Agent Control Center's create flow in agent-dashboard.ts) constructed a fresh AsyncJobManager, overwrote the singleton, and then cleared it on its own dispose. Secondary sessions now leave the live singleton untouched, and their dispose-time cleanup is scoped so it can no longer cancel the primary session's running bash/task jobs. bash / task / job tools and session job snapshots now resolve the manager through session-scoped async manager wiring rather than AsyncJobManager.instance(), so a secondary in-process top-level session cannot accidentally register background work on the owning session's manager or report the owning session's jobs; subagents still inherit the parent's manager via their scoped async manager. Startup failures after a top-level session installs its manager now clear and dispose that manager before the next session decides whether it can create its own (#1923).
Fixed the task tool returning a hard Async execution is enabled but no async job manager is available. error when async.enabled was true but AsyncJobManager.instance() returned undefined, leaving task non-functional for the rest of the session. The tool now falls back to the existing synchronous execution path (which still runs subagents concurrently via mapWithConcurrencyLimit), and logs a warning so the missing-manager state stays diagnosable (#1922).
Fixed Hindsight retain/recall/reflect calls staying pinned to the bank that was selected when the session started after the operator edited hindsight.bankId, hindsight.bankIdPrefix, or hindsight.scoping mid-session. The backend now subscribes to those settings via onHindsightScopeChanged and rebuilds the active HindsightSessionState against the recomputed scope, disposing the old state after flushing its queue so in-flight tool-initiated retains still land in the bank they were enqueued for. Also renamed ensureBankMission to ensureBankExists so a blank bankMission no longer skips bank creation entirely, and called it before mental-model bootstrap so createMentalModel is never the first POST against a missing bank. AgentSession.dispose now flushes the retain queue before clearing #hindsightSessionState, since the queue's identity guard would otherwise drop the spliced batch (#1902).
Fixed /tree rendering a bare "No entries found" line on a fresh session where the only persisted entries are the model_change + thinking_level_change written by sdk.ts at startup — both are hidden by the tree-selector's default filter, so #filteredNodes.length === 0 while tree.length === 2 and the controller's tree.length === 0 short-circuit never fired. The selector now splits the empty-state into three distinct shapes — truly empty tree, search query with no matches, and filter mode rejecting every entry — surfacing the cause and the recovery key (Alt+A to show all, Backspace to clear a stale search) so users on a fresh session can see immediately that the panel isn't broken (#1909).
Fixed remote MCP OAuth refresh failures leaving stale credentials in agent.db: when the token endpoint returns a definitive failure (invalid_grant, invalid_token, revoked, plain 401/403 not classified as transient), MCPManager#resolveAuthConfig now drops the credential via AuthStorage.remove(credentialId) and skips re-attaching the dead Authorization: Bearer … header. Previously a revoked refresh token kept producing 401 invalid_token on every MCP request and survived restarts, so users had to hand-clear the credential row to recover; the next connect now surfaces a clean auth error and /mcp reauth <server> (or /mcp unauth) recovers without restarting. Transient refresh failures (network/fetch failed/ECONNREFUSED) still fall back to the existing access token (#1908).
Fixed omp://docs and omp://docs/... internal documentation URLs in the distributed package to resolve through the embedded documentation index instead of failing with Documentation file not found (#1898).
Fixed the github discovery provider silently ignoring .github/skills/<name>/SKILL.md, GitHub's documented Agent Skills layout. The provider now registers a skills capability (priority 30, project-only) that scans .github/skills/ non-recursively via scanSkillsFromDir with requireDescription: true, matching the Agent Skills spec and the sibling native/omp-plugins providers (#1906).
Fixed inline images rendering as a wall of empty PUA box glyphs with laggy scrolling on Kitty-protocol terminals that do not honor Unicode placeholders (most notably WezTerm and tmux/screen passthrough to a non-Kitty outer terminal). The 15.9 placeholder rollout enabled the U=1/U+10EEEE grid for every Kitty-protocol path; it now defaults on only for kitty and ghostty, with PI_NO_KITTY_PLACEHOLDERS=1 as a hard opt-out and PI_KITTY_PLACEHOLDERS=1 as opt-in for terminals (e.g. wezterm nightlies) that have since added support (#1877).
Fixed auto session-title generation failures being swallowed without an actionable diagnostic. Title generation now logs structured start, missing-model/API-key, provider-error, empty-result, and exception outcomes with the session id and resolved title model; the interactive auto-title caller also logs uncaught persistence/generation errors instead of dropping them. (#1892)
Fixed TranscriptContainer reporting the live block boundary to the TUI again, so ED3-risk foreground streaming can append newly sealed transcript blocks to native scrollback once while deferring only the active live block.

[15.9.1] - 2026-06-04

Added

Added deferred session-title generation so greetings no longer become the session title. A first user message that is only a greeting / acknowledgement / filler ("hi", "thanks", "ok", a bare number, emoji-only, etc.) is now detected deterministically and skips titling entirely — no title model is invoked. Title generation then retries on each subsequent user message while the session stays unnamed, so the title is deduced from the first message that actually describes work. A capable online title model may additionally answer none to decline a non-greeting taskless message (normalized to "no title").

Changed

Changed mid-turn user steers to reach the model inside a wire-only interjection envelope, while transcripts and persisted session history keep the user's original text.
Changed the system prompt to treat user requests for parallel work as task subagent fan-out rather than parallel tool calls.
Changed the Agent Control Center's new-agent description field to use the multiline TUI editor, with Enter inserting lines and Ctrl+Enter generating the spec.
Changed the Agent Control Center and Extension Control Center to accept Left/Right arrow keys for switching tabs (source / provider), in addition to Tab / Shift+Tab — matching the model and settings selectors, whose TabBar already supported arrow navigation.
Refreshed the Ctrl+R history search overlay: the selected row now renders as a full-width selectedBg highlight bar, matched query tokens are highlighted in the accent color, each result shows a right-aligned relative timestamp, and the panel gained an icon'd accent title plus a two-tone keyhint footer. The selector also gained PageUp/PageDown (via the configurable tui.select.pageUp/pageDown keybindings) and Home/End navigation.
Changed Perplexity API-key web search to return more comprehensive results: web_search_options.search_context_size is now high (was medium) for maximum retrieval grounding, the default num_search_results is 20 (was 10) so twice as many sources are surfaced, and return_related_questions is enabled with the response's related_questions now parsed into relatedQuestions (previously dropped). On an identical query this lifted the result from 10 sources / ~410 output tokens to 20 sources / ~1900 output tokens with a structured, multi-section answer; latency tracks model output length, not context size, so the 60s hard timeout headroom is unchanged.

Fixed

Fixed a streamed assistant message freezing at a partial prefix (e.g. only "Nat" of "Natives built, now…") on ED3-risk terminals (Ghostty/kitty/iTerm2/Alacritty), with the final text appearing only after a resize. TranscriptContainer freezes each non-live block by replaying its last live render, but render coalescing can finalize a block's content and append the next block within the same throttled frame — so the block was sealed at its stale mid-stream snapshot and never repainted until the next thaw. The block that was live on the previous render is now recomputed once on the live→frozen transition, sealing it at its final content.
Fixed ACP/RPC stdio startup so protocol frames are no longer consumed as one-shot piped prompt input before the JSON-RPC transport starts.
Fixed omp completions to await the completion script write before exiting.
Fixed AssistantMessageComponent exposing its stable-prefix completion API again so streamed assistant messages remain unstable until explicitly completed.
Fixed session restoration to ignore transient fallback model switches (such as automatic context-promotion or retry fallback) so resumed or resumed-switch sessions revert to the configured default model unless the last change was a user-selected temporary model
Fixed in-session /resume to restore both the last user-selected temporary model and persisted plan/goal mode state instead of falling back to the default model with plan mode off.
Fixed the /resume session picker overflowing short viewports: the visible window was hardcoded to 5 entries (and assumed 3 lines each), but titled sessions render 4 lines, so on a typical-height terminal the picker's header and search box scrolled off the top and the first entry was hidden until you scrolled the terminal up. The visible-entry count is now derived from the live terminal height (budgeting the worst-case 4-line titled entry plus the picker's chrome), so the whole picker fits the viewport and grows on taller terminals.
Fixed the Agent Control Center and Extension Control Center dashboards overflowing the terminal: they were mounted inline below the chat transcript, so the combined height exceeded the viewport — the tab bar and controls scrolled off the top into native scrollback, and every state change yanked the view back to the bottom. Both dashboards now render as full-screen overlays sized to the live terminal height (process.stdout.rows), re-fit on resize, fill the viewport, and reserve space for the footer keyhints so the controls stay visible.
Fixed Ctrl+R history search results to remain globally sorted by prompt recency after merging FTS prefix matches with substring fallback matches.
Fixed Exa web search with no stored or environment credential to use the public Exa MCP fallback again, preserving the auth storage → EXA_API_KEY → mcp.exa.ai resolution order (#1860).
Fixed ACP plan-mode writes to local://PLAN.md so session-local plan artifacts are written to OMP's local artifact root instead of being routed through the editor writeTextFile bridge, avoiding Zen's Internal error and making the plan readable after creation (#1863).
Fixed ACP plan mode stranding the agent at plan approval: entering mode: "plan" now registers a standing resolve handler so the agent's resolve { action: "apply" } no longer fails with No pending action to resolve. Nothing to apply or discard. The handler validates the plan file, asks the ACP client to confirm via unstable_createElicitation when the client supports forms, renames the approved plan to local://<title>.md, and exits plan mode so the agent regains write tools for execution (#1869).
Fixed provider.appendOnlyContext: "auto" staying inactive for Xiaomi Token Plan/SGLang endpoints, preserving prefix-cache hits without forcing append-only mode globally (#1851).
Fixed models.yml compatibility parsing to preserve compat.cacheControlFormat: "anthropic" for custom OpenAI-compatible Claude proxies. (#1845)
Fixed the TUI's Settings → Plugins panel reporting "No plugins installed" when only marketplace plugins were installed. The panel now merges PluginManager.list() with MarketplaceManager.listInstalledPlugins() — the same data source the /plugins list slash command and omp plugin list CLI already used — and tags each row with an [npm] / [marketplace] kind badge, a scope tag, and a shadow indicator for project-shadowed user installs. Selecting a marketplace row opens a new MarketplacePluginDetailComponent whose single Enabled toggle calls MarketplaceManager.setPluginEnabled(pluginId, enabled, scope), with read-only metadata (version, install path, installed-at, last-updated, git commit SHA) listed below the toggle. The empty-state now lists both install commands (omp plugin install <package> and omp plugin install <name>@<marketplace>) (#1842).
Fixed scoped mnemopi recall in MnemopiSessionState.collectScopedRecallResults/recallResultsScoped to await the async Mnemopi.recallEnhanced so the new auto-derived queryEmbedding flows through. Without this, the embedding-enabled mnemopi backend silently kept running FTS-only on every recall. (#1832)
Fixed the SSH tool renderer inlining multiline remote commands into its single-line status header, which produced a malformed cell where the bordered output block opened mid-command. The renderer now drops the command from the header (which keeps only [host]) and renders the full command in a framed section above Output, mirroring the bash renderer. renderStatusLine also flattens any embedded CR/LF in description, meta, and title so no tool can accidentally expand the header into multiple rows (#1828).
Fixed tsc --noEmit against packages/coding-agent/tsconfig.json reporting 56 errors under TypeScript 5.x (builtin-registry.ts × 46, agent-session-openai-responses-replay.test.ts × 10). The repo's own gate (tsgo / TypeScript 6.x) already accepted the () => void slash-command handlers, but 5.x rejects them because it does not coerce a void-returning function value into a () => T | undefined slot. The SlashCommandSpec.handle / handleTui signatures and the test's createPersistedSession populate callback are now expressed as a union of two function types (one returning a SlashCommandResult / target, one returning void), so the existing handler bodies typecheck on both compilers (#1821).
Fixed omp update leaving @oh-my-pi/pi-natives and the platform-specific @oh-my-pi/pi-natives-<tag> leaf at the previous version on bun install -g updates, so the next launch loaded a stale .node file and aborted at validateLoadedBindings with The .node file on disk is from a different release than this loader. omp update now pins the native addon core and the platform leaf to the same version it installs for @oh-my-pi/pi-coding-agent (#1824).

[15.9.0] - 2026-06-04

Breaking Changes

Removed synchronous readTextSync from SessionStorage and core implementations (MemorySessionStorage, FileSessionStorage, RedisSessionStorage, SqlSessionStorage), requiring callers to use async text reads
Replaced the public SessionStorage readTextPrefix(path, maxBytes) and readTextSuffix(path, maxBytes) methods with readTextSlices(path, prefixBytes, suffixBytes): Promise<[string, string]>; custom session storage backends must implement the new combined slice API.

Added

Added env-driven OpenTelemetry trace export. When OTEL_EXPORTER_OTLP_ENDPOINT (or OTEL_EXPORTER_OTLP_TRACES_ENDPOINT) is set, omp registers a global OTLP/proto trace exporter and switches on the agent loop's telemetry, so the invoke_agent / chat / execute_tool spans actually reach a collector instead of a no-op tracer. Honors the standard OTEL_* env contract (endpoint, headers, OTEL_SERVICE_NAME, OTEL_SDK_DISABLED and OTEL_TRACES_EXPORTER=none parsed case-insensitively) and the OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT capture toggle; it is a no-op when no endpoint is configured. Only the http/protobuf transport is supported — a grpc or http/json OTEL_EXPORTER_OTLP*_PROTOCOL declines rather than misrouting spans. This makes the existing telemetry usable from headless hosts that run omp as a spawned child process, where an in-process TracerProvider registered by the parent can't reach the child. Uses the @opentelemetry/exporter-trace-otlp-proto 2.x line, which exports cleanly under Bun.

Fixed

Fixed the status line session name (and the editor border / status-line gap fill) being nearly illegible on light themes.
Added IndexedSessionStorage and SessionStorageBackend exports to support shared metadata-indexed session backends
Added the tui.maxInlineImages setting (default 8) capping how many inline images render as live terminal graphics. Once a new image pushes the count past the cap, the oldest images are hidden via a full redraw — replaced by their [Image: …] text placeholder and purged from the terminal's graphics store — so long sessions with many screenshots/diagrams stop piling up images (and, on Kitty, stop leaving scrollback ghosts). Set to 0 to keep every image inline.
Added a "View: terminal state" item to the /debug menu that prints the detected terminal, live geometry and cell size, multiplexer, and the negotiated subprotocols actually in use — graphics (Kitty/iTerm2/Sixel), desktop notifications (BEL/OSC 9/OSC 99, plus whether OSC 99 was confirmed via a device-attributes probe), OSC 8 hyperlinks, 24-bit color, DECCARA rectangular-SGR background fills, and DEC 2026 synchronized output — alongside the scrollback-clear strategy (CSI 22 J vs CSI 2 J redraw / ED3 eager-erase risk) and the raw TERM/TERM_PROGRAM/COLORTERM detection signals.
Added a "Test: terminal protocols" item to the /debug menu that renders one live sample of every special escape protocol the renderer can emit — SGR text attributes (bold/italic/underline/strikethrough/inverse/dim), themed and 24-bit truecolor, OSC 8 hyperlinks, OSC 66 text sizing (large text), and an inline graphics swatch via the active image protocol (Kitty/iTerm2/Sixel, with a text fallback) — and fires a desktop notification, so you can eyeball which protocols the current terminal actually honors. The sample image is a gradient PNG generated in-process, so the graphics test needs no asset on disk.
Added the tui.textSizing setting (default off) that renders Markdown H1 headings at 2x scale via Kitty's OSC 66 text-sizing protocol. It replaces the undocumented PI_TUI_TEXT_SIZING env var with a real setting, and only takes effect on Kitty terminals (where OSC 66 is implemented) — it is ignored everywhere else so headings never emit raw escape bytes.
Added a lifecycle status to the /resume session picker. Each session's tail (last 32 KiB) is now read alongside the existing header window in a single pass, and its final message classified as done (the agent ended its turn and yielded control back), interrupted (a trailing tool call or tool result the loop never continued from), aborted, error, or pending (a trailing user message with no reply). The status renders as a colored segment on each session's metadata line. When the final message is larger than the tail window the status is omitted rather than guessed.
Added support for disable-model-invocation: true frontmatter field from the Agent Skills standard. Skills using this field are now hidden from the system prompt listing, matching the behavior of hide: true.

Changed

Changed the task tool description to tag read-only agents and explicitly forbid assigning them file edits/commands or offloading reasoning to quick_task/explore.
Changed Redis and SQL session storage initialization to load only indexed metadata (size, mtimeMs) instead of full session content
Changed SessionStorage read paths to rely on backend-backed metadata/indexed storage, so session content is fetched on demand rather than cached as full in-memory mirrors
Changed session-list slice reads to go through SessionStorage.readTextSlices across all backends, removing the file-only single-open branch and caller-managed buffers. FileSessionStorage now reads both windows via peekFileEnds, while Redis and SQL backends encode session content once per combined read.
Changed the ask tool transcript renderer to mark single-choice questions with circular radio glyphs (○/◉) instead of the rectangular checkbox glyphs (☐/☑) it shares with multi-select questions, so a "pick one" combo box visually reads as a radio group rather than a checklist. Multi-select questions keep checkboxes. Added a radio.selected/radio.unselected symbol pair across the unicode, nerd-font, and ASCII presets.
Changed the ask tool transcript renderer to mark the chosen answer inside the question form rather than re-listing the questions in a detached summary block below it. Once a question is answered, the standalone prompt preview is dropped and the result redraws the same form — every offered option still shown, with the selected one(s) filled in (◉/☑, highlighted) and the rest dimmed (○/☐); custom free-text answers and cancellations render in place as the final entry. This removes the duplicate question/option listing that previously appeared once as the call preview and again as the result.
Changed task-completion and ask desktop notifications to structured terminal notifications (title, body, type, and a focus-on-click action). On Kitty these render through OSC 99 as a proper title/body with click-to-focus; terminals without confirmed OSC 99 support collapse them to the previous single-line message (BEL/OSC 9).
Updated the "each kitty/tmux split" tip to include cmux.

Fixed

Fixed tiny-model startup in compiled binaries by resolving @huggingface/transformers and its runtime dependencies from the installed cache using package.json exports/main metadata, preventing module-resolution failures when launching models
Fixed tiny runtime installation flow in compiled binaries by using the build-time resolved @huggingface/transformers version and ensuring the runtime lock directory’s parent exists before acquiring the install lock, preventing mismatch and setup failures on fresh installs
Fixed the terminal protocol debug probe reusing one stable Kitty graphics id across repeated panels, which could move/replace an earlier swatch instead of rendering a new one.
Fixed selector dialogs (the ask tool, hook prompts) collapsing to a single visible option on shorter terminals when options carried long descriptions: the highlighted option's wrapped description consumed the entire row budget, hiding every other option and making the menu feel unnavigable (down moved the lone visible entry, left/right did nothing). When the fully-expanded list overflows, HookSelectorComponent now renders a compact list — every option label stays on screen and only the highlighted option expands its description, truncated to the remaining rows — so the whole menu is always visible and the detail pane follows the cursor.
Fixed read failing with "Path not found" on web URLs whose scheme // collapsed to a single / (e.g. https:/github.com/...), which happens when a URL is routed through Node's path.normalize/path.resolve. The fetch URL recognizer now accepts a single-slash scheme and repairs it back to // before fetching, so collapsed URLs resolve instead of falling through to filesystem lookup.
Fixed subagent slow-model priority falling through to older Claude Opus aliases when Opus 4.8 is available by adding Opus 4.8 and 4.7 aliases ahead of older Opus fallbacks (#1753).
Fixed the web-search provider selectors in TUI settings/setup to derive from the shared provider metadata, so newly added providers cannot be omitted from the preference list.

[15.8.3] - 2026-06-03

Fixed

Fixed Jujutsu workspace detection failing in non-default workspaces created by jj workspace add, whose .jj/repo is a FILE pointing at the shared repo dir rather than a directory. Detection now matches jj's own criterion (.jj/repo present, file or dir) instead of requiring a .jj/repo/store directory, and jj.repo.resolve's storeDir follows the file indirection to the shared store.

Changed

Changed the todo-write prompt to require initializing every item from a user-supplied multi-step plan as an individual todo task before execution
Changed context compaction (prune/shake) to protect reads of the active plan file the same way it already protects skill:// reads, so the plan stays intact through automatic and manual compaction. Both the canonical local://PLAN.md alias and the session's current plan reference path (e.g. a titled local://<title>.md after approval) are kept, tolerating read selectors and local:/ scheme spelling.

[15.8.2] - 2026-06-03

Added

Added a bundled TypeScript rule that warns against leaving @deprecated compatibility shims behind instead of finishing a refactor.

Fixed

Fixed /review's uncommitted-change mode in Jujutsu repositories to read jj diff --git from the current workspace, so non-default JJ workspaces include their working-copy changes instead of falling back to the colocated Git checkout.
Fixed empty assistant stop retry continuations preserving auto-retry state until a non-empty assistant turn completes or recovery reaches its retry cap.
Fixed TTSR rule conditions never matching streamed edit/write tool calls whose wire format obscures the real content (hashline + body rows, apply_patch envelopes, JSON-escaped write content). The edit and write tools now expose a matcherDigest normalization and TTSR matches against the introduced source text, so rule regexes stay universal regardless of the active edit mode.

Changed

Changed the JJ utility API to mirror Git's scoped helpers: repository operations now live under jj.repo (root, resolve, is, clearRootCache), and diff file listing is available as jj.diff.changedFiles.
Changed the search_tool_bm25 tool description to name the hidden discoverable built-in tools (e.g. write, find, search, lsp, task) when tools.discoveryMode: "all" is active, so a model can form a targeted discovery query by name instead of guessing or falling back to shell. mcp-only mode is unchanged (no built-ins are advertised) and the Total discoverable tools available: N count still includes them.

[15.8.1] - 2026-06-02

Fixed

Fixed an unhandled EPIPE rejection when an MCP stdio server exits between returning the initialize response and the client's notifications/initialized send. StdioTransport.notify() and #sendResponse() now route stdin writes through a shared helper that catches synchronous sink failures: notify() tears the transport down (firing onClose) and surfaces a Transport closed while sending notification rejection so connectToServer() treats the handshake as a failed connection instead of returning a "connected" handle wrapping a dead transport; #sendResponse() stays silent because a dead subprocess has no use for the response. StdioTransport.close() is now the authoritative resource teardown — it no longer early-returns when #handleClose() has already flipped #connected, so the subprocess and read loop are always cleaned up (including in the connectToServer() failure path) (#1710).
Fixed startup model resolution ignoring cached discovery rows for special built-in providers (google-antigravity, google-gemini-cli, openai-codex) until the background refresh completed (#1721).
Fixed Windows clipboard-image paste keeping Ctrl+V unregistered by default. The TUI now registers Ctrl+V plus the Windows Terminal-safe Alt+V fallback, and the keybinding docs call out when to use the fallback (#1708).

[15.8.0] - 2026-06-02

Added

Added an all-projects scope to the session picker (pi --resume / /resume). Press Tab to toggle between the current folder's sessions and every session across all projects; the all-projects list is loaded lazily and shows each session's directory. When the current folder has no sessions the picker now opens straight into all-projects scope instead of printing "No sessions found".
Migrated the Kagi web search provider to Kagi's V1 Search API (POST /api/v1/search), replacing the sunset V0 endpoint while keeping the kagi provider id, KAGI_API_KEY credential, and /login kagi flow unchanged (#1272 by @thismat)
Added Anthropic anthropic-ratelimit-unified-* response-header warming for /usage and the status-line usage segment, throttled to reduce direct OAuth /usage probes during active use.

Changed

Changed Anthropic API request metadata user_id.device_id to be derived from the persistent install ID so it remains stable across Anthropic account changes on the same install
Changed the eval parallel() / pipeline() helpers to drop their concurrency argument and run as wide as a task tool batch. The worker-pool ceiling now tracks the task.maxConcurrency setting (default 32; 0 = run every item at once) resolved live from the host via a new __concurrency__ bridge, instead of the old per-call concurrency option that defaulted to 4 and capped at 16. This stops eval fan-outs from under-using the parallelism the session is configured for.
Changed subagent / agent:// output ids from a numeric prefix scheme (0-Anna, 1-Bob, nested 0-Anna.1-Bob) to a name-first scheme: the requested name is used verbatim and a -2/-3/… suffix is added only when the same name recurs within a session (Anna, Anna-2, Anna-3). Nested ids stay grouped under the parent (Anna.Bob) and the live task widget renders them as Anna>Bob. The main agent's IRC id is now Main (was 0-Main). AgentOutputManager still scans existing .md outputs on resume so it never reuses a name that would clobber a prior output.
Changed resuming a session that belongs to a different project to switch the process into that project's working directory. pi --resume and the in-session /resume picker now chdir into the resumed session's cwd and re-scope every cwd-derived input — project dir, project settings (.claude/settings.yml, .omp/settings.json, path-scoped enabledModels/disabledProviders), plugin roots, capabilities, slash commands, and the ssh tool — so tools, discovery, configuration, and commands all follow the resumed project. The SessionManager adopts the resumed session's own cwd/session directory on load (rolled back if the switch fails).
Documented ANTHROPIC_SEARCH_API_KEY and ANTHROPIC_SEARCH_BASE_URL more thoroughly in docs/environment-variables.md, surfaced them in docs/tools/web_search.md's Anthropic provider section, and added ANTHROPIC_SEARCH_BASE_URL to omp --help's env-var summary so the search-only overrides are discoverable alongside ANTHROPIC_SEARCH_API_KEY (#1694).
Migrated the Kagi web search provider to Kagi's V1 Search API (POST /api/v1/search), replacing the sunset V0 endpoint while keeping the kagi provider id, KAGI_API_KEY credential, and /login kagi flow unchanged (#1272 by @thismat)

Fixed

Fixed read, search, find, ast_grep, and ast_edit recovering when a model flattens multiple existing paths into one comma-, semicolon-, or space-delimited string while preserving real paths that contain delimiters.
Fixed Exa web search reporting available without Exa credentials, which could route searches into the unauthenticated public MCP fallback and stall before trying the next provider. Availability and searchExa() now resolve through the standard AuthStorage cascade (EXA_API_KEY env or stored credential) (#1695).
Fixed Anthropic web search ignoring ANTHROPIC_SEARCH_BASE_URL when credentials came from stored Anthropic auth or generic Anthropic env fallback rather than ANTHROPIC_SEARCH_API_KEY (#1694).
Fixed web_search returning 401 from corporate Anthropic API gateways. ANTHROPIC_CUSTOM_HEADERS is now forwarded to web-search requests whenever ANTHROPIC_BASE_URL points to a non-Anthropic host, not only in Foundry mode (#1693).
Fixed opening exported local files from WSL by sending existing paths through wslpath -w and launching wslview directly when available, avoiding xdg-open's broken file-handler path translation (#950 by @rxreyn3).
Fixed /move (and cross-project resume) not re-scoping the live project settings to the destination directory. Changing a session's working directory now reloads the project settings layer in place (via Settings.reloadForCwd) so project-scoped configuration and path-scoped enabledModels/disabledProviders follow the move instead of remaining pinned to the launch directory.
Fixed read <db.sqlite> freezing the TUI on large databases. Listing tables ran an unbounded SELECT COUNT(*) per table, and since bun:sqlite executes synchronously on the same JS thread that drives rendering and input, a multi-GB database's full-table scans blocked the UI for seconds. The listing now reads the planner's sqlite_stat1 estimate for tables above a scan cap (shown as ~N rows) and only counts exactly when a table is provably small, reading at most cap + 1 rows (a capped table shows N+ rows). On an 8.4 GB stats database the listing dropped from multi-second full scans to ~2 ms.
Fixed a module-load crash (ReferenceError: Cannot access 'evalToolRenderer' before initialization) triggered whenever tools/eval was imported before tools/renderers. The eval JS backend statically pulls the agent/task/sdk/extension chain, which re-enters the root barrel → modes/components → tool-execution → renderers while eval.ts was still initializing, so renderers.ts read evalToolRenderer in its TDZ. The eval TUI renderer is now split into a dependency-light tools/eval-render.ts that renderers.ts imports directly (decoupling pure rendering from the eval runtime); eval.ts re-exports evalToolRenderer/EVAL_DEFAULT_PREVIEW_LINES for compatibility.
Fixed /logout offering providers that were only authenticated by env/config fallbacks, which made OpenCode Go/Zen appear to remain logged in after their stored credentials were removed (#1658).
Fixed omp --resume <id> crashing with an uncaught exception when an interactive user declined the cross-project fork prompt. createSessionManager now returns undefined for that cancellation, while non-interactive invocations still fail with a diagnostic when they cannot answer the fork prompt (#1668).
Fixed history.db never recording the originating session id: the session_id column documented for 15.6.0 was missing from the shipped storage layer, so the column was never created/populated on the write path and every prompt row had session_id NULL. Restored the session_id column, schema migration (ALTER TABLE history ADD COLUMN session_id for pre-existing databases), and HistoryEntry.sessionId; wired interactive mode to register setSessionResolver(...) so prompts are stamped with the session active at submission time (tracking fork/resume switches); and re-enabled prompt-history ranking in the --resume and in-session session pickers via HistoryStorage.matchingSessionIds().
Fixed /quit shutdown leaving the parent shell prompt at the top of the viewport after the final TUI teardown render on Linux terminals (#1620).
Fixed omp update failing with No version matching "X" found for specifier "@oh-my-pi/pi-coding-agent" (but package exists) when bun saw an older catalog than the update check did. The version is resolved by querying https://registry.npmjs.org/ directly, but bun install -g would then consult its on-disk manifest snapshot or a configured npm mirror (corporate proxy, Taobao, …) that hadn't replicated the release. The bun install step now runs with --no-cache --registry=https://registry.npmjs.org/ so it hits exactly the registry the version check used (#1686).
Fixed ACP mode resetting configured async/background job settings to disabled RPC defaults, so explicit ACP background-job opt-ins are preserved during startup (#1324).
Fixed legacy Pi extensions loading their __dirname-relative assets as empty — e.g. @plannotator/pi-extension reading plannotator.html/review-editor.html via readFileSync(join(__dirname, …)), which left planHtmlContent empty and dropped the plugin into its "no UI support" auto-approve path. The compat layer previously mirrored every extension module into a flat temp directory, so import.meta.url (and thus __dirname) pointed at the mirror root and sibling asset reads ENOENTed. Extensions now load in place from their real on-disk location — import.meta.url is the real source file, so asset reads resolve exactly as under the original Pi runtime. A Bun.plugin() onLoad hook scoped to the entry's relative-import graph (the exact set of source modules, never the host, other extensions, node_modules deps, or unrelated project files) rewrites only the legacy @(scope)/pi-* and @sinclair/typebox imports; relative siblings, the extension's own node_modules deps, and bundled assets all resolve natively, with no temp-directory mirroring and no asset copying (#1674).
Fixed plan approval keep-context usage to use the execution model context window, avoid zero-usage aborted turns, and disable keeping context when preserved context exceeds 95%.

[15.7.6] - 2026-06-01

Added

Added ask option descriptions so agents can keep short labels and render explanatory text as separate muted rows in the selector.
Added an extension API for rendering supplemental UI below visible assistant thinking blocks.
Added default-on lsp.diagnosticsDeduplicate support so post-edit LSP diagnostics already shown for a file are suppressed within the session and only new or changed diagnostics are surfaced.

Fixed

Fixed a single ESC press both dismissing the @/slash autocomplete popup and aborting the running agent operation. ESC now drains the popup first; only when no popup is visible does it route to the global interrupt handler (matching the standard TUI/IDE pattern). The shouldBypassAutocompleteOnEscape editor hook is removed — it had become the trigger for this bug (#1655).
Fixed Claude Code slash command discovery to load subdirectory commands recursively while preserving basename commands (e.g. /apply) and adding namespace aliases (e.g. /opsx:apply) for tools that install colon-namespaced workflows (#1523).
Fixed the eval tool's per-cell timeout killing cells that were not stalled. The timeout is now a plain wall-clock budget on the cell's own work that is paused only while a host-side agent()/parallel()/llm() bridge call is in flight — those calls pump a heartbeat that re-arms the watchdog, so a long fanout or a slow (e.g. reasoning-tier) completion runs to completion instead of being aborted mid-flight (a subagent's time-to-first-token, a long quiet nested tool, or an entire oneshot llm() request no longer trip it). Nothing else re-arms the budget: ordinary compute, print/stdout, log()/phase(), and non-agent tool calls all count against it, so a cell that is not delegating to an agent/llm is bounded by the regular wall-clock timeout (and the timeout message no longer says "of inactivity"). The heartbeat is a pure keepalive — never persisted or rendered.

[15.7.5] - 2026-06-01

Fixed

Fixed streaming assistant responses leaving duplicated tail rows in WSL/Windows Terminal scrollback by enabling eager native-scrollback rebuilds while assistant text is actively streaming (#1615).
Fixed the task tool mangling subagent prompts when a model double-JSON-encodes a string argument: context and each task's assignment/description are now repaired when they arrive uniformly double-escaped (literal \n, \", \uXXXX), so the subagent receives the intended prose and the call preview renders real newlines. The repair is guarded by a JSON-string round-trip and a double-encode signature, so legitimate backslashes/quotes (Windows paths, regexes, embedded quotes) are left untouched, and it is scoped to these natural-language fields only (never code-bearing tools).
Fixed pr:// PR views omitting formal review submissions and approvals when comments are enabled (#1600).
Fixed subagent yield-reminder loop logging benign user/compaction aborts as ERROR. The catch around session.prompt/waitForIdle in task/executor.ts now demotes ToolAbortError and signal-aborted exits to debug and keeps ERROR for genuine prompt failures only (#1623).
Fixed read local://<file> resolving to the wrong session's artifacts directory in multi-session ACP hosts (e.g. cmux). LocalProtocolHandler.resolve now honors context.localProtocolOptions supplied by the calling tool before falling back to the process-wide override or the first main-kind session in the global AgentRegistry; read, find, search, ast_grep, and ast_edit thread their session's options through so a local://PLAN.md lookup hits the calling session's local root instead of a sibling session's (#1608).
Fixed omp segfaulting on exit on Windows after the tiny title/memory model loaded onnxruntime-node (issue #1606). The tiny model now runs in a Bun subprocess instead of a Worker thread, so the NAPI finalizer that crashes during shutdown never executes in the agent's address space; the subprocess is SIGKILL'd on dispose to skip every native destructor on every platform.
Fixed unbounded MCP reconnect loop that could fork-bomb the host when a stdio MCP server completes the initialize/tools/list handshake and then exits. MCPManager now enforces a per-server crash circuit breaker (5 reconnects per 30 s window) on the automatic transport.onClose path; manual /mcp reconnect resets the window so users can recover after fixing the misconfiguration (#1592).

[15.7.4] - 2026-05-31

Removed

Removed /shake summary, the shake-summary auto-compaction strategy, and the providers.shakeSummaryModel setting. Use /shake or compaction.strategy: shake for mechanical artifact-backed elision without local-model CPU.

Fixed

Fixed plugin install failing for sources pinned to a SHA: git.clone() no longer adds --depth 1 when options.sha is set, so the checkout of arbitrary commits succeeds instead of bailing out with "shallow clone may not contain this commit" (#1589).

[15.7.3] - 2026-05-31

Added

Added support for decimal and k/m suffix turn-budget directives, enabling budgets like +1.5k and +2m in eval message parsing
Changed eval budget resolution to honor a user +Nk directive over an active Goal Mode limit while falling back to Goal Mode when no per-turn ceiling is set
Added agent() eval options agent_type/agentType, model, context, and label, and returned structured JSON when schema is provided in JS and Python eval cells
Added a live, Task-tool-style progress tree for eval agent() calls, drawn below the notebook (code cell) box. Each subagent surfaces as a status line (icon · id · tool count · context · cost, plus duration on completion) with its current tool/intent while running, and updates mid-execution rather than only at the cell's final result. Progress events coalesce per subagent id so the persisted event list stays bounded across many throttled ticks.
Added agent() to the eval runtime so JS and Python cells can spawn one subagent through the existing task executor; JS eval also gained bounded parallel() and pipeline() helpers for orchestrating subagent calls.
Added a workflow magic keyword (mirrors orchestrate/ultrathink): the standalone word glows amber→green in the editor and appends a hidden notice steering the model to author deterministic multi-subagent fan-outs in eval (agent/parallel/pipeline). Matching is whitespace-delimited and case-sensitive (lowercase only); the singular and plural both trigger, but capitalized forms, inflections like workflowed, and path-embedded occurrences like workflow.ts do not.
Added parallel() and pipeline() to the Python eval runtime (thread-pool over the synchronous agent() bridge), mirroring the JS helpers: bounded pool (default 4, max 16), input-order preservation, a barrier between every pipeline stage, and contextvar propagation so agent() works inside worker threads.
Added log(), phase(), and a budget object to both eval runtimes (Python and JS). log/phase emit progress/phase status lines; budget.total/budget.spent()/budget.remaining()/budget.hard expose a real per-turn output-token budget. A +Nk directive in the user's message sets an advisory budget (the model self-limits via budget.remaining()); +Nk! (or an active Goal Mode budget) makes it a hard ceiling that blocks further eval agent() spawns once reached. budget.spent() counts output tokens spent this turn across the main loop and all eval-spawned subagents.
Added search support for virtual internal URLs (including omp:// roots) by resolving and scanning in-memory internal resources as search targets alongside filesystem paths
Added expansion of virtual internal URL search targets so search can match multiple internal documents when given omp://
Added /omfg <complaint> slash command that drafts a TTSR rule from a complaint, validates it against the current conversation, saves it to project or ~/.omp/agent/rules, and registers it live.
Added /shake slash command and the shake / shake-summary compaction strategies that reduce context by mechanically dropping heavy content instead of LLM summarization. /shake (alias /shake elide) strips heavy tool-call results and large fenced/XML blocks, offloads the originals to one session artifact, and leaves a recoverable artifact://<id> placeholder; /shake summary compresses the same regions with a local on-device model (providers.shakeSummaryModel, default qwen3-1.7b) and falls back to elide per region when the model is unavailable; /shake images strips image blocks. Auto-maintenance honors the shake / shake-summary strategies (16k protect window); on context overflow a shake that reclaims nothing falls back to context-full summarization.
Added providers.shakeSummaryModel setting selecting the local on-device model used by /shake summary and the shake-summary compaction strategy. Runs entirely on-device (downloads on first use) and never calls a remote/cloud LLM.

Changed

Changed the eval cell timeout from a hard wall-clock deadline to an inactivity (idle) budget: a cell is now interrupted only after going the full window with no progress signal, and every status event — agent() progress snapshots, log()/phase(), and tool-bridge activity — re-arms the watchdog. Long agent()/parallel() fanouts that keep reporting progress no longer time out mid-run (previously the kernel was killed at the fixed deadline even while subagents were actively progressing). Raw print/stdout does not reset the watchdog, so pure-compute runaway loops stay bounded; the timeout is driven entirely by the abort signal, so neither runtime arms a competing fixed timer.
Fixed turn-budget parsing to match +Nk directives only at token boundaries, preventing values like version 1.2.3, c++, and +500kfoo from triggering a budget rule
Changed overflowing provider, hook-option, branch-message, agent, extension, and session-tree pickers to support fuzzy type-to-filter search.
Changed Shift+Ctrl+P to cycle role models backward instead of cycling forward without persisting.
Changed empty prompt input so ? inserts a literal question mark instead of opening /hotkeys; use /hotkeys explicitly for the shortcut reference.
Changed search output to preserve full virtual and internal URL paths in grouped results and details.files instead of collapsing them to file basenames
Changed /omfg to run up to three generation attempts with validation feedback and only prompt saving when no draft matches assistant history
Changed /omfg to show a live draft panel with generation/validation/saving status and allow canceling an active rule request with Esc
Changed keybindings config to use ~/.omp/agent/keybindings.yml, with automatic migration from legacy keybindings.json and continued support for keybindings.yaml.
Changed the local SQLite memory backend identifier from mnemosyne to mnemopi. Existing configs are migrated automatically on load: memory.backend: mnemosyne becomes mnemopi and the mnemosyne.* settings block is renamed to mnemopi.* (skipped when an explicit mnemopi block already exists).
Changed the ultrathink/orchestrate/workflow magic keywords to be markdown-aware: the standalone word now also glows in the rendered user message bubble (matching the live editor), and neither the glow nor the hidden steering notice triggers when the keyword sits inside a fenced code block, an inline `code` span, or an XML/HTML section.

Fixed

Fixed Ctrl+O tool-result expansion on POSIX terminals so offscreen tool blocks rebuild native scrollback instead of leaving stale collapsed rows above the viewport.
Fixed final agent() completion status emissions in eval cells so the last live progress snapshot now preserves accumulated subagent metrics such as tool count and cost
Fixed agent() in eval to enforce plan-mode, spawn allowlist, and disabled-agent checks before launching subagents
Fixed recursive agent() calls from eval by enforcing the existing max subagent depth limit
Fixed runtime model switches (Ctrl+P cycling, --model, /model, model picker selections, and programmatic changes) so they no longer overwrite the persisted modelRoles.default; only the model picker's explicit "Set as default" action and settings changes persist the default.
Fixed search to honor line-range suffixes on virtual internal URL targets so matches outside the requested ranges are no longer returned
Fixed search to handle internal URLs without source files without incorrectly reporting Path not found, returning matches from virtual content instead
Fixed /omfg parsing to tolerate fenced or noisy model output, normalize generated rule names, and reject invalid regex conditions before saving
Fixed auto-thinking sessions to persist the concrete resolved effort after classification, so resuming the session restores that level instead of returning to pending auto.
Fixed extension-registered CLI flags (e.g. --spawn-peer <value>) leaking into the initial prompt: argv is re-parsed once the extension flag set is known so flag values are consumed instead of becoming messages or being misread as @file arguments. Registered flags shadow same-named built-ins, so a colliding flag (e.g. plan-mode's --plan) is parsed with the extension's semantics rather than being consumed by the built-in branch (which would otherwise eat the following message and corrupt the built-in field). Extension flags and @file arguments are now resolved before the session is created, so an unreadable initial @file exits without leaving a junk session/terminal breadcrumb behind. (#1503)
Fixed footer status-line truncation: the left stats and right model segments now truncate by terminal cell width (via truncateToWidth) and strip all VT/ANSI escapes (via stripVTControlCharacters) instead of a SGR-only regex plus code-point substring, so wide glyphs, OSC hyperlinks, and non-SGR sequences can no longer overflow the line.
Fixed the streaming edit diff preview rendering a tall, half-empty box (and the earlier "box grows and shrinks repeatedly" stutter). A whole-file Myers re-diff is recomputed on every streamed chunk and its alignment is not monotonic in payload length, so a hunk-aware window that kept whole change segments gained and lost rows tick to tick; the prior high-water row reservation hid that stutter but padded the reserved height with blank rows, leaving a large empty rectangle whenever the diff shrank below its peak. The preview now pins a fixed-height trailing window to the bottom of the diff ("accept from the back"), so the box stays a steady, full window of real diff context instead of blank padding.
Fixed duplicated/stale scrollback above a streaming tool result on POSIX terminals (macOS/Linux). A tool whose output grows and re-lays-out (e.g. an edit diff gaining hunks) re-renders rows that already scrolled into native scrollback; the unknown-viewport anti-yank deferral left the old copy in place while the new one rendered below, showing the block twice. The event controller now enables the TUI's eager native-scrollback rebuild while a foreground tool is executing (setEagerNativeScrollbackRebuild), so those offscreen re-renders rebuild history cleanly — a snap to the tail is acceptable mid-tool. Background-running tools and plain assistant-text streaming keep the no-yank deferral; the mode resets at each turn start.

Removed

Removed the /drop-images slash command; use /shake images, which strips every image from the session through the same dropImages() path.

[15.7.2] - 2026-05-31

Added

Added providers.autoThinkingModel setting so users can choose the auto thinking classifier backend (online smol or local tiny-memory model)
Added an auto thinking level that classifies each real user turn and resolves to a concrete low-through-xhigh effort, with online smol classification by default and an opt-in local on-device classifier.

Changed

Updated the interactive thinking selectors in model/model-role pickers and ACP thinking options to include auto as a selectable level
Updated footer and status-line rendering to show auto while auto-thinking is being resolved and auto → <level> once it resolves
Changed the local tiny-model device default to CPU on every platform; explicit providers.tinyModelDevice / PI_TINY_DEVICE values still opt into accelerated ONNX providers.

Fixed

Prevented auto-thinking classification from running on non-user synthetic turns and non-reasoning models, keeping the session on its provisional concrete effort
Added a bounded auto-thinking classification path that falls back to the provisional effort on failures/timeouts so prompts continue without interruption
Bypassed auto classifier for ultrathink prompts and resolved directly to the highest supported auto effort
Fixed the JavaScript eval kernel crashing the whole process with a segfault (SIGTRAP, getImportedModule on a null record) when imported code reached a local module whose relative-import graph contains a cycle — e.g. await import("…/edit/streaming.ts"), or any workspace path with cyclic re-exports. The LocalModuleLoader linked and evaluated each local module individually inside the recursive vm.SourceTextModule linker callback, which re-entered Bun's node:vm module linker mid-instantiation and detonated JSC on the first cycle. The loader now constructs the entire local module graph first and drives a single link() + evaluate() from the graph root, so cyclic graphs instantiate in one pass; external (node_modules) modules stay eagerly loaded since they carry no imports and cannot form a cycle.
Fixed the streaming edit preview rendering a blank box for hashline edits whose payload sits on the trailing in-flight line (the common single-op replace/insert case). The preview path trimmed that still-typing line before diffing, so a single-payload op collapsed to a "No changes" result — shown as an empty box — for almost the entire stream. Hashline previews now feed the raw in-flight text through applyPartialTo, whose streaming-tolerant parser drops a payload-less trailing op and projects a partially-typed payload line as it grows, so the diff appears and fills in live. Transient errors from the actively-typed trailing section are also suppressed while streaming (regardless of section count) so a mid-typed op can't wipe an already-good preview frame; real errors still surface once args are complete.

[15.7.0] - 2026-05-31

Added

Added a Web search setup tab that lets users choose the preferred providers.webSearch provider during onboarding
Added manual authorization-code/redirect URL prompts for OAuth providers that require non-callback login in the setup wizard
Added an omp completions <bash|zsh|fish> command that prints a shell completion script generated from the live command/flag metadata, so completions never drift from the actual CLI. Subcommands, flags, and enum values complete statically; --model/--smol/--slow/--plan resolve against the bundled model catalog and --resume against on-disk sessions via a hidden __complete helper.
Added a /switch slash command that opens the temporary model selector for the current session, mirroring the alt+p keybinding.
Added replace block N: and delete block N operators to the edit tool: they resolve the syntactic block beginning on line N via tree-sitter (native blockRangeAt) and replace or delete its full line span, so a construct can be rewritten or removed without counting its closing line. Unresolvable blocks (unsupported language, blank/closing-delimiter line, or a parse error) are rejected with guidance to use an explicit replace N..M: / delete N..M range.
Added an animated pending border for bash and eval execution blocks: while a command/cell is running, a single dark segment glides clockwise around the block's outer edge (top → right → bottom → left), replacing the previous static accent border. Motion is eased per edge (decelerating into each corner) and timed against a fixed lap duration mapped onto the live perimeter, so streaming a new output line or resizing the terminal nudges the segment proportionally instead of resetting its position. Driven by the existing spinner cadence and gated on the display.shimmer setting (no motion when disabled).
Added providers.tinyModelDevice and providers.tinyModelDtype settings (Providers tab) controlling local tiny-model acceleration for session titles and Mnemosyne memory tasks. providers.tinyModelDevice selects the ONNX execution provider (default keeps the platform pick — DirectML on Windows, CUDA on Linux x64, CPU elsewhere); providers.tinyModelDtype selects quantization/precision (default keeps each model's shipped q4, e.g. fp16 trades speed for fidelity). The PI_TINY_DEVICE / PI_TINY_DTYPE env vars override the matching setting. Also added PI_TINY_DTYPE as the env counterpart to PI_TINY_DEVICE; an unrecognized device/precision fails loudly at worker startup instead of silently loading a different one.
Added a bundled set of default rules shipped with the agent (TypeScript/Rust convention rules registered as TTSR conditions). They load via the new lowest-priority builtin-defaults discovery provider, so any user/project/tool rule of the same name overrides the bundled copy. Disable the whole set with ttsr.builtinRules: false, or drop individual rules (bundled or your own) by name via ttsr.disabledRules.

Changed

Changed setup onboarding to a tabbed Set up your providers scene with dedicated Sign in and Web search panels
Changed the glyph mode picker to preselect the currently configured symbol preset instead of always defaulting to Unicode and to show live glyph samples in the picker rows
Changed OAuth sign-in flow in the setup wizard so users can authenticate multiple providers before leaving with Escape
Changed the plan-approval model-tier slider and the ctrl+p/alt+p role-cycle status to share one status-line-style chip track: each tier renders in its own role color and the active tier is filled as a powerline chip with a luminance-matched label. The role-cycle status now shows only the chip track — the resolved model and thinking level already live on the status line — instead of the verbose Switched to <role>: <model> (cycle: …) line.
Changed the in-flight bash tool-call preview to render as a full bordered block as soon as the call appears, instead of a one-line Bash: $ … status that only expanded into a block once the command produced its first output chunk. Silent commands (e.g. sleep 30) now show the framed command block — with the animated pending border — for their whole runtime.
Changed local tiny-model inference to request a worker-safe accelerated ONNX execution provider where available (DirectML on Windows, CUDA on Linux x64), with CPU retry if acceleration cannot initialize. PI_TINY_DEVICE=cpu restores CPU-only behavior; PI_TINY_DEVICE=metal is accepted as a WebGPU alias but guarded back to CPU in the production macOS worker because WebGPU currently hard-crashes Bun on worker teardown.

Fixed

Used the native block resolver for hashline operations so replace block edits now derive block ranges from file-aware parsing
Fixed OAuth login handling to cancel cleanly when users press Esc or Ctrl+C during authentication
Fixed the read tool description advertising inspect_image ("for visual analysis, call inspect_image") even when the inspect_image tool was disabled, which left the model hunting for a tool absent from its function list. The image section is now gated on inspect_image.enabled: when disabled it instead states that reading an image path returns the decoded image inline.
Fixed session-title generation latching onto literal text inside fenced code blocks — a pasted UI mockup containing "Welcome to Claude Code v2.1.158" titled the session "Setup Screen for Claude Code v2.1.158" instead of capturing the actual request. The first user message now has fenced code blocks stripped before titling (both the online pi/smol and local on-device model paths share the same preprocessing), with a fallback to the original message when stripping would leave too little to title from (e.g. a message that is essentially just a code block).
Fixed slash-command autocomplete repaint requests so Windows Terminal sessions with unknown native viewport state keep updating the input box and candidate list. (#1550)
Fixed Python eval failing the whole session when the managed ~/.omp/python-env interpreter exists on disk but no longer runs (e.g. a stale uv-managed Python that was removed or upgraded). Availability resolution now enumerates every candidate — active/project venv, the managed env, then the system interpreter — and probes each in priority order, falling through to the first that actually executes instead of failing fast on the first resolved path. The kernel spawns whichever interpreter the probe selected, so a working system Python takes over transparently.

Removed

Removed the recipe tool and its recipe.enabled setting. Task-runner targets (just/package.json/Cargo/make/Taskfile) are invoked directly through bash.

[15.6.0] - 2026-05-30

Added

Added prompt-mode autocomplete for supported internal URL schemes (skill://, rule://, agent://, artifact://, local://, memory://, and omp://) so typing those tokens now suggests existing resources as completion candidates
Added fuzzy matching and ranked suggestion ordering for internal URL completion, including rule and skill descriptions, with accepted completion replacing just the typed token and inserting the chosen URL followed by a space
Changed internal URL completions now include nested local:// path suggestions from the configured local workspace
Added Mnemosyne memory inference model selection with an online mode or local transformers.js options (qwen3-1.7b, gemma-3-1b, qwen2.5-1.5b, lfm2-1.2b) so memory extraction and consolidation can run via the shared tiny-model worker
Changed memory tiny-model handling to route local memory prompts through the same queueed tiny-model worker pipeline with bounded completion output
Added a Providers → Tiny Model setting for session titles, defaulting to the online pi/smol path with five optional local CPU transformers.js models. A local model — and the one-time @huggingface/transformers runtime install in compiled binaries — is downloaded and loaded only when explicitly selected (or via omp tiny-models download); the default online path never spawns the title worker for inference. Selecting a local model adds a delayed pi/smol fallback so titles never block, plus in-chat download progress.
Added a persistent live agent roster pinned below the editor (focus it with Ctrl+S or Alt+Down), including view-as switching into delegated agent sessions with human-readable delegate names and UI pinning to suppress idle reaping while viewed. The roster stays hidden until at least one delegated agent exists and releases focus back to the editor once the last one is gone.
Recorded the originating session ID alongside each prompt in history.db (new session_id column, surfaced as HistoryEntry.sessionId), so recalled prompts can be traced back to the session they came from. Existing history databases gain the column automatically on next launch.
Added compact inline TUI renderers for the retain, recall, and reflect memory tools. retain now shows one themed bullet line per stored item (truncated to width) under a status header with the stored/queued count, and recall/reflect collapse to a single query header (recall reports the match count and hides recalled memories until expanded) instead of dumping the raw JSON argument tree.
Added a randomly picked tip beneath the welcome screen, sourced from an embedded tips.txt (one tip per line). The line is italicized with a purple Tip: label and a dimmed light-blue body, and the tip is chosen once per welcome instance so intro-animation and LSP re-renders don't shuffle it.
Added a Mnemosyne-only memory_edit agent tool for updating, forgetting, or invalidating recalled memories by id, and added /memory stats plus /memory diagnose slash commands for backend maintenance visibility.
Added an orchestrate magic keyword that mirrors ultrathink: dropping the standalone word in a message paints it with a cool teal→violet gradient in the editor and appends a hidden system notice that switches the model into the multi-phase, parallel-subagent orchestration contract. Matching is word-bounded and case-insensitive, so orchestrated/orchestrating never trigger it.
Added a model-tier slider to the plan-approval prompt ("Plan mode - next step"). Left/right arrows move it from any list position to pick which configured role model (cycleOrder, e.g. smol › default › slow) executes the approved plan, with each tier colored by its role and the resolved model name shown beneath the track. The chosen tier is applied before dispatch and carries through the fresh/compacted execution session; the slider is hidden when fewer than two role models resolve.

Changed

Changed irc to treat the attached human as a first-class User peer, merging human prompts into irc call User with optional structured question payloads and adding /dm <agent> <message> for user-to-agent routing without switching views.
Changed the --resume session picker (and the in-session resume selector) to also rank sessions by prompt-history matches from history.db, not just the session-list metadata. Because the session list only indexes the first 4KB of each file, this surfaces sessions by prompts typed deep into long conversations. Sessions matched by both signals lead, then metadata-only matches, then history-only matches — no metadata match is dropped.
Changed the task tool's streaming call preview to list each dispatched agent's id and UI description as a tree instead of a bare N agents count, so the individual agents are visible while the tool-call arguments are still streaming. The collapsed view caps at 12 entries (… N more agents); the expanded view shows all.
Changed Mnemosyne recall tool output to include memory ids for explicit recall results so agents can target memory_edit; auto-injected memory context and reflect remain id-free.
Changed the system prompt to advertise memory://root only when the local memory backend is active.
Changed todo_write result rendering to animate completed items in place: the checkbox flips checked first, then the strikethrough reveals across the task text.

Removed

Removed the standalone ask, task, and yield tools along with their obsolete prompts, docs, and tests; delegation now routes through persistent delegate agents plus IRC coordination.
Removed the /orchestrate slash command; orchestration is now triggered by the orchestrate keyword (see Added) so the contract rides alongside the user's own prompt instead of replacing it.
Removed the sticky Todos panel all-done drop/collapse animation; completed todo state now stays visible until the next explicit todo update changes it.

Fixed

Fixed Mnemosyne session shutdown to flush queued memory extractions before exit so the last turn’s facts are not lost
Fixed a native crash (malloc: pointer being freed was not allocated / NAPI FATAL ERROR) when quitting after the local transformers.js title model had run. The tiny-title worker no longer calls pipeline.dispose() on shutdown — disposing the onnxruntime session freed native memory that Bun's worker/NAPI teardown then freed again. The worker is torn down immediately after, so the OS reclaims the model memory regardless.
Fixed the tiny-title download progress bar flashing on every first message even when the local model was already downloaded. A cached model emits the same download/progress events as a real download, so the bar is now revealed only when in-flight progress events keep arriving past a short grace window — cache hits finish (or fall silent during onnxruntime init) before then and never show the bar.
Fixed the Mnemosyne memory backend lifecycle so auto-retain counts the full session transcript, delegated agents inherit the parent Mnemosyne state, /memory clear removes scoped project-bank databases, session disposal closes Mnemosyne SQLite handles, session switches rekey/reset Mnemosyne tracking, and project bank names include an absolute-root hash with safe bank-name sanitization.
Fixed the streaming edit preview showing no diff for single-line hashline edits. The preview-diff coalescing keyed only on the arg text, so the final (args-complete) pass — which computes an untrimmed diff — was skipped because the payload was byte-identical to the last streamed chunk whose trailing line had been trimmed. The dedup key now pairs the streaming state with a content hash.
Fixed Esc in a delegated agent view returning to the main session instead of aborting the delegated agent's active turn.
Fixed the subagent stats line to separate the cost with the theme dot separator (was a stray literal .) and to render context usage as <pct>%/<window> (e.g. 21.3%/272K) matching the status line gauge, via a shared formatContextUsage helper now used by the footer, status-line segment, session observer overlay, and task renderer.
Fixed the agent roster staying pinned under the editor when all delegated agents are idle or dormant; it now reappears when explicitly focused with Alt+Down / session observe.
Fixed selector-style UI components to honor tui.select.up and tui.select.down keybindings instead of hard-coding raw Up/Down arrow bytes (#1535).
Fixed the bash (and recipe) tool result footer not rendering for failed commands. A non-zero exit threw a ToolError, which dropped the result details, so the styled ⟨Wall … | Timeout …⟩ footer was replaced by the raw Wall time: … seconds / Command exited with code N lines. Non-zero exits now resolve as a non-throwing error result that keeps wallTimeMs/timeoutSeconds/exitCode, and the footer shows ⟨Wall … | Timeout … | Exit: N⟩ with the textual notices folded out of the output pane. Aborts, timeouts, and missing-exit-status still throw as before.
Fixed selector-style UI components to honor tui.select.up and tui.select.down keybindings instead of hard-coding raw Up/Down arrow bytes (#1535).

[15.5.15] - 2026-05-30

Changed

Enabled the agent loop's tool-call batch cap for Anthropic Claude sessions, cutting oversized streamed tool-use bursts into runnable batches before continuing the conversation.

Removed

Removed the calc tool (deterministic arithmetic evaluator) and its calc.enabled setting. The model can compute via eval instead.

Fixed

Fixed Anthropic Claude tool-call batching to clear and reapply the Claude-specific batch cap whenever the session model changes

[15.5.14] - 2026-05-29

Added

Added progress status output for llm() calls in eval, including the resolved model, tier, and returned character count
Added an llm(prompt, opts) helper to both eval runtimes (JavaScript and Python) for oneshot, stateless LLM calls. opts.model selects a tier — "smol" (pi/smol), "default" (the session's active model, falling back to pi/default), or "slow" (pi/slow, with high reasoning effort on reasoning-capable models). Pass system for a system prompt and a plain JSON-Schema schema to force a structured response (the helper returns the parsed object instead of the completion string). Calls carry no conversation history and expose no agent-visible tools; they route host-side through the existing tool bridge under the reserved name __llm__ (packages/coding-agent/src/eval/llm-bridge.ts).

Fixed

Fixed a rewind/restore loop when the session leaf lands on an assistant turn that emitted tool calls (e.g. selecting such a turn in /tree, or restoring a session whose head is a mid-batch turn). That turn's tool results are persisted as its children — below the leaf — so they fall off the leaf→root path that buildSessionContext walks, leaving the assistant turn's tool_use blocks dangling as the final message. transformMessages then fabricated one synthetic "aborted"/"No result provided" result per call plus a <turn-aborted> developer note, which both rendered as phantom failed calls on a turn that "hadn't run anything yet" and re-injected the whole failed batch into the model's context, prompting it to re-issue the batch (the spiral). buildSessionContext now strips dangling tool_use blocks from a trailing assistant turn (dropping the turn entirely if nothing else remains), so a rewound turn resumes cleanly from its reasoning/text. Live turns are unaffected — their results persist and the leaf advances past the assistant before any rebuild.
Fixed external extension loading on Windows compiled binaries: bare @oh-my-pi/pi-* value imports (e.g. import { AssistantMessageEventStream } from "@oh-my-pi/pi-ai") failed with Cannot find package '\$bunfs\root\packages\…' because legacy-pi-compat.ts built shim override paths from a hardcoded POSIX /$bunfs/root/packages literal. Win32 normalised the leading slash to a backslash and the resulting path never resolved against the real bunfs mount (<drive>:\~BUN\root\…). The bunfs package root is now derived from import.meta.dir, so override paths stay platform-native on Windows, Linux, and macOS (#1514).
Fixed the interactive prompt showing no cursor in Ghostty. A prior change wired the editor's cursor mode to a new getUseTerminalCursorMarker() (which always reported the requested preference) instead of the resolved hardware-cursor visibility, so when Ghostty force-hid the hardware cursor the editor stayed in terminal-cursor (marker-only) mode and drew no glyph — leaving no visible caret with either showHardwareCursor/PI_HARDWARE_CURSOR value. The editor now follows ui.getShowHardwareCursor(): a hidden hardware cursor falls back to the steady software-cursor glyph (which still emits CURSOR_MARKER for IME positioning).

Changed

Changed the eval tool's display() JSON tree in the transcript to use the shared renderJsonTreeLines renderer (the same one behind tool args, MCP results, and subagent output) instead of its own format. This drops the redundant Object(N) / Array(N) type labels and the per-output JSON output N header in favor of type icons plus bare keys; the display[N] header is now shown only when a cell emits more than one display() value.
Removed the animated spinner from the sticky Todos panel. In-progress tasks and pending tasks with a matching in-flight subagent still highlight (accent colour + the static theme.status.running glyph), but no longer tick through theme.spinnerFrames, so the panel paints once per state change instead of on an 80 ms timer. Subagent auto-checkmarking, the advancing window (selectStickyTodoWindow), todoMatchesAnyDescription highlighting, and the all-done close animation are unchanged.

[15.5.13] - 2026-05-29

Breaking Changes

Changed hashline edit syntax to verb-based v4: body-bearing ops are replace N..M:, insert before N:, insert after N:, insert head:, and insert tail:, while bodyless delete N..M handles deletion. Removed >A..B repeat rows and the old prepend: / append: virtual insert headers; - rows remain rejected with a teaching error.

Changed

Changed hashline tag generation to use full-file snapshots for read/search/ast-grep and related outputs, so hashline anchors now validate only when the complete file matches
Changed hashline tagging to omit file headers for files over 4 MiB or that cannot be snapshotted, so those files are returned without editable hashline anchors
Changed hashline context generation for line edits from partial/sparse snippets to complete-file fingerprints, reducing stale anchors for partially read files

Fixed

Restored automatic repair of edit range hunks that break bracket balance — the failure class that previously left a duplicated closing line (a </> / ); / } echoed just below the range) or dropped one (the range swallowed a }); the payload never restated), leaving the file syntactically broken until a follow-up edit. The hashline applier now normalizes each replacement so its payload preserves the deleted region's delimiter balance, dropping a duplicated bordering closer or sparing a deleted one, and surfaces a warning on the tool result. Always on and balance-validated (no edit.hashlineAutoDropPureInsertDuplicates setting); see @oh-my-pi/hashline for the contract.

[15.5.12] - 2026-05-29

Added

Added the omp-plugins discovery provider, which scans every extension package directory configured via extensions: (in ~/.omp/agent/settings.json or <cwd>/.omp/settings.json) or --extension/-e on the CLI for skills/, hooks/pre|post/, tools/, commands/, rules/, prompts/, and .mcp.json. Prior to this, only the extension's TypeScript factory module ran; every sibling capability the docs (https://omp.sh/docs/extension-authoring) advertised was silently ignored (#1496).
Added the top-level omp install <target> subcommand documented at https://omp.sh/docs/extension-authoring. Local paths route to omp plugin link (so the directory is symlinked into the plugin set), and npm/marketplace specs route to omp plugin install. Before this, install was not a registered subcommand and the CLI runner silently forwarded install ./my-extension to launch as an initial LLM prompt (#1496).

Changed

Changed the sticky Todos panel above the editor to advance as tasks close, instead of pinning to the first 5 tasks of the active phase. selectStickyTodoWindow now shows up to 5 open (pending / in_progress) tasks in original phase order and reports the count of remaining open tasks for the +N more hint, so every todo_write flip produces a visible row shift. Closed-phase tail falls back to the last 5 tasks (with the +N more line suppressed) until getActivePhase walks to the next phase.
Linked the sticky Todos panel to the live SessionObserverRegistry so pending todos that have an in-flight subagent doing their work light up green with an animated spinner — the same theme.spinnerFrames ("status" preset) the task tool uses for its agent rows — instead of staying greyed out as if nothing is happening. A new exported todoMatchesAnyDescription(content, descriptions) does case- and whitespace-insensitive equality first with a 6-char minimum-overlap substring fallback in either direction, so "Sonnet #2: shallow bug scan" and a subagent description of "Sonnet #2" still link up. Completed todos now render with theme.status.success (✔ / \uf00c / [ok] per symbol preset, still wrapped in the success colour so themed palettes can keep their purple/green/whatever) and in_progress rows render with theme.status.running, matching the task tool's icon vocabulary. The spinner interval only ticks while at least one visible open todo has a matched active subagent, and self-stops once subagents finish, so plain in_progress todos do not animate forever in the absence of subagent activity.
Extracted the top-level CLI command table from src/cli.ts into a side-effect-free src/cli-commands.ts so test code can introspect the registered subcommands without triggering the entrypoint's top-level await.

[15.5.11] - 2026-05-29

Added

Added SqlSessionStorage, a bun:sql-backed implementation of SessionStorage that persists session JSONL into PostgreSQL, MySQL/MariaDB, or SQLite. Pass a connected Bun.SQL instance (the constructor accepts postgres://, mysql://, or sqlite: URLs) to SqlSessionStorage.create({ client, table?, adapter?, createTable? }) and hand the returned storage to any SessionManager factory. The dialect is auto-detected from client.options.adapter and used to pick the correct DDL plus upsert-with-append syntax (ON CONFLICT … DO UPDATE for PG/SQLite, ON DUPLICATE KEY UPDATE for MySQL), so the agent's append-only persist pattern works in a single round-trip per line. Same in-memory mirror and drain() semantics as the Redis backend; blobs and tool artifacts still live on disk via ArtifactManager/BlobStore.
Added RedisSessionStorage, a bun:redis-backed implementation of the SessionStorage interface that lets API consumers route session JSONL through Redis instead of local disk. Pass a connected Bun.RedisClient (or any compatible adapter) to RedisSessionStorage.create({ client, prefix? }) and hand the returned storage to SessionManager.create(cwd, sessionDir, storage) (or any other static factory that accepts a storage argument). An in-memory mirror is loaded on creation so the interface's synchronous methods (existsSync, statSync, listFilesSync, …) keep their contracts; drain() waits for queued background writes. Tool artifacts and image blobs still live on disk via ArtifactManager/BlobStore — Redis only owns the session JSONL keyspace under the configured prefix.
Exported the SessionStorage / SessionStorageWriter / FileSessionStorage / MemorySessionStorage symbols (already reachable via the ./session/session-storage subpath) from the package root so SDK consumers can construct alternative storage backends without deep-importing.
Added a fresh ¶<relative-path>#TAG snapshot header to the write tool's success text in hashline display mode, covering plain disk writes, ACP-bridge writes, and conflict resolutions (bulk resolutions emit a trailing Snapshots: block with one header per successfully written file). The header records a current snapshot in the file-snapshot store so the next edit can land without an extra read round-trip. Suppressed when the session is not in hashline mode and skipped for archive/SQLite writes and host-managed internal URL targets where hashline anchors do not apply.

Changed

The edit tool's stale-snapshot rejection message now distinguishes "file changed between read and edit" (the section's hash was recorded in this session but the file has since drifted — a prior in-session edit advanced it, or an external write changed it) from "hash #X is not from this session" (a fabricated or carried-over cross-session tag), the latter carrying explicit "never invent the tag" guidance. Both messages include the current file hash plus 2 lines of context around each anchor so the next attempt has everything it needs. Snapshot-based recovery still runs first; the sharper diagnostics only surface when recovery cannot reconcile the edit.

Fixed

Fixed Autonomous Memory phase 1/phase 2 failing with Thinking effort low is not supported by <provider>/<model> on models whose supported reasoning efforts exclude low/medium (e.g. deepseek/deepseek-v4-pro). Both stage1 (Effort.Low) and consolidation (Effort.Medium) call sites in packages/coding-agent/src/memories/index.ts now route through clampThinkingLevelForModel, lifting the requested effort to the model's lowest supported level instead of letting requireSupportedEffort throw (#1480).

[15.5.10] - 2026-05-28

Added

Added /drop-images slash command that strips every ImageContent block from the current session's branch — user/developer/custom/hookMessage/toolResult content arrays plus toolResult.details.images and fileMention.files[].image — rewrites the session JSONL, rebuilds the agent's in-memory message list, tears down Codex Responses provider sessions, and rebuilds the TUI chat container so the change is visible immediately. ACP clients receive the same handler (returns "Dropped N images …" / "No images found …" through runtime.output). Stripping content that would leave a toolResult or user message with zero blocks inserts a single [image removed] placeholder so providers do not reject empty content arrays.

Fixed

Fixed compaction surfacing raw HTTP 401/403 envelopes (e.g. Compaction failed: 401 {"type":"error","error":{"type":"authentication_error",…}}) instead of routing to an authenticated fallback model. The compaction layer now attaches the provider-reported HTTP status onto the thrown error, and AgentSession's auth-failure detector branches on error.status === 401 || 403 in addition to the existing auth_unavailable regex. When a fallback model role (e.g. modelRoles.smol) is configured, compaction retries it transparently; otherwise the user sees the actionable "Compaction requires usable credentials for …" hint instead of the raw provider envelope.

[15.5.8] - 2026-05-28

Breaking Changes

Changed hashline edit parsing to require wrapped hunk headers such as @@ A..B @@ (including @@ BOF @@ and @@ EOF @@), with empty @@ A..B @@ blocks deleting the anchored range and legacy inline payload forms treated as malformed

Added

Added vault.enabled setting (Tools → Obsidian Vault, default false) gating the vault:// internal URL. When disabled, VaultProtocolHandler.resolve / write, resolveVaultUrlToPath, and hasObsidian() all refuse — the latter hides the vault:// entry from the system prompt's Handlebars {{#if hasObsidian}} block. Tests can opt in via vi.spyOn(vaultProtocol, "isVaultEnabled").mockReturnValue(true).
Added support for vault:// URLs in path resolution utilities, including plan mode and internal selector parsing so read and edit paths can target Obsidian vault files directly
Added vault:// internal URLs for editable Obsidian vault files, with filesystem-backed read/write/listing and CLI-backed vault index operations.
Added strict-mode indicators to omp auth-gateway check output by appending [strict] to strict-mode text headers and adding a top-level strict field in --json output
omp auth-gateway check --strict exercises each broker-supplied credential against its provider's chat-completion endpoint (cheapest bundled chat model per provider, with 15s/attempt timeout and up to 4 catalog fall-throughs on "model not found / invalid model" errors). Surfaces failures where the usage endpoint reports 200 but the chat endpoint 401s the same bearer (revoked OAuth scope, mislabeled provider row, …). Output gains a [chat: ok|FAIL|skip] column in text mode and a completion field on each credential in --json mode; the chat-failed count contributes to the non-zero exit code.

Changed

Changed hashline apply behavior to preserve duplicated boundary and context lines in replacement and insert payloads instead of auto-absorbing or dropping them
Updated hashline syntax: replaced ↑/↓ payload sigils with ^ repeat syntax and | literal rows for clearer edit semantics
Changed hashline delete syntax from bare A: or A-B: to explicit A-B:- inline delete marker
Modified hashline anchor syntax to require explicit range notation A-B: instead of shorthand A: for single-line operations
Updated hashline description in settings to clarify pure insert context behavior without arrow notation

Removed

Removed the edit.hashlineAutoDropPureInsertDuplicates setting
Removed the edit.hashlineAutoDropPureInsertDuplicates setting from configuration and execution paths
Removed the edit.hashlineAutoDropPureInsertDuplicates setting
Removed the edit.hashlineAutoDropPureInsertDuplicates setting from configuration and execution paths

Fixed

Fixed agent yielding silently on response.incomplete (OpenAI Responses / Codex stopReason: "length"). The agent now treats output-side incompletion as a recovery case: drops the truncated/reasoning-only assistant turn, attempts context promotion to a larger model, and falls back to compaction or handoff. AutoCompactionStartEvent.reason and the custom-tool auto_compaction_start.trigger discriminator gain an "incomplete" value. The handoff strategy is honored for "incomplete" (unlike "overflow", where the input is broken and handoff would hit the same wall).
Fixed eval tool to resize large displayed images and append dimension notes to text output
Fixed write tool to strip malformed or loose hashline section headers before writing file content
Fixed eval tool image rendering to resize displayed images before returning them and append image-dimension notes to text output
Fixed write tool output sanitation to strip malformed or loose hashline section headers before writing file content
Fixed omp auth-broker serve crashing at startup with logger.setTransports is not a function — switched the call site to import { setTransports } from "@oh-my-pi/pi-utils/logger", bypassing the logger namespace re-export that some Bun versions failed to expose at runtime
Fixed omp auth-gateway returning 502 upstream_error and refusing to rotate credentials when a provider responded with a non-401 usage-limit error (Codex usage_limit_reached, Anthropic usage_limit_reached, Google resource_exhausted). classifyGatewayError now reuses pi-ai's central isUsageLimitError heuristic and reports those failures as 429 rate_limit_error. streamSimple's pre-emit retry hook fires on usage-limit phrasing in addition to HTTP 401; the gateway's refresh callback branches on the error type and calls AuthStorage.markUsageLimitReached(provider, sessionId, { retryAfterMs }) — temporarily blocking just the exhausted credential and surfacing the next sibling — instead of invalidateCredentialMatching, which would have suspect/deleted the row. The same branching is wired into the coding-agent streamFn callback so subscription multi-account rotation works the same on both surfaces.
Fixed extractRetryHint not recognising Codex's Try again in ~N min. / … hour / … hours phrasing, which left the gateway and TUI without a server-suggested retry window when an upstream account hit its usage cap. The shared try again in pattern now accepts min, minutes, mins, h, hr, hour, hours units in addition to ms / s / sec, and tolerates a leading ~ and embedded whitespace.
Fixed the auth-gateway threading sessionId: undefined into AuthStorage.getApiKey, which left #sessionLastCredential empty and made markUsageLimitReached a no-op for gateway-mediated requests. Both /v1/chat/completions-style endpoints and the /v1/pi/stream fast path now derive a stable sessionId from the client's prompt_cache_key (or the existing model+system+tools+first-message hash when absent) and reuse the same identity for credential-stickiness and prefix-cache routing.
Fixed eval tool to resize large displayed images and append dimension notes to text output
Fixed write tool to strip malformed or loose hashline section headers before writing file content
Fixed eval tool image rendering to resize displayed images before returning them and append image-dimension notes to text output
Fixed write tool output sanitation to strip malformed or loose hashline section headers before writing file content
Fixed omp auth-broker serve crashing at startup with logger.setTransports is not a function — switched the call site to import { setTransports } from "@oh-my-pi/pi-utils/logger", bypassing the logger namespace re-export that some Bun versions failed to expose at runtime
Fixed user shortcut Python execution to namespace session IDs like eval, so both paths share one kernel

Security

Secured vault:// reads and writes by validating URL paths and blocking traversal, absolute paths, and symlink escapes outside the selected vault root

[15.5.7] - 2026-05-27

Added

providers.openrouterVariant setting (Settings → Providers → "OpenRouter Routing") to default OpenRouter requests to a routing-variant suffix (:nitro, :floor, :online, :exacto). Selectors that already name a variant (e.g. openrouter/anthropic/claude-haiku:nitro) keep precedence.
generate_image supports xAI Grok Imagine via providers.image=xai. Supports grok-imagine-image (default) and grok-imagine-image-quality at aspect ratios 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3. Uses the xAI Grok OAuth credential when available, otherwise XAI_API_KEY.
New tts tool synthesises speech via xAI Grok Voice behind the disabled-by-default tts.enabled setting. Built-in voices ara, eve (default), leo, rex, sal; custom voice IDs also accepted. Output codec inferred from the output_path suffix (.wav → wav, else mp3). Up to 15,000 characters per request.

Fixed

Fixed plan-mode re-entry after approval reopening a fresh local://PLAN.md instead of the approved titled plan artifact, which could duplicate plan content and fail approval on an existing destination.
Fixed read URL reader mode aborting after a stalled Jina request instead of falling back to trafilatura/lynx/native: Jina (and Parallel extract) now have their own per-attempt sub-budget capped at 10s, the catch handler honours only real user cancellation, and the in-process native renderer is always attempted on already-loaded HTML (#1449)

[15.5.6] - 2026-05-27

Added

Support for multi-range line selectors on URLs (e.g., :5-10,20-30) to fetch and display multiple non-contiguous sections
Support for combining :raw mode with line range selectors on URLs (e.g., :raw:1-120 or :1-120:raw)
Support for line range selectors on directory listings (e.g., :30-40 to view lines 30–40 of a directory tree)
Clear error message when requesting a line offset beyond the end of a directory listing

Changed

URL selector parsing now supports multiple trailing selector tokens (e.g., :raw:N-M), applying them left-to-right

Fixed

Fixed :raw selector being ignored for JSON and feed URLs, causing them to be pretty-printed or converted to markdown instead of returning raw content
Fixed directory listing line selectors silently dropping the offset parameter and only applying the limit

[15.5.5] - 2026-05-27

Changed

Removed the model-facing path property from hashline edit tool parameters; hashline edit targets now come from ¶PATH headers in input.

Fixed

Fixed legacy pi-* extension loading regression where import { Type } from "@(scope)/pi-ai" (e.g. @earendil-works/pi-ai used by @plannotator/pi-extension) failed with Export named 'Type' not found after pi-ai 15.1.0 removed the root Type runtime export; the legacy-pi compat layer now redirects bare @oh-my-pi/pi-ai root imports through a sibling shim that re-exports the canonical pi-ai surface plus the Zod-backed Type runtime from the same TypeBox shim served to @sinclair/typebox imports (#1437)

[15.5.4] - 2026-05-27

Breaking Changes

Removed the package root hashline export so imports from the top-level entrypoint can no longer access hashline helpers directly

Added

Added read.summarize.minTotalLines setting (default 100) to set the minimum file length that triggers read summarization
Added <file>:<lines> support to search paths, allowing file-scoped constraints such as :N-M, :N+K, and comma-separated ranges

Changed

Changed multi-section hashline edit execution to defer LSP diagnostics flushing until the final section is written
Changed read to return verbatim contents for files shorter than read.summarize.minTotalLines instead of summarizing them
Changed search path line-range filtering to include only matches and context lines that fall inside the requested ranges

Fixed

Fixed multi-section hashline edits to reject duplicate canonical targets and preflight write guards before any section is committed
Fixed createAgentSession() dropping the hidden resolve tool from the registry when no active tool sets deferrable: true, even though plan mode dispatches the plan-approval resolve { action: "apply", ... } call through a standing handler. Read-only plan-mode toolsets (e.g. read, search, find, web_search) silently activated plan mode without resolve, leaving the agent unable to submit the finalized plan and forcing the user to exit plan mode manually. resolve is now kept whenever plan.enabled is true, so the standing handler always has a callable tool (#1428)
Fixed omp startup and /changelog reading the host project's CHANGELOG.md as omp's — getPackageDir() no longer falls back to the user's cwd when no owning package.json is locatable, preventing spurious lastChangelogVersion writes (#1423)

[15.5.3] - 2026-05-27

Breaking Changes

Disallowed inline payload on hashline ↑, ↓, and : operations (including BOF/EOF inserts), requiring payload text to be supplied on standalone + continuation rows

Changed

Warned when legacy inline LINE:TEXT lines are accepted as payload continuations only when inside a pending multi-line A-B: replacement

[15.5.2] - 2026-05-26

Breaking Changes

Changed the hashline patch format so payload continuation lines now require a leading +, rejecting unprefixed multiline payload rows that were previously accepted as fallback payload text

Changed

Changed hashline payload parsing so blank lines are only preserved when prefixed with +, so blank separator lines between operations are ignored unless explicitly marked
Changed payload escaping so a line beginning with + is now represented as ++... while the leading marker is stripped before writing
Changed the default task.simple mode from default to schema-free, so task-call schema inputs are disabled by default while shared context and user prompt/session-defined output schemas remain available
Changed tools.approvalMode: yolo to auto-approve tool calls even when a tool marks override: true; user tools.approval.<tool> policies (allow/prompt/deny) now remain the only controls for yolo mode.
Changed the hashline edit executor to coalesce two consecutive A-B: ops on the identical range last-wins (the model painted a before/after pair) and append a warning, instead of throwing anchor line X is already targeted by the :/! op on line Y. Other overlap shapes (different ranges, A-B:+!, !+!) still throw.

Fixed

Fixed nested replace parsing so line-anchored N: rows inside a pending A-B: replacement now trigger overlap errors instead of being silently folded into the replacement payload

[15.5.1] - 2026-05-26

Breaking Changes

Removed the href, hrefr, and hline Handlebars prompt helpers along with the shared hashline anchor state; none were referenced by any built-in or user prompt template

[15.5.0] - 2026-05-26

Added

Added per-tool approval declarations so each built-in, custom, or extension tool can declare its capability tier and approval prompt details.
Added OMP_MCP_TIMEOUT_MS environment variable to override MCP client request timeout for every server (in milliseconds); set to 0 to disable client-side timeouts. Invalid (negative or non-numeric) values are ignored with a warning and fall back to the per-server timeout or default 30s (#1415).
Added omp auth-broker list (with --json for machine-readable output) to enumerate supported OAuth providers, replacing bunx @oh-my-pi/pi-ai list.
Added interactive provider selection to omp auth-broker login and omp auth-broker logout when invoked without a provider argument (replacing the equivalent bunx @oh-my-pi/pi-ai login/logout flows). The logout picker is sourced from stored credentials so it only lists providers the user is actually signed in to.
Added directory matches to the find tool: glob searches now return directories alongside files, with directory hits emitted with a trailing / marker. find for **/tests no longer requires a follow-up read to surface a directory hit; tool prompt and output docs were updated to reflect that paths may be either kind.
Added the RFC 8414 §3.1 path-ful issuer form (/.well-known/oauth-authorization-server<issuer-path>) as a third candidate in MCP OAuth discovery, after origin-root and path-prefixed well-known URLs, so deployments that publish authorization-server metadata at the path-ful location resolve correctly. Single-segment authorization URLs (e.g. https://gateway/my-service) are now treated as the gateway prefix instead of being dropped back to the origin root.

Changed

Changed tool-approval prompts to use an explicit Approve/Deny selector and surface the denial reason as contextual help text instead of a bare confirm/cancel toggle.
Changed approval safety overrides to prompt even in yolo / --auto-approve sessions, so critical tool-declared patterns no longer run unattended.
Changed omp auth-broker login to drive the per-provider OAuth/API-key flow in-process via AuthStorage.login() instead of spawning the pi-ai CLI subprocess. The pi-ai bin is being removed; the same login surface now lives entirely inside omp.
Changed the default per-line truncation cap for search/grep output (DEFAULT_MAX_COLUMN) from 1024 to 512 characters to keep wide minified single-line bundles from blowing out the model's context.
Changed edit.hashlineAutoDropPureInsertDuplicates to also fire on A-B:payload replacement ops when a single non-structural boundary payload line duplicates the line immediately above the deleted range (prefix) or immediately below it (suffix). Catches the common mistake of A-B:foo where the user meant A-B! but typed a payload that happens to match adjacent context. The existing 2+-line block absorb and balance-validated structural single-line absorb already covered the multi-line and structural-delimiter cases; this closes the single-line non-structural gap.

Fixed

Fixed the agent loop and bash executor busy-spinning when scheduler waits returned early. napi uv_async_send callbacks can wake the event loop after only ~1–2 ms even when the caller asked for a longer sleep, so yieldIfDue() now compensates by retrying Bun.sleep() until the requested wall-clock duration has elapsed, and the bash executor wraps its runPromise/timeout/abort race in an ExponentialYield (20 ms → 10 s) so long-running commands stop hot-looping on the race. Yields are additionally throttled by a module-level 50 ms gate, and losing timers in the race are aborted via AbortSignal so they don't keep firing after a winner is selected (#1396 by @hezhiyang2000, closes #1384).
Fixed streaming edit previews going blank for inline-payload ops. LINE↓payload / A-B:payload are valid single-line writes, but the natural-order preview was skipping the entire op token until a newline arrived — so the first character the model typed after the sigil only appeared after the next line ended, and the renderer would fall back to rendering the raw A-B: bla bla bla input. Inline-body content on op-insert and op-replace tokens is now emitted as a +payload line on the same op tick.
Fixed /usage and the status-line reset countdown rendering stale negative deltas after a Codex window had elapsed: Codex keeps reporting the prior window's reset_at until a new request opens a fresh window, which turned the resets in … suffix into a meaningless negative duration. formatDuration now clamps non-positive, NaN, and infinite inputs to 0ms, and both renderers suppress the resets in … suffix entirely once resetsAt <= now.
Fixed auto-handoff race at the context threshold: when compaction.strategy = handoff fired at agent_end with an active checkpoint or incomplete todos, the deferred handoff post-prompt task and the rewind/todo-completion path both scheduled work concurrently, so a fresh agent.continue() streamed a new assistant turn alongside the handoff LLM call (visible as the "Auto-handoff" loader plus an assistant message still streaming, with the chat container then rebuilt mid-stream). #checkCompaction now reports whether it deferred a handoff and the agent_end handler short-circuits the rewind/todo passes; #scheduleAgentContinue also skips when isCompacting || isGeneratingHandoff. The pre-prompt #checkCompaction call now forces inline execution (allowDefer = false) so the new turn cannot begin until the maintenance settles.
Fixed /exit and Ctrl+C-double-tap hanging when a deferred handoff was mid-flight: AgentSession.dispose() now aborts retry/compaction (auto-compaction + handoff) and the agent stream before draining #cancelPostPromptTasks, so the post-prompt task awaiting generateHandoff rejects and Promise.allSettled can resolve. Tool work (bash/eval/python) is intentionally still left for the existing dispose paths so shared kernels continue to survive across session dispose.

[15.4.3] - 2026-05-26

Fixed

Fixed Google Vertex cached project discovery replacing the bundled fallback catalog so /models does not keep showing outdated Gemini entries after authoritative Vertex discovery (#1412).

[15.4.2] - 2026-05-26

Fixed

Fixed plan-mode subagents being unable to terminate because yield was registered but missing from the active tool set when requireYieldTool was combined with an explicit toolNames list (#1408)

[15.4.1] - 2026-05-26

Breaking Changes

The vim edit mode option is no longer available; configurations using edit.mode: vim will be automatically mapped to hashline mode
Hashline payload semantics are now strictly inline-first: the first payload line is whatever follows the sigil on the op line itself, and subsequent lines append after it. A newline immediately after ↑/↓/: is no longer a free separator — it produces a blank first payload line. Use LINE↓content for a one-line insert, LINE↓firstline\nsecondline for two lines; bare LINE↓ / LINE↑ / LINE: (no inline payload) still insert/replace with one blank line as before.

Added

Added irc.timeoutMs setting to configure IRC message timeout duration with a default of 120 seconds
Added timeout enforcement for IRC send operations to prevent indefinite hangs when recipients are unresponsive
Added evaluator state inheritance for task-spawned subagents so JavaScript and Python variables are visible between a parent agent and its child sessions
Added hashline-per edit mode to restore the legacy per-line hashline dialect alongside the default file-hash dialect
Added file-hash computation and validation for hashline sections to detect stale edits
Added file-read snapshot caching with multi-snapshot ring per path for recovery from agent's own writes
Added delete operation (!) support to hashline grammar for explicit line deletion
Added structural bracket/brace balance warnings when deleting lines with unclosed constructs
Added file-hash computation and validation for hashline sections to detect stale edits
Added file-read snapshot caching with multi-snapshot ring per path for recovery from agent's own writes
Added delete operation (!) support to hashline grammar for explicit line deletion
Added structural bracket/brace balance warnings when deleting lines with unclosed constructs

Changed

Changed Python shared eval sessions to be keyed by sessionId and cwd so code state no longer leaks across different directories when reusing a session
Changed shared JavaScript and Python startup to deduplicate concurrent first-time session initialization so parallel first calls share one warm session
Changed shared JavaScript and Python execution output handling so interleaved async runs keep their display output scoped to the originating run
Changed Python tool bridge to use per-run identifiers alongside session IDs for correct routing of tool responses and output in concurrent evaluations
Changed JavaScript and Python eval execution to allow overlapping asynchronous cells on the same session ID to run concurrently instead of being strictly queued
Updated the edit mode option set to support replace, patch, hashline, and apply_patch variants
Bare A: / A-B: (no payload, no inline body) now replaces the line/range with a single blank line, symmetric with bare A↑ / A↓ inserting a blank line; previously rejected as ambiguous
Simplified hashline anchor format from LINE+HASH to bare LINE numbers in edit operations
Updated hashline file headers to include 4-hex file hash: ¶PATH#HASH format for anchored edits
Changed hashline line separator from | to : in editable output (e.g., 42:content instead of 42ab|content)
Removed per-line hash validation; file-level hash now validates entire section integrity
Updated read/search output to emit file-hash headers (¶PATH#HASH) followed by numbered lines for hashline mode
Modified hashline grammar to accept optional file hash in headers and removed hash requirements from line anchors
Changed hashline diff preview format to use LINE:content instead of LINE+HASH|content
Updated prompt documentation to reflect new ¶PATH#HASH header and bare line-number syntax
Bare A: / A-B: (no payload, no inline body) now replaces the line/range with a single blank line, symmetric with bare A↑ / A↓ inserting a blank line; previously rejected as ambiguous
Simplified hashline anchor format from LINE+HASH to bare LINE numbers in edit operations
Updated hashline file headers to include 4-hex file hash: ¶PATH#HASH format for anchored edits
Changed hashline line separator from | to : in editable output (e.g., 42:content instead of 42ab|content)
Removed per-line hash validation; file-level hash now validates entire section integrity
Updated read/search output to emit file-hash headers (¶PATH#HASH) followed by numbered lines for hashline mode
Modified hashline grammar to accept optional file hash in headers and removed hash requirements from line anchors
Changed hashline diff preview format to use LINE:content instead of LINE+HASH|content
Updated prompt documentation to reflect new ¶PATH#HASH header and bare line-number syntax

Removed

Removed the installH2Fetch() activation from CLI startup; HTTPS fetches now use Bun's default transport
Removed the vim edit mode along with the VimTool module, prompt, and supporting buffer/engine/renderer stack
Removed per-line hash anchors (2-letter bigram hashes) from hashline format
Removed RANGE_INTERIOR_HASH constant; multi-line ranges no longer use ** filler
Removed HashMismatch type and hash mismatch error reporting; replaced with file-level validation
Removed per-line hash anchors (2-letter bigram hashes) from hashline format
Removed RANGE_INTERIOR_HASH constant; multi-line ranges no longer use ** filler
Removed HashMismatch type and hash mismatch error reporting; replaced with file-level validation

Fixed

Fixed JavaScript module reloading to refresh local re-exports when transitive dependency files are edited
Fixed Python tool calls in warm kernels to initialize once bridge environment variables appear after startup and to return a clear tool bridge is unavailable error when missing
Fixed IRC send handling to preserve recipient incoming messages when auto-reply timeouts instead of dropping them
Fixed Python session disposal to cancel all concurrent active executions in a shared kernel
Fixed JavaScript eval imports to preserve module-level singletons across re-imports of unchanged local files and reload them only after edits
Fixed concurrent Python evaluator tool calls to use per-run identifiers so tool responses and output are routed to the correct execution
Fixed the search tool argument validation to accept a single string paths value as a one-path search.

[15.4.0] - 2026-05-26

Breaking Changes

Replaced the hashline patch format from §, «, », ≔, and .. to ¶, ↑, ↓, → with range separators written as A-B, requiring users to migrate hashline edit inputs

Added

Added codex and gemini to the web search provider settings so users can configure OpenAI and Gemini web search directly from provider selection
Added OpenAI (codex) and Gemini web search options with updated setup descriptions for omp /login openai-codex and Gemini OAuth login
Added pretty-printing for wide JSON data: payloads in the raw provider-stream debug viewer so streamed event bodies expand across multiple data: lines instead of getting clipped by the per-line truncator, and updated the viewer header to read raw provider stream (SSE + WS) now that Codex WebSocket frames also flow through the buffer

Changed

Updated hashline operation syntax to support inline payload text on the same op line for insert and replace (ANCHOR↑/↓/...→) while still accepting payload lines that follow
Updated hashline anchor parsing so copied |TEXT decorations remain cosmetic and payload must be provided on or after the operator
Unified subagent output-schema validation into a single shared module (tools/output-schema-validator.ts) used by both the in-process yield tool (validates before the subagent yields) and the executor's post-mortem finalizeSubprocessOutput path (validates after subprocess exit). Previously each side ran its own normalizeSchema → jtdToJsonSchema → validateJsonSchemaValue chain in parallel, which was semantically equivalent but invited drift: a future tweak on one side could silently disagree with the other and cause yields that pass in-tool to fail post-mortem (or vice versa). The unification preserves both call sites' existing behavior (yield throws an actionable per-issue error for the model; executor produces a schema_violation outcome with the first issue and missing-required fields) by exposing two output formatters (formatAllValidationIssues for retries, formatValidationIssueHeadline for headlines).
Changed web search provider credential lookup to use the shared AuthStorage pipeline (getApiKey/getOAuthAccess) for API-key and OAuth auth instead of direct AgentStorage access
Changed the codex web search provider display label from Codex to OpenAI
Updated anthropic and openai/gemini web search option descriptions to reflect their native web_search/OAuth requirements

Fixed

Fixed hashline inline payload parsing so same-line payloads containing whitespace, including tab-indented text, are preserved instead of being rejected
Fixed Bun HTTP/2 transport errors (HTTP2StreamReset, HTTP2RefusedStream, and HTTP2EnhanceYourCalm) to be treated as transient so the assistant now retries automatically instead of stopping on these recoverable failures
Fixed web search OAuth-backed providers (including Codex and Gemini) to use broker-managed token retrieval and account metadata, avoiding direct token-store refresh behavior that could cause search authentication failures
Updated Tavily missing-credential feedback to prompt users to configure an API-key provider setting instead of referencing agent.db directly
Refreshed expired OpenAI Codex OAuth tokens during web_search execution and persisted the updated credentials so searches continue working after token expiry
Fixed built-in explore agent failing every invocation with schema_violation: files.0.ref: must not be present on releases prior to 15.3.2 by renaming the files[].ref property to files[].path in the agent's output schema; ref is a JTD-reserved keyword (RFC 8927) and collides with JSON Type Definition's schema-reference form, so the converter previously dropped it from the generated JSON Schema. Defense-in-depth alongside the 15.3.2 converter fix (#1379).
Increased the yield tool's schema-validation retry budget from 1 to 3 so subagents whose first structured-output attempt mismatches the declared output schema get up to three retries before the parent's post-mortem schema_violation check hard-fails the task. The tool now also surfaces remaining retry attempts and an explicit "call yield again with the corrected shape" directive in each rejection message, giving the model the context it needs to converge — particularly helpful for models like GLM that tend to invent per-element field names instead of following the declared schema.

[15.3.2] - 2026-05-25

Added

Added inline |TEXT payload support to » and « hashline insert operations, allowing single-line inserts on the op line and still supporting additional payload lines
Added support for using inline payloads with BOF/EOF inserts so |TEXT is treated as inserted content at file boundaries
Added live nested-task rendering: while a subagent is mid-flight, the parent UI now surfaces both completed nested task sub-calls and the in-flight nested snapshot (forwarded from tool_execution_update), matching the finished-result tree
Added omp auth-gateway check (and matching GET /v1/credentials/check endpoint) — probes each broker-supplied credential against its provider's auth-verifying usage endpoint and prints per-credential health, so when a multi-account pool starts returning 401s you can identify which row in the broker is the bad one. The existing /v1/usage endpoint silently drops failed credentials, which is the wrong shape for diagnosing auth — the new endpoint captures errors and surfaces the credential's id, provider, type, email/accountId, and the upstream error string. CLI groups results per-provider, exits non-zero when any credential failed, and supports --json for scripting. The probe also exercises OAuth refresh on expired tokens, so a working refresh + working access reports as ok and a revoked refresh token reports as oauth refresh failed: … instead of being masked by the cached expired access token.

Fixed

Fixed parsing of inline |TEXT payloads containing whitespace on » and « inserts, which previously failed with unrecognized-op errors
Fixed anchored insert handling so an inline |TEXT body matching the anchor line is treated as anchor decoration and no longer inserted as a duplicate

[15.3.0] - 2026-05-25

Added

Added OMP_NO_WEBP environment variable to disable WebP encoding in image resize, fixing HTTP 400 errors when attaching browser snapshots to vision models running on local llama.cpp (which uses STB library that lacks WebP support)
Fixed loop mode submitting the next prompt while a background async-job delivery turn (idle flush) was still pending, which could cause the job result to be silently dropped and make the session appear to keep firing while work was ongoing (#1294)
Fixed clipboard image paste (Ctrl+V) silently failing on WSL2 by routing image reads through a powershell.exe bridge when WSL interop is detected, since arboard returns ContentNotAvailable under WSLg (#1280)
Fixed extension ctx.ui.notify() messages emitted during session_start being cleared before the first interactive render (#1316).
Fixed append-only context mode not being recomputed after model switches — the mode was frozen at session construction time using the initial model's provider, so provider.appendOnlyContext=auto left append-only enabled after switching away from DeepSeek (or disabled after switching to DeepSeek) for the rest of the session

Fixed

Fixed clipboard image paste (Ctrl+V) silently failing on WSL2 by routing image reads through a powershell.exe bridge when WSL interop is detected, since arboard returns ContentNotAvailable under WSLg (#1280)

[15.2.4] - 2026-05-22

Breaking Changes

Replaced the legacy @@ header and +/</=/- hashline syntax with the new §PATH header and «/»/≔ operation format, so existing hashline scripts and prompts using old symbols must be updated

Added

Added one-anchor ≔ANCHOR shorthand equivalent to ≔ANCHOR..ANCHOR for single-line replace/delete

Changed

Changed ≔A..B so an omitted payload now deletes the range, and added an explicit empty payload line to keep a literal blank replacement line

Removed

Removed the hsep prompt helper and PI_HL_SEP payload-prefix configuration because hashline payloads are no longer line-prefixed

Fixed

Fixed hashline payload handling in parser and streaming preview to preserve blank lines as actual payload text until the next op, file header, or envelope marker

[15.2.3] - 2026-05-22

Breaking Changes

Changed PR and task-isolation worktree directory layout to hash-based ~/.omp/wt/<identifier>-<path-hash> style paths, replacing the previous nested encoded-repo layout

Added

Added omp worktree command (alias wt) to list and manage agent-managed worktrees under ~/.omp/wt
Added omp worktree clear to remove orphaned worktree directories, with --all to include live PR-checkouts, --dry-run for preview, and --json reporting
Added machine-readable JSON output to omp worktree list for scripted inspection
Added display.shimmer appearance setting with classic, kitt (Knight Rider K.I.T.T. scanner), and disabled modes

Changed

Changed the welcome intro animation to a 3-second eased gradient sweep with a diagonal shine highlight across the logo
Changed background job completion follow-ups to batch multiple finished jobs into a single async-result message, showing each completed job and its result in one place
Changed MCP notification follow-ups to combine multiple resource updates into a single consolidated message and suppress duplicate server/uri entries
Updated PR checkout to reuse hashPath-based worktree roots when creating and scanning worktrees for cleanup
Updated worktree cleanup logic to gracefully prune parent git metadata after removing worktree directories
Reworked working-message shimmer animation for 60fps rendering: ANSI sequences are coalesced per same-tier run instead of emitted per code point, palettes compile once and cache per active theme, and the band position is now fractional so motion is smooth at any frame rate
Switched MCP health-check and "Connecting to…" spinners from hard-coded ASCII (|/-\) to theme.spinnerFrames so they pick up the active symbol preset (braille on unicode/nerd, ASCII when explicitly themed)

Fixed

Fixed PR checkout failures when the default worktree path was already registered or occupied by stale leftovers by automatically selecting suffixed alternatives (-2, -3, etc.)
Fixed Memory tab in the settings UI not revealing or hiding the Hindsight-only rows (Hindsight API URL, Hindsight Bank ID, Hindsight Scoping, etc.) when Memory Backend was switched via the inline submenu. The selector now rebuilds its item list after every change so condition-gated rows appear/disappear immediately instead of requiring a tab switch

[15.2.2] - 2026-05-22

Fixed

Fixed RULES.md not being injected. The documented sticky-rules file at ~/.omp/agent/RULES.md and <repo>/.omp/RULES.md was never read by any discovery provider; only .omp/rules/*.md was scanned. The native provider now loads both as always-apply rules so they re-attach every turn as documented (#1266).

[15.2.1] - 2026-05-21

Fixed

Fixed compaction routing to the wrong provider when modelRoles.default is set to a different model than the active chat. Auto- and manual compaction now prefer the active session's model and only fall back to role-based candidates when the current model has no usable credentials. Previously, an Anthropic chat with modelRoles.default = openai/gpt-5 would compact through OpenAI (including the remote-compaction endpoint), even though the live conversation never used OpenAI.
Fixed context overflow not being detected when the session's model (e.g. hf:zai-org/GLM-5.1 via synthetic provider) returns a 400 with the upstream "400 status code (no body)" message wrapped inside a JSON envelope. The isContextOverflow check now matches the no-body status phrase anywhere in the error string rather than requiring it at the very start, so auto-compaction fires correctly instead of leaving the session silently stuck (#1251).
Fixed formatCapturedHttpError not extracting the error text from {"error":"string"} response bodies (only object-valued error fields were previously handled), resulting in raw JSON in error messages instead of the human-readable string.

[15.2.0] - 2026-05-21

Changed

Changed the interactive working loader and slash-command progress bars to use the active theme's accent shimmer instead of static muted text.

Fixed

Fixed false-positive hashline ~ TEXT separator-padding warning firing on YAML, JSON, Python, Markdown, TOML, and other indent-sensitive file edits. The padding check is now skipped entirely for indentation-sensitive extensions (.py, .yml/.yaml, .md/.mdx, .json/.jsonc/.json5, .toml, .rst, .tf, .nix, .coffee, .haml/.slim/.pug, .sass/.styl, .nim, .cr, .elm, .fs, …), and tightened on every other extension to flag only the ~ beta typo shape (exactly one leading space before non-space content) rather than any leading-space payload.
Fixed goal state machine: get now returns paused goals (was returning null for enabled=false), complete now works on paused goals after interrupts, create is allowed after a previous goal reaches complete status, and the goal tool is re-added to the active tool set on session reload when a paused goal is persisted. Added resume and drop ops to the goal tool so the agent can re-engage or discard a paused goal without requiring user slash commands. (#1249)

[15.1.9] - 2026-05-21

Fixed

Fixed disabledProviders still probing local discovery endpoints for Ollama, llama.cpp, and LM Studio during background model refresh. Disabled providers are now excluded before implicit and built-in discovery managers are created. (#1232)
Fixed omp acp auto-discovering host .mcp.json servers in parallel with the ACP client's session/new.mcpServers, which shadowed client-supplied MCP tools in search_tool_bm25 and the session tool registry. The ACP session factory now forces enableMCP: false, so MCP ownership stays with AcpAgent#configureMcpServers. Non-ACP modes keep on-disk discovery. (#1234)
Fixed binary omp update rollbacks so a downloaded replacement that fails post-install version verification no longer remains installed over the previous working binary. (#1240)
Fixed /force <tool> rejecting Ollama/local models before the requested tool could run; Ollama now receives a named forced choice that the provider transport narrows to the selected tool. (#1236)
Fixed web_search freezing the session when an upstream provider stalled. Bun's WinHTTP backend on Windows can silently drop AbortSignal once a TCP/TLS connection hangs (oven-sh/bun#15275, oven-sh/bun#18536), so Esc never reached the in-flight fetch and the only recovery was Ctrl+C + omp --resume. Every web-search provider's outbound fetch (anthropic, brave, codex, exa, gemini, jina, kagi, kimi, parallel, perplexity, searxng, synthetic, tavily, z.ai) now composes the caller signal with a 60s hard timeout via a shared withHardTimeout helper, guaranteeing the request settles within a minute even when Bun's abort fails to propagate. Independently, executeSearch's provider-fallback loop was masking real cancellations as ordinary provider errors and returning "All web search providers failed"; it now re-throws as ToolAbortError the moment the caller's signal aborts, so the session sees a clean cancel on every platform. (#1221)

[15.1.8] - 2026-05-20

Fixed

Fixed streaming edit previews for apply_patch and hashline jittering as the model typed +added lines. Two root causes addressed: (1) the trailing partial line of the streaming text input is now trimmed at each tick so a half-typed +added line no longer flickers; (2) the preview is rendered in the model's input order during streaming instead of re-deriving a unified diff via Diff.structuredPatch, whose coalescing previously reshuffled existing +added lines downward each time a new -removed line arrived. Existing additions now stay put and the preview only grows at the bottom while streaming. A residual trailing -removed/hunk-header block whose matching +added companion has not yet arrived is also suppressed until the additions land.
Fixed Perplexity web search appearing "logged out" roughly an hour after omp auth login perplexity. The search provider's findOAuthToken was honoring the bogus expires = login_time + 1h written by older logins (Perplexity JWTs typically omit exp because sessions are server-side) and silently dropping the credential. The loader now decodes the JWT's exp claim directly and only skips when the JWT itself is expired; tokens without an exp claim are treated as non-expiring.

[15.1.7] - 2026-05-19

Fixed

Fixed debug launch/attach failures so configurationDone no longer masks the underlying DAP launch error, early stop-outcome watchers cannot emit unhandled rejections, and directory-valued launch programs are rejected before adapter selection. (#1187)
Fixed hashline edit payloads that use a readability space after ~ by warning on separator-padding-shaped payload blocks and tightening the model prompt. (#1166)
Fixed ACP bash permission requests to include execute tool metadata and command content so clients can render command approval prompts consistently. (#1189)
Fixed the status-line fast-mode indicator (⚡) rendering for scoped service tiers (openai-only, claude-only) even when the active model's provider didn't realize them — e.g. serviceTier: "openai-only" would still show the indicator next to a Claude model the wire request couldn't apply fast mode to. The indicator now consults a new AgentSession.isFastModeActive() predicate that runs the configured tier through resolveServiceTier(tier, model.provider) and only lights up when the result is "priority" for the current model. isFastModeEnabled() keeps its scope-aware semantics so /fast on|off|toggle and /fast status continue to reflect the user's configured intent.

Added

Added scoped service tier values to the serviceTier setting: priority (OpenAI only) and priority (Claude only). They let you opt into premium processing on one provider family without paying premium costs on the other when switching models mid-session. /fast on continues to set the unscoped "priority" (active everywhere supported); /fast status and isFastModeEnabled() now report on for any scoped value too.

Changed

Changed /fast to be a single provider-agnostic toggle: enabling the command sets serviceTier: "priority" for every provider, and the anthropic-messages provider translates priority into speed: "fast" plus the fast-mode-2026-02-01 beta. Anthropic fast mode is currently supported on Claude Opus 4.6 and 4.7; the server rejects other models, which triggers the provider's auto-fallback (request retried without the priority signal, providerSessionState.fastModeDisabled persisted for the rest of the session). The session listens for the "priority" marker in AssistantMessage.disabledFeatures, syncs /fast off, and emits a warning notice. Re-running /fast on clears the per-session disable so the next request actually re-tries priority.

[15.1.6] - 2026-05-19

Fixed

Fixed plan-mode resolve looping when grammar-constrained models (e.g. Qwen3.6-35B-MTP via llama.cpp) emit extra: { title: {} } instead of a string — the open Record<string, unknown> schema for extra lets such models drop in an empty object, and the apply guard then hard-threw on every retry. Plan approval now derives the title from extra.title when usable, falling back to the plan's first # Heading, then the plan filename stem (local://PLAN.md → PLAN), then the literal "plan". Prompt language relaxed from "MUST" to "SHOULD" for extra.title. (#1179)

[15.1.5] - 2026-05-19

Fixed

Fixed ast_grep and ast_edit tool details retaining every per-file parse error — a scan over hundreds of files with syntax-error nodes inflated details.parseErrors to one entry per file, leaking into traces and the renderer's "X more" overflow. Errors are now capped at PARSE_ERRORS_LIMIT (20) at the source, with the original total preserved in a new parseErrorsTotal field for accurate count labels.

[15.1.4] - 2026-05-19

Fixed

Fixed normalizePlanTitle rejecting plan titles that contain spaces or common punctuation (e.g. "My Feature Plan") — spaces are now converted to hyphens and other invalid characters are dropped, so models that produce natural-language plan titles no longer loop forever trying to call resolve. (#1176)
Fixed ask tool prompt example showing the legacy question/options top-level format instead of the current questions: [{id, question, options}] array format; models that closely followed the example generated calls that always failed schema validation. (#1176)
Fixed ACP command and custom tool-call notifications to carry the original tool arguments in replayed and final updates, so command text is preserved and raw input is no longer wrapped
Fixed ACP async-job draining to be scoped by session owner so getAsyncJobSnapshot and drainAsyncJobDeliveriesForAcp no longer consume or expose jobs from other sessions
Fixed async job status reporting to include in-flight completions so queued/delivering indicators remain accurate while callbacks are still running
Fixed deferAgentInitiatedTurns handling during ACP async-job draining so background completion follow-up turns are delivered even when agent-initiated turns are deferred
Fixed ACP ordinary file-editing calls (edit, write, ast_edit) incorrectly requesting session/request_permission before every call, while keeping permission prompts for edit operations that delete or move files; permission requests now report the gated tool call as pending so clients can render the approval UI instead of returning Permission request cancelled without a visible prompt. (#1134 by @jiwangyihao)
Fixed the session tree selector to preserve a readable message column when deeply nested branch gutters would otherwise consume the viewport. (#1144)

[15.1.3] - 2026-05-17

Breaking Changes

Renamed the embedded-documentation internal URL scheme from pi:// to omp://. OmpProtocolHandler replaces PiProtocolHandler; update any external references accordingly.
Removed the StringEnum re-export from @oh-my-pi/pi-coding-agent. Custom tools and extensions should use z.enum([...]) directly via the injected pi.zod.
Replaced the eval tool's LARK-grammar input string with a structured cells array. Each cell is { language: "py" | "js", code, title?, timeout?, reset? }. Removed the implicit/sniffed language path, the *** Cell / *** End / *** Abort markers, and the per-cell t:<duration> unit suffixes — timeout is now seconds (1-600).

Added

Added providers.<name>.transport: "pi-native" to models.yml. When set, every model under that provider routes its streaming dispatch through the auth-gateway's POST /v1/pi/stream endpoint instead of the per-provider SDK. The provider's baseUrl must point at a compatible omp auth-gateway and apiKey must carry the gateway bearer. The slot's models.json still resolves locally for pricing/capabilities/thinking config; only the wire dispatch is redirected. Use case: containerized omp installs (robomp slots, swarm extension) where the slot must stay credential-free and a sidecar gateway holds the real provider tokens. Also surfaced as transport on ProviderConfigInput for extension-registered providers.
Added optional backend push for the auto-QA grievance database (dev.autoqaPush.enabled, dev.autoqaPush.endpoint, dev.autoqaPush.token; env overrides PI_AUTO_QA_PUSH, PI_AUTO_QA_PUSH_URL, PI_AUTO_QA_PUSH_TOKEN). When enabled, every report_tool_issue call schedules a background flush that POSTs pending rows to the configured endpoint and deletes them on HTTP 2xx. Each push carries a stable per-install UUID (installId) generated on first use and persisted at ~/.omp/install-id via getInstallId() (new export from @oh-my-pi/pi-utils), so the receiver can dedup retries across host renames and autoqa.db wipes. Single-flight, 5s request timeout, 30s in-memory cooldown after failure, and a row-id watermark so rows inserted during an in-flight push survive and ship next time. Tool execution remains non-blocking and never throws.
ModelRegistry now promotes models.yml providers.<name>.apiKey entries to AuthStorage's new config-override tier (above OAuth, below --api-key). Pinning a bearer in models.yml was previously a no-op when the broker had an OAuth credential for the same provider — the OAuth access token won and got sent unmodified to whatever baseUrl you redirected to, which an auth-gateway in front of that endpoint rightly rejected with 401. The override is now honored, and is cleared/repopulated atomically on models.yml reload (#reloadStaticModels calls clearConfigApiKeys before re-parsing). Use case: route anthropic / openai-codex to http://llm-gateway.internal:4000 with the gateway's own bearer.
Added omp auth-broker subcommand for running and consuming a hosted credential vault.
serve [--bind=host:port] — boots a local broker against the SQLite store at $AGENT_DB_PATH.
token [--regenerate] — prints (and rotates) the bearer token stored at ~/.omp/auth-broker.token.
login <provider> [--via=user@host] [--dry-run] — drives the OAuth flow locally or via SSH -L tunnel into a remote broker (callback ports pinned per provider).
logout <provider> — disables every credential for the given provider in the local SQLite store.
import <file|dir> [--provider=<id>] [--include-disabled] [--dry-run] — imports CLIProxyAPI-style JSON credential dumps (~/.cliproxy/auth/*.json). When OMP_AUTH_BROKER_URL is configured, credentials are uploaded to the remote broker via POST /v1/credential; otherwise they go into the local SQLite store. JSON type is mapped to omp providers (claude → anthropic, codex → openai-codex, gemini[-cli] → google-gemini-cli, antigravity → google-antigravity); --provider overrides the mapping for unrecognized types.
status — pings the configured remote broker (OMP_AUTH_BROKER_URL).
Added remote credential vault support to discoverAuthStorage. Configure via env (OMP_AUTH_BROKER_URL / OMP_AUTH_BROKER_TOKEN) or by setting auth.broker.url and auth.broker.token in ~/.omp/agent/config.yml (hidden from the settings UI; supports !command resolution). Falls back to ~/.omp/auth-broker.token when no token is provided inline. Otherwise behavior is unchanged.
Added omp auth-broker migrate --from-local [--include-env] [--include-oauth] [--dry-run] — uploads local SQLite credentials (and optionally env-var API keys) to the configured broker. Skips anything already on the broker via identity-key matching. OAuth is skipped by default (handled via cliproxy import). Idempotent on re-runs.
Added omp auth-gateway subcommand for running a forward-proxy that hides access tokens from less-trusted clients:
serve [--bind=…] — boots the gateway against the configured broker. Listens on 127.0.0.1:4000 by default.
token [--regenerate] — manages the gateway bearer token at ~/.omp/auth-gateway.token (separate from the broker bearer).
status — verifies gateway config and authenticated broker readiness.
One wire surface: POST /v1/chat/completions (OpenAI chat-completions), POST /v1/messages (Anthropic messages), POST /v1/responses (OpenAI Responses), GET /v1/usage (aggregated, 5-min per-credential cache), GET /v1/models (catalog). Model id in the request body selects which omp provider/model services it; the gateway translates wire format ↔ omp canonical Context and dispatches through pi-ai streamSimple(). Container deployments (robomp, etc.) get inference auth without ever holding access tokens or the broker bearer.

Changed

Changed TTSR interruptMode semantics so a non-interrupting decision on a tool-source match now folds the rule reminder into that specific tool's toolResult content instead of queuing a loop-wide deferred follow-up turn. Text/thinking matches keep the previous deferred-injection behavior.

Fixed

Fixed streaming API requests to recover from provider auth errors by invalidating stale credentials and retrying with a fresh key
Fixed auth-broker migration, auth-gateway startup, and discoverAuthStorage to fail fast with a clear error when the broker snapshot endpoint returns a non-200 response
Fixed omp auth-broker migrate to skip local placeholder <authenticated> API credentials (not real keys) when exporting to a remote broker
Fixed auth-gateway token initialization to avoid clobbering an existing token when multiple processes initialize it concurrently
Fixed omp auth-gateway request handling to reject unsupported OpenAI/Anthropic protocol controls with 400 instead of accepting and ignoring them, propagate upstream error/abort terminal states as failures, preserve Responses reasoning and completed text items, accept string/system Responses messages, and keep Anthropic tool-result ordering valid.
Fixed gateway usage reporting to include cached-token totals for OpenAI Chat/Responses and to serve the last good cached report during transient upstream usage fetch failures.
Fixed auth-gateway request cancellation for requests that are already aborted before dispatch.
Fixed /login and /logout provider selector overflowing tall provider lists off-screen on small terminals. The selector now scrolls a 10-item window centered on the highlighted entry, shows a (n/total) indicator when windowed, and accepts PageUp/PageDown for faster navigation.

[15.1.2] - 2026-05-15

Fixed

Fixed SSH host additions/removals made inside a running session not refreshing the live ssh tool. /ssh add and /ssh remove now update the model-visible host list immediately, while /reload-plugins and /move refresh SSH discovery for external or project-scope config changes without restart.
Fixed loose object output schemas in YieldTool so non-strict schemas (for example additionalProperties: true) are preserved and accepted instead of being forced into strict mode
Fixed unconstrained output schema modes (outputSchema: true or absent/non-strict schemas) to run in loose mode for successful results
Fixed bash tool calls with pty: true hanging indefinitely on Windows by falling back to the non-PTY executor instead of entering the ConPTY-backed interactive path. (#1103)

Changed

Updated MCP and theme schema metadata to reference JSON Schema draft-2020-12

[15.1.0] - 2026-05-15

Breaking Changes

Changed the extension and hook runtime API by moving schema typing from direct TypeBox imports to TSchema from @oh-my-pi/pi-ai, requiring callers who use TypeScript imports of Type to migrate via provided injected modules

Added

Added a cancellable handoff progress indicator in /handoff that displays while handoff generation runs and can be aborted with Esc
Added apiKey as a supported provider override field in model config, allowing API-key-only overrides to provide fallback credentials for built-in models
Added supportsMultipleSystemMessages, allowsSyntheticReasoningContentForToolCalls, disableReasoningOnToolChoice, and levels model-thinking compatibility fields to model configuration schemas
Added zod to the Extension, Custom Tool, Hook, and Custom Command APIs as pi.zod so extension and plugin authors can define tool schemas with Zod without separate imports
Added pi.zod as a canonical schema API for examples and extension plugins while keeping typebox available as legacy compatibility
Added a telemetry option to createAgentSession for passing OpenTelemetry configuration through to the underlying Agent

Changed

Changed handoff generation to run as a one-shot handoff request and switch to the new session only after it completes, avoiding an extra assistant handoff turn in chat history
Changed pi.typebox.Type.Composite to merge all object schemas in the provided list, enabling more than two object inputs
Changed pi.typebox.Type.Record to validate record keys against the provided key schema instead of forcing string keys
Changed pi.typebox.Type.Array with uniqueItems: true to reject duplicate items while preserving the constraint in wire schemas
Changed pi.typebox.Type.Object with additionalProperties: false to reject unknown properties during parsing
Changed pi.typebox.Type.Enum in the compatibility shim to preserve numeric TypeScript enum values
Changed tool parameter schemas across the agent to use the shared Pi schema pipeline (TSchema plus Zod/JSON Schema validation) instead of direct AJV/TypeBox compilation for stricter schema validation compatibility
Changed GitHub tool input schema shape to expose operation fields in a flat schema form without legacy run_watch-style nesting
Changed Python session pooling to remove the previous 4-session retention cap and 5-minute idle-session eviction, so kernels now stay alive for a session until explicitly disposed via disposeKernelSessionsByOwner or disposeAllKernelSessions
Changed kernel cleanup behavior to avoid automatic eviction by idle timeout and capacity pressure, so additional Python sessions are not queued behind retained-session shutdown retries
Replaced the bundled @sinclair/typebox runtime dependency with an in-repo Zod-backed shim exposed through pi.typebox.Type.*. Common builders (Object, String, Number, Integer, Boolean, Array, Tuple, Union, Intersect, Literal, Enum, Optional, Nullable, Record, Partial, Required, Pick, Omit, Composite, …) keep their existing call signatures but now return Zod schemas that flow through the same validation/wire pipeline as pi.zod. Bare @sinclair/typebox imports inside extensions are transparently remapped to the same shim by the runtime plugin shim, so plugins that authored against import { Type } from "@sinclair/typebox" keep working unchanged. Plugins that relied on TypeBox-only submodule APIs (@sinclair/typebox/compiler, @sinclair/typebox/value, TypeRegistry, the Symbol(TypeBox.Kind) marker) must vendor @sinclair/typebox in their own package — only the root import is remapped.

Deprecated

Deprecated direct TypeBox-only examples for plugin schemas by updating example documentation to prefer pi.zod

Fixed

Fixed auto-triggered handoff flow to perform only a single handoff-generation model call instead of an extra prompt cycle
Fixed handoff cancellation behavior so a pre-cancelled signal returns Handoff cancelled without starting generation and aborting handoff now propagates through the handoff request signal
Fixed create_conventional_analysis parsing to ignore harmless extra fields and still parse the required conventional fields
Fixed BashTool async request validation flow so async execution remains disabled and returns the explicit Async bash execution is disabled error
Fixed task.simple invalid schema and context argument handling to still reject unsupported fields after tool-argument validation
Fixed subagent execution hangs by enforcing task.maxRuntimeMs as a wall-clock limit even when inference streaming stalls, so stuck subagents now abort and report runtime-limit exceeded
Fixed tool schema compatibility validation by routing TypeBox schemas through shared conversion and Zod-based validation to avoid strict-schema provider mismatches
Fixed Python execution cancellation and timeouts by escalating to kernel shutdown if SIGINT did not terminate a running cell within 2 seconds, preventing indefinite hangs in queued or stuck sessions
Fixed cleanup blocking during long-running executions by forcing a kernel shutdown path when interrupt-based cancellation is ignored
Fixed bash output emitting a spurious [… 0 lines elided (NB) …] marker (and reordering the artifact link before the command output) when the shell minimizer rewrote a small command's output. After OutputSink.replace() swapped the minimized text into the buffer, the subsequent sink.push("[raw output: artifact://N]\n") chunk was funneled back into the (now empty) head-retention window while the pre-replace #totalBytes still tracked the original raw stream — so dump() composed <head=artifact-link> + <middle-elision marker against stale totals> + <tail=minimized text> instead of <minimized text> + <artifact link>. replace() now realigns #totalBytes/#totalLines/#sawData/#truncated to the authoritative buffer and disables head retention for the lifetime of the sink, so further pushes append to the tail buffer in order. The bash executor also drops the leading \n on the artifact-link push when the minimized text already ends with one so the separator stays single-newline.
Fixed legacy plugin extensions failing to load on Windows when they import a bare-specifier dependency from their own node_modules (e.g. import YAML from "yaml" in supipowers). The legacy-pi mirror resolved the dependency to its absolute path and then ran the path through isUrlLikeSpecifier, whose ^[A-Za-z][A-Za-z\d+.-]*: regex matched the Windows drive letter (C:) and short-circuited the pathToFileURL conversion. The raw path was emitted into the mirrored TS source as import x from "C:\\Users\\...\\dep\\dist\\index.js", where \n, \U, \y and other backslash sequences were eaten by the TS string-literal parser, producing nonsense package specifiers like C:Usersjames.ompagentextensionssupipowers\node_modulesyamldistindex.js that Bun's resolver rejected with Cannot find package …. isUrlLikeSpecifier now rejects ^[A-Za-z]:[\\/] first, so Windows absolute paths flow through pathToFileURL like every other absolute path and reach the mirror as proper file:///C:/... URLs.
Fixed Python session queued executions silently resurrecting kernels after disposeAllKernelSessions or disposeKernelSessionsByOwner removed the session: queued work now checks the session is still registered before replacing or executing on a kernel and rejects with cancellation otherwise
Fixed Python session disposal treating an unconfirmed PythonKernel.shutdown() result as success: sessions whose kernel shutdown returns { confirmed: false } (or rejects) are now retained in the registry and a warn is logged so a later dispose can retry instead of orphaning the subprocess
Fixed task.maxRuntimeMs losing wall-clock aborts that fired during pre-prompt session setup by re-checking the abort signal immediately before issuing the model prompt, so a stalled subagent now exits with the runtime-limit reason instead of hanging through setup races
Fixed late yield events landing after a wall-clock timeout from flipping a timed-out subagent to a successful exit, so the reported aborted flag and exit code now always reflect the runtime-limit breach while yield payloads remain in extractedToolData
Fixed async-task progress consumer to copy contextTokens and contextWindow from the completed SingleResult onto AgentProgress, so UI gauges keep showing per-turn context after a backgrounded task finishes
Fixed the status-line path segment ignoring stripWorkPrefix: false when selecting the folder icon for scratch directories. The icon selection now respects the same gate as the scratch path stripping, so disabling stripWorkPrefix keeps the regular folder icon even when the project directory is inside a scratch root.
Fixed YieldTool constructor to fall back to the loose record schema when the session outputSchema contains unresolved $ref strings (e.g. external or cyclic references that survive dereferencing), instead of installing a validator that would reject every payload with an unresolved-reference error
Fixed pi.typebox.Type.String dropping minLength/maxLength/pattern constraints when a format (e.g. email, url, uuid) was also supplied; length and pattern checks are now applied to the format-specific schema instead of being gated on an instanceof z.ZodString check that never matched the format subclasses.
Fixed pi.typebox.Type.Object stripping unknown properties when constructed without explicit additionalProperties. TypeBox preserves extras by default, so the shim now installs .loose() for the omitted/true cases while keeping .strict() for additionalProperties: false and .catchall(schema) for a schema value.

[15.0.2] - 2026-05-15

Added

Added the set_host_uri_schemes RPC command so hosts can register and replace writable/read-only internal URI schemes with scheme metadata (writable, immutable) at runtime
Enabled the write tool to dispatch write(url, content) to registered internal URL handlers, allowing edits to non-filesystem resources via host-managed URI schemes
Added host-owned internal URI read/write over RPC, including abort support, so URI operations are resolved by the host transport for read and write requests
Added handling of host URI request results in RPC mode so host services can stream completion frames for internal URI operations
Added scratch-directory awareness to the status-line path segment. When the project directory is inside an OS-level scratch root (the platform os.tmpdir(), /tmp and /var/tmp plus their macOS /private/... aliases, ~/tmp, or — on Windows — %TEMP% / %TMP% / %SystemRoot%\Temp), the segment now (1) renders the new icon.scratchFolder symbol instead of icon.folder, and (2) strips the scratch root from the displayed path so only the trailing folder (and any subpath beneath it) is shown — mirroring how /work and ~/Projects are already abbreviated. Both behaviors honor the existing stripWorkPrefix option. Icon defaults: 🗑 (emoji), `` (nf-fa-trash) for Nerd Font, [T] for ASCII, ◌ in the poimandres themes; themes can override icon.scratchFolder independently of icon.folder.

Changed

Changed the github tool's search ops (search_issues, search_prs, search_code, search_commits) to default the repo scope to the current checkout's owner/repo when repo is omitted. The auto-scope is skipped when the query already carries an explicit repo:/org:/user:/owner: qualifier or when gh repo view cannot resolve a github remote (in which case the search proceeds across all of GitHub as before). search_repos is unchanged — repository-scoping there must live in the query.
Changed bash command preprocessing to strip trailing | head and | tail pipelines (including |&) from each top-level segment in command chains separated by ;, &&, ||, or &
Changed bash fixup notices to state that stderr is already merged into stdout and to reflect that fixes were applied for multiple stripped segments when several transforms fire
Changed shell-minimizer per-line truncation marker from a bare … to …[+N], where N is the count of dropped Unicode scalars. The bracketed tally disambiguates minimizer-driven cuts from genuine … characters in the source (paths, JSON, stack traces, etc.) and gives the agent an exact count so it can decide whether the missing tail is recoverable inline or warrants reading the [raw output: artifact://<id>] footer the bash wrapper already emits when the minimizer rewrites output. Affects pipeline Stage 5 (truncate_lines_at in defs/*.toml) and the internal callers in filters/git.rs, filters/listing.rs, and filters/lint.rs. (#1046)
Changed bash command preprocessing to use the real brush-parser AST via pi-natives applyBashFixups instead of a hand-rolled top-level mask scanner. The previous regex/character-walking implementation reimplemented quote/heredoc/$(...) tracking with conservative bail-outs (notably refusing to fixup commands containing here-strings); the AST-driven version inherits the full shell parser, so semantics-preserving rewrites like stripping | head -5 off cat <<<'content' | head -5 now succeed instead of being skipped. No public API change — applyBashFixups(command) returns the same { command, stripped } shape.

Fixed

Fixed bash command fixups to remove a redundant standalone trailing 2>&1 redirect when no other pipe or redirection remains
Fixed command-fixup notices to list all stripped segments instead of reporting only one
Fixed summarized read output stalling agents on elided regions by appending an explicit footer like [NN lines across MM elided regions; read <path>:raw or a line range like <path>:1-9999 for verbatim content]. The footer fires whenever the structural summarizer elided at least one span, so the model gets a concrete recovery selector instead of having to guess from a bare ... / { .. } marker. Surfaces elidedLines on ReadToolDetails.summary alongside the existing elidedSpans. (#1046)
Updated the read tool prompt to describe the new elision footer and instruct the model to follow :raw (or an explicit line range) when the elided body is actually needed, rather than guessing.
Fixed plugin extensions failing to load when their peerDependencies reference internal pi-* packages under any scope other than @mariozechner (e.g. Cannot find module '@earendil-works/pi-tui' from @juicesharp/rpiv-ask-user-question, or Cannot find module '@oh-my-pi/pi-utils' from @oh-my-pi/swarm-extension). The legacy-pi specifier shim now treats @mariozechner, @earendil-works, and the canonical @oh-my-pi itself as aliases for the same set of bundled in-process packages (pi-agent-core, pi-ai, pi-coding-agent, pi-natives, pi-tui, pi-utils), and additionally rewrites the upstream-only pi-ai/oauth subpath onto our pi-ai/utils/oauth layout. Restored the Key runtime helper export on @oh-my-pi/pi-tui to match upstream — plugins using Key.enter / Key.ctrl("c") (e.g. @plannotator/pi-extension, @juicesharp/rpiv-ask-user-question) no longer fail with Export named 'Key' not found. End-to-end verified against @juicesharp/rpiv-ask-user-question, @oh-my-pi/swarm-extension, and @plannotator/pi-extension — each now loads cleanly with all of its tools/commands/handlers registered. Plugins importing any of those scopes are remapped to the omp binary's own copy at load time, so peer deps are no longer dragged in from npm and there is exactly one module instance per package regardless of which scope name the plugin's manifest happened to declare.
Fixed omp commit hanging after a successful commit instead of returning to the shell. The command now mirrors the runPrintMode exit pattern and calls postmortem.quit(0) once the pipeline resolves so lingering HTTP/2 keep-alive sockets, the Settings autosave timer, and other AgentSession background handles don't keep the event loop pinned. (#1041)
Fixed hashline payload parsing to silently treat truly-blank lines as empty ~-prefixed payload lines when more payload follows in the same run. The previous behavior broke at the blank ("payload line has no preceding +, <, or = operation.") even though the intent is obvious — the only ambiguity is between in-payload blanks and end-of-section blanks, and a one-line lookahead resolves it: blanks that precede a non-payload op still end the run cleanly as section separators. Recovers the common case of forgetting the leading separator on a blank inserted line without changing how trailing blanks between ops behave.
Rewrote the hashline edit prompt examples to use an ASCII-only TITLE = "Mr" → "Mrs" / "Dr" motif instead of the previous " • " and "·" separators. Some agents had been copying the middle-dot literal characters into real edits as if they were format scaffolding (e.g. emitting payload lines like ~ ·), since the demo inserts were near-twins of the existing string. The new example keeps every original op shape (single-line replace, multiline replace, insert AFTER/BEFORE, append, delete, blank, plus both anti-patterns) but uses content that is obviously domain-specific and clearly distinct from any payload separator. Pure prompt change; no parser, schema, or runtime behavior is affected.
Fixed startup fallback-chain validation to recognize cached runtime-discovered standard provider models, including Ollama Cloud models listed by --list-models, so retry.fallbackChains no longer warns that valid ollama-cloud/<model> selectors are unknown. (#1052)
Fixed discoverAgents() ignoring disabledProviders for the claude-plugins provider. Plugin roots from ~/.claude/plugins/ were scanned unconditionally, so agents from Claude Code marketplace plugins continued to appear in /agents and the Agent Control Center even when disabledProviders: [claude-plugins] was set. The discovery path now checks isProviderEnabled("claude-plugins") before calling listClaudePluginRoots(), matching how every other capability respects the disabled-providers set. (#1075)
Fixed $env:VAR PowerShell variables being mangled on Windows when commands invoked PowerShell as a subprocess (e.g. powershell -Command "Write-Host $env:SystemRoot"). Brush-core applied POSIX parameter expansion to $env before spawning the child, leaving a dangling :NAME. The fix lives in pi-shell at env-var application time: every brush session now defines env=$env as an internal shell variable so $env:NAME expands to the literal $env:NAME token that PowerShell expects. The fallback is not exported, only influences brush's own expansion, and is shadowed by any user assignment to env (e.g. env=prod; echo "$env:8080" still prints prod:8080), so the POSIX bash contract is preserved. (#1079)

[15.0.1] - 2026-05-14

Breaking Changes

Removed the dedicated exit_plan_mode tool and its prompt, requiring plan-mode completion to use the existing resolve tool path instead

Added

Added optional extra metadata object to the resolve tool so callers can pass context-specific payloads, including plan approval titles
Added hide: true frontmatter option for skill SKILL.md files. Hidden skills are still loaded and remain reachable via skill://<name> URLs and (when enabled) /skill:<name> slash commands, but are omitted from the rendered system prompt's <skills> listing so the model won't auto-discover them. Use for skills the user opts into explicitly rather than ones the model should pick up from descriptions.
Added middle elision for streaming tool outputs (bash, ssh, python, js eval) and post-execution tool result spill. When tools.artifactHeadBytes is set (default 20 KB), large outputs now keep both the first N KB and the last N KB with an inline [… N lines elided (M KB) …] marker between them, instead of dropping everything before the trailing tail. Setting tools.artifactHeadBytes = 0 reverts to the previous tail-only behavior. The full output is still mirrored to the session artifact (artifact://<id>) regardless of elision mode. Exposes truncateMiddle and formatMiddleElisionMarker from @oh-my-pi/pi-coding-agent/session/streaming-output, extends OutputSinkOptions with headBytes, and adds direction: "middle" plus headRange / tailRange / elidedLines / elidedBytes to TruncationMeta.
Added per-line column cap shared across streaming tool outputs (bash, ssh, python, js eval) and the read tool. Lines wider than tools.outputMaxColumns bytes (default 768) are ellipsis-truncated at write time and remaining bytes up to the next \n are dropped — bounded memory even on multi-MB single-line outputs (e.g. cat /dev/urandom). The cap lives on OutputSink as the new maxColumns option, persists state across chunk boundaries so split-mid-line writes still respect the budget, and exposes columnDroppedBytes / columnTruncatedLines on OutputSummary. Middle-elision byte math subtracts column drops so the "elided from middle" count stays honest. read reuses the same setting but trims its already-collected lines via truncateLine. Skipped when the read selector is :raw. The artifact file (artifact://<id>) keeps the full uncapped stream. Set tools.outputMaxColumns = 0 to disable.
Added Bun HTTP/2 fetch opt-in. Dev scripts (bun run dev, bun run stats) now pass bun --experimental-http2-fetch so every fetch() advertises h2 in the TLS ALPN list and falls back to HTTP/1.1 when the server doesn't select it. Multiplexing collapses parallel requests to the same origin onto one TLS connection. For the installed omp binary, export BUN_FEATURE_FLAG_EXPERIMENTAL_HTTP2_CLIENT=1 in your shell to enable the same behavior (the flag has to be set before Bun starts; process.env from inside JS is too late). Requires Bun 1.3.14.
Added per-subagent cost display ($X.XX in the task progress tree and the session-observer stats line). Cost is accumulated incrementally from message_end events and shown only when non-zero, using the statusLineCost theme color. Providers that do not report per-turn cost data (e.g. subscription/OAuth usage) continue to show nothing.

Changed

Changed plan-mode completion to use resolve { action: "apply", reason, extra: { title } } to request plan approval rather than calling exit_plan_mode
Changed resolve pending-action previews to trim and truncate long reason text for cleaner status-line rendering
Raised the image downscaling default JPEG quality from 75 to 80 in resizeImage output generation
Changed image resize metadata notes from coordinate-scale hints to a simple Image resized from <original> to <displayed> message and hide the note when the resized dimensions are unchanged
Removed utils/image-convert.ts and its convertToPng helper; callers now inline new Bun.Image(bytes).png().toBase64() from Bun.Image (Bun 1.3.14+).
Changed image decode/resize/encode in utils/image-resize.ts from the native PhotonImage binding to Bun.Image. Same PNG/JPEG/WebP quality+dimension ladder, but pipelines run off-thread on Bun's statically-linked codecs with no native-addon round-trip. Bumped the minimum Bun runtime requirement to 1.3.14.
Changed search pagination in multi-file scopes so skip now skips entire files and pages results in groups of up to 20 files, with output guiding the next skip value via Showing files X-Y of N
Changed multi-file search result selection to cap each file at 20 matches and round-robin across files, so one noisy file no longer suppresses visibility of hits in other files and truncation now reports per-file limits
Changed search truncation metadata/renderer output from match/result-based limits to file-based limits (fileLimitReached, perFileLimitReached) and updated truncation labels accordingly
Lowered read.defaultLimit default from 500 to 300 lines, and split the per-range context padding into asymmetric RANGE_LEADING_CONTEXT_LINES = 1 / RANGE_TRAILING_CONTEXT_LINES = 3 (was symmetric RANGE_CONTEXT_LINES = 3). Replay analysis over post-summarizer sessions (scripts/session-stats/optimize_read_config.py) showed that bare-path reads are over-provisioned at the median (file p50 = 220 lines) and that most follow-up reads are disjoint hops rather than adjacent extensions — so a smaller default plus narrower leading context reclaims tokens without measurably changing first-cover rate. Trailing context stays at 3 lines to keep anchor-stale recovery on narrow reads. Explicit read.defaultLimit overrides in settings are honoured unchanged.

Fixed

Fixed abrupt process termination data loss during session persistence by moving steady-state session writes to a synchronous path that writes each entry to the kernel page cache before returning
Fixed --help startup to avoid a config/model-registry load cycle so the root CLI help command now exits successfully in a clean environment
Queued /skill:<name> [args] invocations now show as compact Steer: /skill:<name> [args] / Follow-up: /skill:<name> [args] chips in the pending-messages bar and disappear when the agent consumes the queued message (parity with plain-text steer/follow-up). Previously the queued skill was invisible while queued and rendered as a full skill block at consumption with no chip ever appearing.
Plan-mode "Approve and compact context" no longer surfaces a red "Operation aborted" line on the plan-mode assistant message; the silent transition into compaction now renders cleanly on both live and replay paths. Real user-cancel aborts on unrelated turns and the existing "Compaction cancelled" path are unchanged.
Auto-recover conflict-resolution write/read paths that the agent malformed as <file>:conflict://<N> (or <file>:conflict://*) by mixing the :conflicts read selector with the conflict:// scheme. The stripped <file>: prefix is stored on ParsedConflictUri.recoveredPrefix and, for writes, surfaces as a trailing note in the result text so the agent learns the correct shape. Clean conflict://… URIs are unchanged.
Fixed hashline edit renderer leaving a stray @ in the displayed file path when the agent emitted a canonical @@ PATH header (or any @-run longer than one). Titles like Edit: @ packages/foo.ts now render as Edit: packages/foo.ts, matching the actual parser in hashline/input.ts which already strips every leading @ before resolving the path. Purely cosmetic — the edit itself was always routed to the correct file.
Fixed model contextWindow and maxTokens defaulting to UNK_CONTEXT_WINDOW (222222) / UNK_MAX_TOKENS (8888) when cached or freshly-discovered provider models replace bundled models through ModelRegistry.#mergeResolvedModels. The merge now preserves the bundled model's values when the replacement only has sentinel fallbacks.
Fixed headless browser.open tab startup on slow Chromium target enumeration by making worker-side stealth user-agent target setup selective, bounded, and best-effort for non-active targets. Worker startup errors are now surfaced directly instead of degrading into the generic tab worker initialization timeout.
Fixed token display for sessions and subagents inflating far beyond the context window. token_total status-line segment and the subagent overlay token counter now show input + output + cacheWrite instead of input + output + cacheRead + cacheWrite. With prompt caching, cacheRead per turn equals the full cached context — summing it across all turns produces a cumulative total that is N×context_size (e.g. a 5-turn session with a 1 M-token context reported ~5 M tokens). Cache activity is still visible via the dedicated cache_read/cache_write status-line segments; billing cost is unaffected.
Fixed ACP clients missing config_option_update notifications when the thinking level changed via any path other than the client's own session/set_session_config_option call (slash commands, model auto-adjust, extension UI). AgentSession now emits a thinking_level_changed event from setThinkingLevel, and AcpAgent subscribes to each managed session for the session's lifetime and pushes a fresh config_option_update whenever the effective level changes — independent of any active prompt turn. The subscription is installed inside #scheduleBootstrapUpdates's 50 ms timer so it shares the same race guard that prevents Zed's Received session notification for unknown session drop when notifications fire before session/new (or fork) returns; the pre-bootstrap thinking level is reported in the response's configOptions. The session/set_session_config_option handler keeps its own push only when the subscription has not yet been installed, so client-driven thinking changes still notify pre-bootstrap, post-bootstrap they flow through the subscription exactly once. Subscriptions are released in #disposeSessionRecord.
Fixed MCP OAuth refresh failing with HTTP 401 invalid_client for servers that require Dynamic Client Registration (RFC 7591) and have no oauth.clientId configured (e.g. mcp.linear.app). MCPOAuthFlow registered a fresh public PKCE client on each authorize and discarded the issued client_id once the flow object went out of scope; refresh then called the provider's /token endpoint without a client_id. The flow now exposes resolvedClientId / registeredClientSecret getters, MCPCommandController#handleOAuthFlow returns them alongside credentialId, and both the initial-connect and /mcp reauth paths persist them into auth.{clientId,clientSecret} (used at refresh) and oauth.{clientId,clientSecret} (used by subsequent /mcp reauth to skip re-registration). The MCPAddWizard onOAuth callback type is now Promise<MCPAddWizardOAuthResult> and #launchOAuthFlow folds the registered credentials into wizard state. Servers with a statically-configured oauth.clientId (Notion, Slack, Datadog) are unaffected — #tryRegisterClient short-circuits and the write-back is a no-op. (#1061 by @ldx).

[15.0.0] - 2026-05-13

Breaking Changes

Removed op: issue_view and op: pr_view from the github tool. Read single issues/PRs via the read tool against issue://<N> / pr://<N> (or the long form issue://<owner>/<repo>/<N> / pr://<owner>/<repo>/<N>); append ?comments=0 to drop the comments section. The issue and comments parameters were removed from the tool schema since no remaining op consumes them. Mutating ops (pr_create, pr_checkout, pr_push), repo_view, search_*, and run_watch are unchanged.
Removed op: pr_diff (along with the nameOnly and exclude schema fields) from the github tool. Read PR diffs through the new pr:// URL family: pr://<N>/diff for the changed-file listing, pr://<N>/diff/<i> for a single file slice (1-indexed), and pr://<N>/diff/all for the verbatim unified diff. Long-form pr://<owner>/<repo>/<N>/diff[/…] works the same way. All three variants share one gh pr diff invocation through a new pr-diff cache row, so the listing and per-file slices reconstruct from cached bytes without re-shelling. Diff content is served as text/plain so the read tool's line selectors (e.g. pr://<N>/diff/all:200-400) page the cached output without falsely advertising hashline anchors.
Renamed ACP custom extension methods from omp/* to _omp/* to comply with the ACP spec's _-prefix requirement for non-spec methods; existing callers must update method names

Added

Added markdown rendering for read results when content type is text/markdown, so GitHub internal-URL outputs are shown as formatted markdown instead of plain code blocks
Added pr://<N>/diff, pr://<N>/diff/<i>, and pr://<N>/diff/all internal-URL shapes covering changed-file listings, per-file slices, and the full unified diff. They share one pr-diff SQLite cache row with the same TTL knobs as pr://<N> views (github.cache.softTtlSec / github.cache.hardTtlSec / github.cache.enabled). Single PR views now advertise the diff entry point via a Diff: pr://<owner>/<repo>/<N>/diff note. Cache schema bumped to user_version = 3; older rows are dropped on first open to add credential-scoped keys and relax the kind CHECK constraint.
Added issue:// / pr:// internal-URL schemes that share a SQLite-backed cache with the rest of the github tool. Single-item reads (issue://<N>, issue://<owner>/<repo>/<N>) return rendered markdown and within github.cache.softTtlSec (default 5 minutes) skip the gh round-trip entirely; within github.cache.hardTtlSec (default 7 days) the cached row is returned and a background refresh is scheduled. Root and repo-scoped reads (issue://, pr://owner/repo) issue a live gh issue list / gh pr list for browsing, supporting ?state=open|closed|all for issues, ?state=open|closed|merged|all for PRs, and ?limit=, ?author=, ?label= query params. Rendered output lands in ~/.omp/cache/github-cache.db (override via OMP_GITHUB_CACHE_DB); disable the cache entirely with github.cache.enabled = false. Cwd→default-repo lookups (gh repo view) are memoized per-process.
Added new Approve and compact context choice to the ExitPlanMode approval selector. Sits between Approve and execute (purge session) and Approve and keep context (full transcript) — runs /compact on the plan-mode transcript with a planning-specific summarization hint, then dispatches the plan-approved execution turn so it lands on a fresh cache anchor with the summarized rationale carried over. Cancelling the compaction (Esc or any other abort source) defers the execution dispatch and surfaces a warning so the operator can resubmit manually; non-abort failures proceed best-effort.
Added CompactionCancelledError typed sentinel and CompactionOutcome ("ok" | "cancelled" | "failed") return type to @oh-my-pi/pi-coding-agent/session/compaction. CommandController.executeCompaction and handleCompactCommand now return the outcome instead of void so callers can discriminate user-driven aborts from generic failures without inspecting error messages.
Added a credential_disabled extension event so extensions can subscribe via pi.on("credential_disabled", handler) and react when AuthStorage automatically soft-disables a credential (e.g. OAuth invalid_grant). Replaces the current agent_end errorMessage regex pattern downstream extensions have to match against. Handler payload is { type, provider, disabledCause }. createAgentSession() subscribes the per-session extension runner to the shared AuthStorage via authStorage.onCredentialDisabled(...) at the very top of session creation — before any startup model probes run — so events fire on every disable regardless of whether the embedder also has a constructor onCredentialDisabled handler attached. The SDK forwards through ExtensionRunner.emitCredentialDisabled(event), which buffers events until runner.initialize(...) runs in the mode controller and then flushes them through emit() so extension handlers see populated UI/runtime context (rather than the constructor's no-op default with hasUI=false, an unset model, and no-op runtime actions). On session.dispose() the subscription is unsubscribed; the embedder's constructor-attached listener keeps firing through its own permanent subscription. The outer createAgentSession() catch also releases the subscription if startup throws before the dispose-wrap is wired, so repeated retries don't accumulate dead listeners.
Added omp acp subcommand for launching as an ACP (Agent Client Protocol) server over stdio
Added explicit type discriminators to ACP initialize auth methods, including a terminal setup method gated on clientCapabilities.auth.terminal
Added ACP equivalents for the remaining TUI slash commands (/jobs, /changelog, /dump, /copy, /hotkeys, /extensions, /agents, /model, /plan, /loop, /btw, /login, /logout, /resume, /tree, /branch, /new, /drop, /handoff, /fork, /session delete, /export, /share, /todo, /memory, /move, /mcp, /ssh, /marketplace, /plugins) so ACP clients reach feature parity with the TUI for non-interactive flows
Added ACP plan mode: when plan.enabled setting is on, ACP session/new/load/resume/fork advertise a plan mode alongside default; session/set_mode toggles plan-mode state so the next agent turn injects the plan-mode system prompt
Added ACP ClientBridge abstraction (packages/coding-agent/src/session/client-bridge.ts) that routes tool I/O through the connected client when capabilities are advertised at initialize; populated from AgentSideConnection in ACP mode
Added ACP terminal/* routing for bash: when the client advertises terminal: true, the tool creates a client-side terminal, embeds its terminalId on the live tool card, polls output, and releases the handle on exit or abort
Added ACP fs/read_text_file and fs/write_text_file routing for the read and write tools: when the client advertises fs.readTextFile / fs.writeTextFile, plain-text reads/writes go through the editor (surfacing unsaved buffer content and letting the editor track agent writes); falls back to disk only for reads, throws on bridge write failures
Added ACP session/request_permission gate around bash, edit, write, and ast_edit when an ACP client is connected; remembers allow_always / reject_always decisions per tool for the session lifetime
Added diff ToolCallContent emission for edit tool results: per-file oldText/newText is threaded through EditToolPerFileResult / EditToolDetails so ACP clients can render inline diffs
Added richer ACP StopReason mapping (max_tokens, refusal, cancelled) derived from the last assistant message's internal stop reason; previously only end_turn/cancelled were emitted
Added _meta.messageCount and _meta.size on session/list SessionInfo entries
Added ACP tool_call_update locations refresh from in-flight tool args and final result details so clients can "follow along" multi-file edits in real time

Changed

Changed issue and pull-request list entries to link to repository-qualified URLs (for example issue://owner/repo/<N>) so list items open correctly outside the default repo
Aligned prompt instruction language by defining NEVER and AVOID as strict aliases for MUST NOT and SHOULD NOT in the system prompt, and standardized agent, tool, and system prompt templates to use those terms consistently
Changed --mode acp to apply the same stdout-quiet overrides as --mode rpc so no banner or status text leaks into the JSON-RPC channel
Changed ACP startup to no longer require a configured model so registry validators and clients can complete initialize and authenticate before any model is selected

Fixed

Deferred flushing of buffered credential_disabled events during extension runner initialization to a microtask so handler failures are now routed through onError() registrations made immediately after initialize(), preserving extension error reporting
Fixed eval tool dynamic await import("./relative.ts") calls failing with Cannot find module ... from .../eval/js/shared/runtime.ts. The static-import rewriter only handled ImportDeclaration nodes, so dynamic-import call expressions resolved their specifier against the worker module's URL instead of the session cwd. The rewriter now walks the full AST and additionally swaps import callees in CallExpression nodes for __omp_import__, which forwards the optional options bag verbatim to native import() so { with: { type: "json" } } round-trips. Renamed rewriteStaticImports → rewriteImports.
Fixed pr://<owner>/<repo>/diff URLs for repositories named diff to continue resolving to PR list lookups instead of being parsed as short-form diff links
Fixed issue://<N>/diff short form previously misparsing as a repo named <N>/diff and surfacing a confusing GraphQL "Could not resolve to a Repository" error. The numeric-host disambiguation that already routed pr://<N>/diff through the diff path now applies to both schemes, so issue://<N>/diff[/…] falls through to the existing "Issue views do not have a diff; use pr://///diff for pull requests." rejection — matching the long-form issue://<owner>/<repo>/<N>/diff behavior. <scheme>://<owner>/diff listings for a repo literally named diff are unchanged.
Fixed PR unified diff parsing so changed-file headers with quoted paths (such as paths containing spaces) are now detected correctly and hunk content lines beginning with ---/+++ are counted in additions/deletions
Fixed GitHub view caching to account for active credential identity and avoid serving cached issue/PR data across different account/token contexts
Fixed read call tracking so calls without an explicit path or URL target no longer appear as regular file reads in the execution tracker
Fixed createAgentSession() subscribing the credential_disabled bridge to a freshly discovered AuthStorage orphan when an embedder supplied only options.modelRegistry (no options.authStorage). Refresh failures emitted by modelRegistry.getApiKey() flow through modelRegistry.authStorage, so a divergent local instance silently swallowed every disable event and also leaked into the mcpManager and session result. The SDK now reconciles authStorage to modelRegistry.authStorage up front and rejects mismatched options.authStorage/options.modelRegistry.authStorage pairs at session construction.
Fixed runSubagent (subagent task executor) carrying the same latent AuthStorage/ModelRegistry divergence as createAgentSession(): when only options.modelRegistry was supplied, the executor previously fell through to a fresh discoverAuthStorage() and handed that orphan into createAgentSession() alongside a registry whose .authStorage was a different instance. The executor now reconciles to modelRegistry.authStorage before any further work and rejects mismatched options.authStorage/options.modelRegistry.authStorage pairs the same way the SDK does, so subagents can no longer silently observe a different storage view than their parent.
Fixed github tool's search_issues/search_prs/search_code/search_commits/search_repos ops always returning 0 results when the query contained more than one qualifier (e.g. is:merged is:pr, is:open author:foo). gh search … since the advanced_search=true rollout in gh 2.92 silently wraps multi-token positional queries in parentheses and quotes everything after the first qualifier as that qualifier's value (is:"merged is:pr"), which GitHub then matches as a literal state filter that no PR can satisfy. The tool now calls gh api -X GET /search/<endpoint> -f q=… -F per_page=… directly so the qualifiers reach GitHub's search API verbatim. is:issue/is:pr and repo:<owner>/<repo> are appended internally to preserve the previous CLI-flag behavior; the user-facing query string in the formatted output is unchanged. state for merged PRs is derived from pull_request.merged_at so the rendered State: line stays merged/closed/open as before.
Fixed read tool renderer rendering failed reads with a success check (✓) and styling the error message as file content while the surrounding box was red. The renderer now branches on isError for both file and URL paths: header shows ✘ Read <path> with a proper error icon and the underlying message is rendered as an error line. renderReadUrlResult got the same treatment so failed URL reads also get the cross icon instead of falling through to the "No response data" Text fallback. Mirrors the bash/find renderer error pattern.
Fixed ACP mode to advertise and handle non-TUI builtin slash commands and /skill:<name> commands
Fixed ACP session/resume and session/close to dispatch correctly under SDK 0.21 by renaming unstable_resumeSession / unstable_closeSession to the stable resumeSession / closeSession method names the SDK now routes to
Fixed ACP tool_call / tool_call_update locations to always emit absolute paths (resolved against the session cwd) so editor clients can reliably open or focus the referenced file
Fixed ACP edit diff metadata for moves to point at the destination path rather than the now-deleted source so post-edit "open file" actions land on the new file
Fixed ACP session/request_permission locations to be absolute and to honor the requestPermission capability bit instead of only checking for the method, matching the read/write/bash capability gating
Fixed ACP authenticate to reject methodId values that were not advertised by initialize so malformed clients fail fast instead of being treated as authenticated
Fixed ACP mode changes made via session/set_session_config_option (MODE_CONFIG_ID) to also emit a current_mode_update notification, matching session/set_mode so clients tracking modes.currentModeId stay in sync
Fixed /model ACP builtin to emit a config_option_update after switching models so clients show the new model in config selectors immediately
Fixed /mcp list (ACP) to redact query strings and userinfo from server URLs before emitting them, so API keys embedded in URLs (e.g. ?exaApiKey=…) are not leaked to clients
Fixed /mcp test|resources|prompts (ACP) to wire the auth storage before prepareConfig so OAuth-backed MCP servers can refresh tokens and inject Authorization headers
Fixed /mcp list, /mcp test, /mcp resources, /mcp prompts, /mcp enable, and /mcp disable (ACP) to preserve project-over-user precedence when the same server name is defined in both scopes, matching the runtime capability merge so toggling the duplicated name flips the effective entry
Fixed /ssh add --port parsing to reject non-integer values (e.g. 22oops) instead of silently coercing them via Number.parseInt
Fixed /ssh list to deduplicate hosts shared between project and user scopes, listing project entries first to match capability-loader precedence
Fixed /export (ACP) to reject clipboard aliases (--copy, clipboard, copy) instead of using them as the output filename
Fixed ACP builtin commands (/compact, /force, /move, /browser) to surface underlying failures via output() instead of swallowing them
Fixed /session save|delete (ACP) to route through the active SessionManager so the persist writer is consulted and stale storage references are removed
Fixed /reload-plugins, /marketplace install|uninstall|upgrade, and /plugins enable|disable (ACP) to refresh slash command registries and emit available_commands_update after plugin state changes
Fixed ACP usage() text emission to be awaited so help and error output is not dropped or reordered when commands return immediately
Fixed ACP bash tool to release the client terminal handle on terminal/output or waitForExit failures, and to race output polling against abort so a stuck RPC cannot delay cancellation
Fixed ACP resource content blocks with image/* MIME types to be routed into the LLM images array instead of being dropped as opaque blobs
Fixed pr:// and issue:// URLs accepting empty, ., or .. path segments. pr://owner//77, pr://owner/repo/77/diff//2, and pr://owner/../77/diff previously slipped past the .filter(Boolean) split and were forwarded to gh; now they throw Invalid <scheme>:// URL: empty or unsafe path segment before any subprocess work.
Fixed read of issue:// / pr:// URLs ignoring the read tool's AbortSignal. Aborting a long pr://<N>/diff/all or stale issue fetch now propagates into the resolver and short-circuits at the handler entry; previously the gh round-trip and cache write ran to completion.
Fixed read <path>:raw (and the raw: true arg) still rendering markdown internal-URL content through the formatted markdown renderer. The TUI now respects the raw selector and falls back to the code-cell renderer so verbatim bytes are shown when requested.
Fixed eval tool JS cells crashing with a structuredClone error when an awaited final expression returned a non-cloneable value (module namespace, function, symbol, etc.). displayValue now falls back to a text representation and logs at debug instead of throwing.
Fixed eval tool JS rewriter missing the final expression when followed by trailing empty statements (e.g. await Promise.resolve(1);;). returnFinalExpression now scans backward past EmptyStatement nodes before deciding there is no final expression.
Fixed github view cache evicting valid rows on open whenever a longer-than-default github.cache.hardTtlSec was configured. openDb() no longer sweeps with the 7-day default before settings load; the per-lookup sweepIfDue() enforces the configured retention exclusively.
Fixed extension ctx.shutdown() being a no-op in the primary interactive path. The handler in initHooksAndCustomTools now sets shutdownRequested (mirroring the backgrounded-reinit path), and the main REPL loop drains the flag at the post-stream idle boundary so queued steering messages still flush before teardown.
Fixed plan-mode "Approve and compact context" dispatching queued user input against the stale plan-mode reference path. setPlanReferencePath(finalPlanFilePath) now runs before handleCompactCommand flushes the compaction queue, so any message typed during compaction is delivered with the approved plan context attached.
Fixed .omp/commands/fix-issues.md and .omp/commands/review-prs.md still instructing agents to call the removed github issue_view / pr_view / pr_diff ops; they now reference read issue://<N> and read pr://<N>[/diff[/all|<i>]].
Fixed ExtensionRunner.initialize() flushing buffered credential_disabled events before mode controllers had a chance to register their onError listener. Mode controllers call runner.initialize(...) immediately followed by runner.onError(...) synchronously; the flush now runs in a microtask after splicing the buffer, so a synchronously throwing credential_disabled handler is routed through the registered error listener instead of being silently dropped.

Security

Secured the GitHub cache store with strict file permissions (0600 files) and private permissions for newly created cache directories (0700) to reduce local cache exposure

[14.9.9] - 2026-05-12

Added

Added new task.isolation.mode values auto, apfs, btrfs, zfs, reflink, overlayfs, projfs, block-clone, and rcopy for native PAL-backed task isolation backends
Added automatic PAL-backed isolation backend selection so task.isolation.mode uses the host's best-available backend
Added input-token and output-token totals to omp stats --summary.

Changed

Changed task.isolation.enabled=true migration to map to task.isolation.mode = "auto" instead of legacy worktree isolation
Updated isolation configuration UI labels and descriptions to expose new back-end names (overlayfs, projfs, etc.) and removed references to deprecated values in guidance text

Fixed

Fixed worktree delta capture to include previously untracked file state by baselining untracked patches for both snapshots
Fixed task isolation startup to try alternate PAL backends when the preferred one is unavailable, allowing successful fallback instead of immediate failure
Mapped legacy task.isolation.mode values worktree, fuse-overlay, and fuse-projfs to their new equivalents during settings migration to preserve behavior with older configs

[14.9.8] - 2026-05-12

Breaking Changes

Changed the eval tool input format to a single-line *** Cell <lang>:"<title>" [t:<duration>] [rst] header per cell, replacing the *** Begin <LANG> / *** End <LANG> envelope and the standalone *** Title: / *** Timeout: / *** Reset directives. The lark grammar enforces a fixed attribute order; the runtime parser remains lenient (alias keys, bare positional tokens, single-quoted titles).

Added

Added :conflicts read selector (read <path>:conflicts) to return a one-line index of all unresolved merge conflicts with stable #N IDs for quick inspection
Added bulk conflict resolution with write({ path: "conflict://*", content }) to resolve all currently registered conflicts across files in one call, expanding @ours/@theirs/@base/@both per conflict and returning per-file counts
Added read support for conflict://<N> and read conflict://<N>/<scope> to inspect unresolved conflict regions captured by a prior read, including ours, theirs, and base side views with original file line alignment
Added shorthand content tokens @ours, @theirs, @both, and @base to conflict-resolution writes using path: "conflict://<N>" so replacement content can be composed from recorded conflict sections
Added conflict count metadata to read results so conflict files now show a warning badge (⚠ N) in the read tool UI
Added support for explicit boolean rst values (rst:true, rst:false, rst:1, rst:0, rst:yes, rst:no, rst:on, rst:off) in *** Cell headers
Added detection of unresolved git merge conflicts in read output: each marker block is registered with a session-stable id and surfaced in a footer with ours/theirs previews. Resolve a block by calling write({ path: "conflict://<id>", content }) — the tool splices the recorded marker region (markers and all sides) with the supplied content and routes through the normal writethrough (LSP format/diagnostics, fs-cache invalidation).

Changed

Changed read conflict warning footers to show X of Y unresolved conflicts when a range only captures part of a file and provide a read <path>:conflicts hint for the full list
Changed conflict scanning in conflict read paths to inspect the whole file (with a 10 MB cap) so totals better reflect hidden conflicts and truncated scans are called out
Changed conflict marker scanning during read to only register fully formed, column-0 merge-marker blocks, so indented or malformed marker-like lines are no longer treated as conflicts
Changed write conflict resolution to validate conflict:// IDs and report clear errors for malformed or unknown conflict URIs
Changed the HTML transcript renderer to parse the new *** Cell headers while keeping the older *** Begin <LANG> and ===== ... ===== formats renderable for historical sessions.
Changed the eval tool parser so a stray non-marker line between cells no longer crashes with null is not an object (evaluating 'BEGIN_RE.exec(lines[i])[1]'); stray content is consumed without aborting parsing.
Changed *** End to be an optional, undocumented per-cell terminator (kept in the lark to satisfy GPT-trained models' natural terminator habit during constrained sampling).

Fixed

Fixed single-conflict write retries to re-locate the recorded conflict block by exact marker content so shifted line numbers from out-of-band edits no longer prevent resolution
Fixed read conflict://* handling by rejecting wildcard reads with a clear write-only guidance error
Fixed conflict resolution to verify the live file still contains recorded <<<<<<< and >>>>>>> markers before splicing, preventing stale conflict IDs from silently corrupting out-of-band-edited files
Fixed @base token handling so two-way conflicts without a base section now return a clear error
Improved *** Cell header parsing to reject invalid rst values with a clear invalid rst value error

[14.9.7] - 2026-05-12

Breaking Changes

Changed the timeoutMs execution option to no longer be enforced during worker-based JS runs, so callers must rely on external cancellation signals for time limits
Replaced the Jupyter kernel gateway + WebSocket protocol behind the Python eval backend with a subprocess-backed runner that speaks NDJSON over stdin/stdout; removed the jupyter_kernel_gateway/ipykernel pip dependencies, the python.sharedGateway setting, the omp jupyter CLI command, and the PI_PYTHON_GATEWAY_URL / PI_PYTHON_GATEWAY_TOKEN environment variables

Added

Added Python tool.<name>(args) support to executePython sessions so evaluated Python code can invoke session tools through the prelude tool proxy
Added per-execution Python tool bridge session registration and loopback endpoint wiring so Python tool calls resolve to host tools and return tool results
Added status-event forwarding for Python tool bridge calls so tool invocations can emit execution status updates
Added browser-tab JavaScript execution through the shared runtime so tab runs now expose the standard helper globals (read, write, sort, uniq, counter, diff, tree, env, output, display, and tool)
Added static ESM import support to browser-tab JavaScript by rewriting top-level imports and resolving them against the tab session context
Added substring fallback matching to HistoryStorage.search so infix and short-token queries that FTS5 prefix matching misses are still returned
Added a live single-line sync progress display to the stats command showing current/total sessions while syncing
Added automatic inline JS evaluation fallback when worker creation failed so script execution still works in environments without worker support

Changed

Changed setup python to only verify a reachable Python 3 interpreter instead of installing Jupyter dependencies
Changed info output to remove the obsolete Python Gateway status block now that shared gateway management is no longer available
Changed JavaScript execution in executeJs to expose the worker\u2019s real process object instead of a restricted, frozen subset
Changed JavaScript evaluation to run per session in a worker-backed runner with explicit initialization and teardown handling
Changed the Python backend to launch one python -u runner.py subprocess per kernel; cancellation now sends SIGINT which raises a real KeyboardInterrupt in user code, and the same subprocess is reused across cells in session mode
Changed Python magic handling so %pip, %cd, %env, %pwd, %ls, %time, %timeit, %who, %reset, %load, %run, %%bash, %%capture, %%timeit, %%writefile, and !shell work without depending on IPython

Fixed

Fixed Python output rendering so text/markdown takes precedence over text/plain and status bundles are emitted as status updates rather than plain text
Fixed query tokenization in HistoryStorage.search so punctuation-delimited terms like git-commit are aligned with indexing and matched correctly
Fixed history search result merging to de-duplicate matches and return full-text matches before substring-only matches while still respecting the requested limit
Fixed JS run cancellation so aborting a run now also cancels in-flight tool calls and terminates the active worker session
Fixed top-level const, let, and class declarations in evaluated JavaScript to persist across subsequent runs by rewriting top-level declarations

[14.9.5] - 2026-05-12

Breaking Changes

Removed the jobs:// internal URL protocol; inspect background jobs via the job tool's list: true operation instead

Added

Added since and until date-range filters to search_issues, search_prs, search_commits, and search_repos, accepting relative durations (m/h/d/w/mo/y), ISO dates, and ISO datetimes
Added dateField support for date filtering (created or updated) so search results can be constrained by creation, update, pushed (for repos), or committer date (for commits)
Added owner-based scoping to async job registration and queries so background jobs can be registered with an ownerId and filtered per agent in getRunningJobs, getRecentJobs, getAllJobs, and cancelAll
Added agent ownership metadata to async jobs started by task and bash tools so their lifecycle and cancellation is attributed to the creating agent
Added list: true operation to the job tool, returning an immediate snapshot of every job spawned by the calling agent without waiting (replaces the deleted jobs:// URL)
Added per-agent visibility scoping to the job tool so list, poll, and cancel only see and act on jobs owned by the calling agent; cross-agent operations now return not_found

Changed

Changed search_issues, search_prs, search_commits, and search_repos to allow date-only queries where query is omitted if since/until is provided
Changed search_code to return a validation error when since/until is supplied because GitHub code search does not support date qualifiers
Changed async job manager ownership so subagents inherit the parent session’s global AsyncJobManager instead of creating and owning separate instances
Changed session lifecycle cleanup so the global async-job manager is disposed only by the owning top-level session
Changed subagent session switches and handoff paths to stop global async-job cancellation and cancel only jobs owned by that session
Changed agent:// and artifact:// URL resolution to search artifact outputs across all active sessions instead of only the current session, allowing parent and subagent sessions to read each other’s generated outputs by ID
Changed memory:// URL resolution to walk all active sessions’ memory roots and return the first matching file, so worktree-based subagents can access their own memory views as well as shared roots
Changed internal URL routing to use a shared process-global InternalUrlRouter and protocol handlers, so built-in tools resolve agent://, artifact://, memory://, skill://, rule://, mcp://, and local:// URLs without requiring session-specific router wiring
Changed mcp:// handler to use the globally registered MCP manager so MCP resource links work for agents sharing session context
Changed the ask.timeout default from 30 (seconds) to 0 (wait indefinitely). Auto-selecting the recommended option after a fixed delay was surprising users mid-deliberation; the timer is now strictly opt-in. The legacy auto-select behavior is preserved when ask.timeout is set to a non-zero value, and the ask tool's prompt has been updated so the model expects unlimited reply time by default.

Fixed

Added ModelRegistry.hasConfiguredAuth(model) to mirror the upstream @mariozechner/pi-coding-agent API surface; external plugins and downstream wrappers that pre-flight auth before launching a subagent no longer crash with this._modelRegistry.hasConfiguredAuth is not a function on the direct agent-launch path. (#993)
Fixed an ESM circular-import TDZ that crashed test suites when modules from the task/ and tools/ graphs were evaluated together (e.g. executor-warnings.test.ts + task-simple-mode.test.ts) by deferring BUILTIN_TOOLS.task's TaskTool.create dereference to factory-call time and sourcing truncateTail from session/streaming-output instead of the tools/ barrel
Treat keyless-by-design providers (llama.cpp, ollama, lm-studio) as authenticated in subagent model resolution; fixes silent fallback to parent remote model when a local model is configured. (#1008)
Fixed subagent disposal and session transitions that previously canceled all running async jobs, preventing inadvertent termination of a parent agent’s background work
Fixed multi-entry edits silently rendering a fake success when every entry failed (e.g. all hit the auto-generated guard), by surfacing isError: true from the single-path edit orchestrator so the renderer takes the error branch instead of falling through to the streaming-preview fallback that displays the proposed diff
Fixed the auto-generated streaming guard being gated behind edit.streamingAbort (default false), so it now pre-empts streaming edit tool calls targeting auto-generated files regardless of that setting
Fixed subagents launched in the same parallel batch not seeing each other in their initial # IRC Peers system-prompt block by pre-registering the agent in the global AgentRegistry before rebuildSystemPrompt runs and attaching the live session afterwards
Fixed plugin manifest extensions whose entry points at a directory (e.g. pi-goal's "pi": { "extensions": [".pi/extensions/pi-goal"] }) failing to load with Failed to load extension: Directories cannot be read like files. The plugin path resolver now resolves directory entries to their index.{ts,js,mjs,cjs} file, matching the behavior of native auto-discovery via resolveExtensionEntries.
Fixed the SSH tool on native Windows by avoiding OpenSSH ControlMaster multiplexing, which Win32-OpenSSH does not support and reports as getsockname failed (#154).
Fixed /export and /tree not showing developer-role messages (including the plan content injected after /plan approval) so the HTML export and TUI session tree now render developer messages dimmed with their actual content instead of hiding them entirely (#753)
Fixed Timed out initializing browser tab worker on prebuilt binaries by rewriting spawnTabWorker to import the worker entry with with { type: "file" } so Bun's --compile bundler statically discovers and embeds tab-worker-entry.ts in the single-file binary (#1011)

[14.9.3] - 2026-05-10

Breaking Changes

Changed the eval tool input format to canonical *** Begin <LANG> ... *** End <LANG> cells with *** Title, *** Timeout, and *** Reset directives, so legacy ===== ... ===== eval inputs are no longer accepted for execution
Removed the sectionSeparator re-export from config/prompt-templates, so existing imports from @oh-my-pi/pi-coding-agent/config/prompt-templates now need to resolve sectionSeparator from its utility package

Added

Added support for the *** Abort recovery marker in eval and hashline parsing to terminate processing safely when stream corruption is detected
Added support for wrapping hashline edits in *** Begin Patch and *** End Patch markers so patch input with these envelopes is parsed and applied
Added support in the HTML export renderer for the new *** Begin/*** End eval cell format
Added a dedicated [now] prompt block to buildSystemPrompt output containing current date, current working directory, and required end-of-turn continuation/verification guidance
Added a new [project] prompt block wrapper around workstation and workspace context and ensured it is emitted as a separate system prompt segment
Added dedicated HTML rendering for eval tool calls, including cell-by-cell parsing of ===== ... ===== blocks with inferred Python/JS/TypeScript highlighting
Added dedicated rendering support for search, recipe, and irc tool calls in transcript exports
Added a collapsible Available Tools section with a tool count and chip-style compact tool names
Added macOS power assertion settings power.preventIdleSleep, power.preventSystemSleep, power.declareUserActive, and power.preventDisplaySleep so users can control what types of sleep are blocked during sessions

Changed

Kept legacy ===== ... ===== eval transcripts renderable in HTML while adding parsing for the new *** Begin format for newer transcripts
Changed the system prompt’s Bash usage guidance to explicitly forbid specific anti-patterns (sed/awk line-range reads, stderr redirects, and | head|tail pagination) and require using dedicated tools for those operations
Changed delegated subagent prompts so shared task context is now rendered only in the system-level [context] block, while the user-facing task message now contains only the assignment prompt text
Changed system prompt rendering to use block markers such as [env], [contract], [role], [coop], and [closure] for more explicit structural instructions
Changed the working-directory value in rendered prompts to use shortenPath before interpolation
Updated subagent prompt assembly to compose prompt blocks and place the [now] block after the subagent-specific instructions
Changed GitHub (gh) tool cards to include operation, PR, branch, and truncated query/title/body details
Changed tool-call output to display internal _i intent separately and hide it from rendered argument JSON
Changed ast_edit and find/search rendering to show resolved path values and option flags such as limit, no-hidden, and no-reply
Changed power assertion behavior to take effect only while a prompt is in flight, replacing session-level persistent assertions

Removed

Removed the unused head and tail parameters from the bash tool schema, along with the dead normalizeBashCommand / applyHeadTail post-processing module — output truncation is already handled by the harness's streaming tail buffer and artifact spillover, so the agent should rely on read (or the artifact link) instead of inline truncation pipes.

Fixed

Fixed eval tool outputs to append a truncation warning and ask users to re-issue remaining work when parsing is aborted by *** Abort
Fixed hashline parsing and input splitting to stop at *** Abort and ignore trailing edits after the marker
Fixed subagent task prompt construction so a trailing [now] block in the base prompts is preserved and not swallowed when rendering subagent-system-prompt
Fixed edit rendering so provided input text is shown in the export even without a file path
Fixed args.paths handling in ast_edit and find so multiple paths are shown as a comma-separated list
Fixed power assertion state handling so subsequent prompts are no longer blocked after an aborted or canceled prompt
Fixed IRC background exchange poll loop leaking after session disposal: #scheduleBackgroundExchangeFlush now stops immediately when dispose() is called, preventing stale setTimeout callbacks from firing against a torn-down agent

[14.9.2] - 2026-05-10

Added

Added agentsMdFiles to WorkspaceTree so AGENTS.md discovery results are returned with the workspace scan output

Changed

Changed startup workspace discovery to use one native listWorkspace walk for both the rendered tree and AGENTS.md directory-context candidates, removing the layered git ls-files orchestration and secondary AGENTS.md glob.

Fixed

Fixed AGENTS.md context discovery to include AGENTS.md files that are explicitly gitignored while still excluding AGENTS.md files under ignored directories
Fixed task tool renderer spamming Tool renderer failed: undefined is not an object (evaluating 'args.tasks.length') warnings while a task call was streaming in (the tasks array is undefined until the partial JSON parser closes it); the renderer now tolerates an absent tasks field and shows 0 agents until the array arrives (#985).
Fixed MCP HTTP streamable transport spamming HTTP SSE stream error: ReadableStream already has a controller after every JSON-RPC request whose response was returned as text/event-stream. The transport used to break out of the SSE iterator once the matching response was captured and then re-open response.body for a background drain, but the body had already been piped through a TransformStream and could not be re-read. The drain now runs from a single iterator that resolves the response promise inline and continues to dispatch piggybacked notifications on the same stream.

[14.9.0] - 2026-05-10

Breaking Changes

Moved hashline APIs to the dedicated @oh-my-pi/pi-coding-agent/hashline module, moved hash helpers to @oh-my-pi/pi-coding-agent/hashline/hash, and removed the legacy edit/modes/hashline and edit/line-hash source subpaths.

Removed

Removed hashline auto-rebase. Anchor mismatches now reject immediately so the model re-reads instead of silently relocating an edit to a hash-collision within ±5 lines, which could otherwise apply the change to the wrong region. Stale-anchor recovery via the cached read snapshot is unaffected.

Fixed

Fixed compaction crashing with auth_unavailable when the current model's provider has no credentials configured; compaction now falls back to an available model role (or fails fast with a clear error) instead of attempting a doomed provider call (#986).
Fixed top-level static import rewriting in JS evaluation to use parser-based detection so only real import declarations are rewritten and import text inside strings, comments, or template literals is preserved
Fixed import ... with attribute handling in rewritten ESM imports so static imports with module attributes now become dynamic imports with matching with options
Fixed model resolution silently falling back to a different provider (e.g. Amazon Bedrock) when modelRoles specified a fully-qualified <provider>/<id> whose exact pair was not in the bundled catalog. Explicit provider prefixes are now honored or surface a clear error (#980).
Fixed session count inflation on Anthropic backend caused by a fresh random metadata.user_id being generated on every API request; all requests within one conversation now share a stable metadata.user_id derived from the session ID, matching the expected one-session-per-conversation counting
Fixed plan mode review resubmits to append each refreshed local://PLAN.md preview to the chat history, preserving the full refined plan in terminal scrollback.
Fixed compaction requests (manual and auto) not carrying metadata.user_id, leaving them unattributed on the backend
Fixed direct session-bound LLM calls (/btw ephemeral turns via runEphemeralTurn, branch summarization, session title generation) bypassing the agent and emitting a fresh random metadata.user_id per request on Anthropic OAuth: the session-level prepareSimpleStreamOptions helper now stamps the agent's session metadata onto direct calls, and generateBranchSummary plus generateSessionTitle accept and forward an explicit metadata option from the call site
Fixed metadata.user_id lacking the authenticated account_uuid on Anthropic OAuth requests; sessions now install a dynamic resolver via Agent#setMetadataResolver that builds { session_id, account_uuid? } per request, looking the live OAuth account UUID up from AuthStorage so it stays in sync with token refreshes and login/logout transitions instead of stranding a stale value
Fixed multi-file legacy Pi extensions failing to load when sibling .ts files import each other via relative paths (#983).
Fixed sub-agent dispatch silently routing to a model whose provider has no working credentials (e.g. an unqualified modelRoles.task id like qwen3.6-plus-free resolving to a provider the user is not authenticated against). Task dispatch now falls back to the parent session's active model — which by definition has working auth — when the resolved subagent model has none (#985).
Fixed legacy Pi plugin extensions failing to load on Windows when their entry path contains a drive letter (#990 by @jiwangyihao).

Added

Added a debug-panel raw SSE stream viewer so stuck model/tool-call streams can be inspected live from the TUI.
Added get_login_providers RPC command to list registered OAuth providers with their current authentication status (id, name, available, authenticated)
Added login RPC command to trigger OAuth login for a given provider; emits an open_url extension UI event (fire-and-forget) carrying the auth URL and optional instructions so headless clients can open the browser, then resolves when the callback-server flow completes
Added open_url variant to RpcExtensionUIRequest for the above
Added getLoginProviders() and login(providerId) methods to RpcClient

[14.8.0] - 2026-05-09

Added

Added hashline stale-anchor recovery by replaying edits against a session-scoped read/search snapshot and 3-way-merging them onto the current file when anchors no longer match

Fixed

Fixed legacy pi extensions failing to import their own bare-specifier dependencies (e.g. import x from "pkg"): files loaded via the omp-legacy-pi-file: namespace now pre-resolve bare imports against the extension's directory so the extension's own node_modules is honored.

Changed

Changed hashline success output to include a warning when stale-anchor recovery is used

[14.7.8] - 2026-05-08

Fixed

Fixed indefinite startup hang on large repos introduced in 14.7.6 (#975) on two fronts: (1) createAgentSession was awaiting buildAgentsMdSearch and buildWorkspaceTree directly in its blocking Promise.all, bypassing the existing 5s preparation deadline that previously protected startup — both scans are now raced against a 5s deadline and fall back to the system-prompt fallback path on timeout; (2) buildWorkspaceTree now derives its listing from git ls-files --cached --others --exclude-standard when the workspace is a git worktree, which is O(index size) and avoids the per-call full-tree gitignore-aware native scan that the previous implementation triggered. Repos without git, or where the call fails / times out, transparently fall back to the previous native-glob path.

[14.7.6] - 2026-05-07

Changed

Changed the "Hide Thinking Blocks" setting (Ctrl+T) to also instruct the provider to omit thinking/reasoning summaries from responses, instead of just hiding them client-side. Anthropic sees thinking.display = "omitted" (where supported); OpenAI Responses / Azure / Codex requests drop reasoning.summary entirely.

Fixed

Fixed the Hide Thinking Blocks toggle so changing it updates the active session’s request settings immediately, ensuring new responses reflect the current hide-thinking preference
Fixed system prompt preparation to keep successful context data and only fall back to minimal defaults for preparation steps that fail
Fixed system prompt preparation timeout to apply per-step instead of all-or-nothing: a single slow step (e.g. buildAgentsMdSearch on a huge directory tree, buildWorkspaceTree, loadProjectContextFiles) now falls back to its own minimal default while the other steps still populate, and the warning names which steps timed out.
Fixed subagents re-running expensive workspace scans (buildAgentsMdSearch, buildWorkspaceTree) on every spawn: parents now forward their already-resolved AGENTS.md search and workspace tree to subagents through createAgentSession, matching how contextFiles, skills, and promptTemplates are already inherited. On large monorepos this removes seconds of redundant work per task invocation and prevents the per-subagent system-prompt timeout warnings.

[14.7.5] - 2026-05-07

Added

Added optional /loop limits: /loop 10 stops after 10 auto-iterations, while duration forms such as /loop 10m and /loop 10min stop after the time limit.

Changed

Changed /loop to include the configured limit and remaining budget in the enabled status message

Fixed

Fixed /loop handling of malformed count or duration arguments by showing usage errors instead of enabling unbounded loop mode
Fixed inherited disabled macOS malloc stack logging variables leaking into shell sessions and spamming Bun subprocess output with MallocStackLogging warnings.

[14.7.4] - 2026-05-07

Breaking Changes

Removed the dedicated notebook tool; .ipynb reads and edits now go through read and edit.

Changed

Changed diff previews to syntax-highlight contiguous context lines in the unchanged sections when file language can be detected
Changed read tool behavior for .ipynb:raw requests to return raw notebook content instead of converting via markit
Changed .ipynb edit and read handling to route through notebook serialization helpers
Changed .ipynb reads to return an editable cell text representation and apply edits back to notebook JSON while preserving cell metadata and outputs where possible.

Removed

Removed the notebook.enabled configuration option from tool settings

Fixed

Fixed hashline edit streaming preview collapsing to a header-only "opaque box" when a second @PATH section header arrived mid-stream — earlier completed sections now stay rendered while the trailing section is still being typed.

[14.7.2] - 2026-05-06

Breaking Changes

Removed the exported BUILTIN_TOOL_METADATA API, including BuiltinEntry-style metadata exports and discoverable-built-in helper exports, which will break consumers relying on those symbols

Changed

Updated discoverable tool search (search_tool_bm25 and related discovery metadata) to read each tool’s own summary field when present, improving discoverability descriptions for built-in tools

Fixed

Fixed SearXNG web search Basic Auth validation to reject RFC 7617 control characters and clarified the equivalent config.yml and environment variable settings.
Fixed extension commands that return without starting a model turn leaving the interactive Working… spinner active indefinitely. (#927)
Fixed authHeader: true provider overrides without custom models so built-in model transport headers receive Authorization: Bearer <resolved-key> (#929).

[14.7.1] - 2026-05-06

Added

Added pr_create operation to the GitHub tool to create pull requests with title/body (or fill), base/head branch, draft, reviewer, assignee, and label options and return a summarized result including the new PR URL
Added read.summarize.prose setting to keep Markdown and plain-text reads out of the structural summarizer by default.

Changed

Changed the PI_GREP_WORKERS environment variable help text to state that it sets filesystem walker workers, defaults to 4, and uses 0 for automatic worker selection
Changed hashline replacement and pure-insert auto-absorb to also drop a single duplicated structural-closing line (}, );, ], etc.) on either boundary when keeping it would unbalance brackets. The pure-insert variant fires regardless of edit.hashlineAutoDropPureInsertDuplicates, while the existing 2+ line generic absorb stays gated on that setting.

[14.7.0] - 2026-05-04

Breaking Changes

Changed session system-prompt APIs to use ordered string block arrays by requiring buildSystemPrompt, CreateAgentSessionOptions.systemPrompt, Session.rebuildSystemPrompt, and extension before_agent_start/getSystemPrompt hooks to accept and return systemPrompt: string[] instead of a plain system-prompt string or separate projectPrompt field
Changed buildSystemPrompt and session rebuildSystemPrompt APIs to return { systemPrompt, projectPrompt }, requiring callers expecting a plain system prompt string to update to the new shape
Removed the top-level sel parameter from the read tool schema, requiring callers to migrate to path-embedded selectors (for example path:50-100, path:raw, or https://...:L1-L40)

Added

Added a separate projectPrompt artifact containing per-session project context (workstation, context files, AGENTS.md rules, workspace tree, and append prompt) so dynamic context is decoupled from the static system prompt
Added Project prompt token accounting to context-usage breakdowns and charts
Added tools.elideFileMutationInputs setting to optionally elide large write, edit, and apply_patch payloads in history after successful mutations
Added hashline-style return data for elided write calls so tools can include the resulting file content without leaking full input text
Added buildDirectoryTree and DirectoryTree exports to generate configurable directory trees with options for depth, entry limits, hidden-file handling, and truncation caps
Added buildWorkspaceTree and WorkspaceTree exports so callers can precompute and pass a workspace context to prompt generation
Added workspaceTree support to buildSystemPrompt options to reuse a prebuilt directory snapshot
Added read.summarize.enabled, read.summarize.minBodyLines, and read.summarize.minCommentLines settings to control whether read returns structural summaries and how many multiline body/comment lines are collapsed
Added edit.hashlineAutoDropPureInsertDuplicates setting to opt into dropping 2+ pure-insert hashline payload lines that duplicate adjacent file context; default is false.

Changed

Updated session dump and HTML export output to serialize ordered system-prompt blocks (including project context) and removed the dedicated project-prompt dump section
Renamed context-usage system-prompt accounting from a separate projectPrompt bucket to systemContext to match the new multi-block prompt structure
Changed prompt delivery to inject non-empty projectPrompt as a leading developer message before conversation messages instead of merging it into the base system prompt
Added projectPrompt to session dumps to expose the injected per-session project context separately
Changed write success output and preview rendering to display hashline-formatted written content from captured file text when mutation inputs are elided
Changed read directory rendering to return a two-level recency-sorted directory tree (including nested folders) instead of a flat alphabetical entry list, while still applying configurable truncation
Changed generated system prompts to include a working-directory tree block after directory context, showing recent files/directories (depth ≤ 3) and truncation notices when entries are elided
Changed read summary rendering to merge opening- and closing-brace boundaries around elided sections into a single .. line (including closers like }; or })), reducing those segments to one concise anchored summary line
Changed default read output for parseable code files without an explicit selector to return a structural summary instead of full verbatim lines, while still supporting full output for :raw and explicit ranges
Changed truncation/pagination hints in read, archive, and SQLite outputs to use colon syntax (Use :<offset>) when continuing reads
Changed the read tool UI preview title to include summary elision counts when a summary is returned
Changed hashline pure-insert duplicate auto-drop to be opt-in through edit.hashlineAutoDropPureInsertDuplicates instead of always enabled.

Fixed

Fixed selector parsing for colon-containing paths by only splitting :<sel> when the suffix matches a valid line-range or raw pattern, preventing paths like db.sqlite:users:42 from being misread as selectors

[14.6.6] - 2026-05-04

Added

Added Ctrl+D draft persistence: pressing Ctrl+D with text in the editor now exits the app and saves the unsent text as a per-session draft. Resuming the same session (e.g. via --resume) restores the draft into the editor (one-shot, removed after restore).

[14.6.4] - 2026-05-03

Added

Added hindsight.mentalModelsEnabled, hindsight.mentalModelAutoSeed, hindsight.mentalModelRefreshIntervalMs, and hindsight.mentalModelMaxRenderChars settings to control curated Hindsight mental-model activation, seeding, refresh cadence, and prompt render budget
Added <mental_models> injection to developer instructions, loading bank-level curated summaries as stable background context
Added built-in /memory mm commands (list, show, refresh, history, seed, reload, delete) to inspect and manage mental models on the active bank
Added scope-aware mental-model seeding for global, per-project, and per-project-tagged banks, including built-in seed models like user preferences, project conventions, and project decisions
Added warning output when hashline block replacements auto-absorbed duplicate boundary lines

Changed

Changed /memory clear and /memory enqueue to apply only to the current agent session’s Hindsight cache instead of all live Hindsight sessions
Changed the prompt assembly order so <mental_models> blocks are appended before <memories> recall blocks in developer instructions

Fixed

Fixed Hindsight memory prompt injection and recall/retain tool execution to resolve against the active session state, preventing context from an unrelated session from being used
Fixed subagent /task sessions to persist memories into the parent agent’s Hindsight bank by explicit parent state wiring
Fixed per-session memory retention behavior when switching or resuming sessions by rekeying Hindsight state and resetting conversation-tracking counters so first-turn recall and nth-turn retain cadence no longer leak across conversations
Fixed the first-turn startup race so <mental_models> appears in the opening system prompt when mental-model loading is enabled
Fixed retention hygiene by stripping <mental_models> blocks from retained content to prevent curated summaries from feeding back into future memory writes
Fixed <mental_models> rendering to honor the configured character budget and truncate with an explicit truncation marker when the snapshot exceeds limits
Fixed hashline replacements so duplicated payload boundary lines adjacent to a replaced block are absorbed into the replacement range instead of being duplicated

[14.6.3] - 2026-05-03

Breaking Changes

Renamed hashline separator configuration from PI_HASHLINE_SEP to PI_HL_SEP and changed the default payload separator from \\ to >

Added

Added inline hashline edit syntax so < ANCHOR${sep}TEXT prepended text to an anchored line and + ANCHOR${sep}TEXT appended text to it without requiring a multi-line payload block
Added a memory.backend setting (off, local, hindsight) under a new Memory settings tab to control which memory subsystem is active
Added Hindsight memory settings (hindsight.*) for API connection, bank identification, and recall/retain policy
Added retain, recall, and reflect tools for direct long-term memory search, retention, and reflection when using the Hindsight backend
Added hindsight.scoping setting (global, per-project, per-project-tagged) that controls whether memories are shared across projects, isolated per cwd, or tagged so global + project memories merge on recall (default: per-project-tagged)
Added search_code, search_commits, and search_repos ops to the github tool so the search surface mirrors gh search's subcommands

Changed

Changed hindsight tool retains to enqueue memory writes and return Memory queued. immediately instead of waiting on a network request
Changed hindsight tool memory writes to use automatic background batching (up to 16 items or 5 seconds) so tool calls do not block
Changed failed hindsight queue flushes to be surfaced as UI warning notices rather than failing the foreground hindsight tool call
Changed hashline read/search previews and diff output to keep | as the anchor-to-text separator while using the separate configured edit payload separator
Mapped invalid hindsight.scoping settings back to the default per-project-tagged behavior with a warning
Changed /memory view, /memory clear, and /memory enqueue to route through the selected memory backend instead of being hardcoded to local memories
Changed compaction context assembly to include backend-provided recall context when available
Updated multi-path search, find, ast-edit, and ast-grep calls to skip missing base paths, returning matches from remaining paths and reporting skipped paths in output
Replaced hindsight.dynamicBankId (boolean) with the explicit hindsight.scoping enum; legacy values are migrated automatically (dynamicBankId=true → scoping="per-project")
Changed search_repos to run as a global repository search using query qualifiers without applying the repo filter

Deprecated

Set explicit hindsight.scoping now takes precedence over legacy hindsight.dynamicBankId when migrating old settings

Removed

Removed legacy hindsight.dynamicBankId and hindsight.agentName fields from the active settings model
Removed hindsight.agentName and the HINDSIGHT_AGENT_NAME / HINDSIGHT_CHANNEL_ID / HINDSIGHT_USER_ID env vars; the legacy agent::project::channel::user bank tuple is gone. A user-set agentName is migrated onto hindsight.bankId.

Fixed

Fixed pending hindsight queued memory writes to flush on agent end, clear, and enqueue operations so tool-invoked facts are not dropped when sessions transition
Fixed retained memory writes from the hindsight tool to include session context and tags consistently in background batches
Fixed inline hashline modify operations to fail fast when combined with a delete or replace on the same target line
Fixed hashline parsing of payload blocks to handle a shared extra leading symbol prefix (such as markdown >>) on all payload lines by stripping it as an auto-correction instead of rejecting the edit
Forwarded project scoping tags to hindsight retain, recall, and reflect operations so manual memory commands honor the active tagging mode
Fixed legacy migrations by mapping existing memories.enabled values to memory.backend on load to preserve prior enable/disable behavior
Fixed memory retention so recalled <memories> blocks and legacy <hindsight_memories> / <relevant_memories> blocks are stripped before storing transcripts and do not feed back as new memory
Fixed search_code output to include each match path, repository, shortened SHA, and a one-line matching fragment
Fixed search_commits output to show shortened SHAs with commit message first lines
Fixed search_repos output formatting to return repository summaries including language, stars, forks, issues, visibility, and key status fields

[14.6.2] - 2026-05-03

Added

Added statusLine.sessionAccent to disable session-name accent coloring for the editor border and status line gap (#918)

Fixed

Disabled repeated OSC 11 background-color polling under WSL to avoid Windows terminal tab crashes while keeping initial and event-driven appearance detection (#914)
Fixed SSH ControlMaster socket paths to use OpenSSH's connection hash (%C) so connections to the same host with different users, ports, or jump hosts do not share a master session.

[14.6.1] - 2026-05-02

Changed

Updated GitHub call headers to display operation-specific titles and contextual metadata such as repository, branch, issue/PR IDs, and search query snippets for supported operations
Changed non-run-watch result rendering to honor terminal width, truncate long lines, and show a +N more lines expansion hint when output exceeds the preview limit

Fixed

Fixed GitHub tool output fallbacks that previously always showed a GitHub Run Watch heading so they now show the actual operation and clear no output/request failed status messaging

[14.6.0] - 2026-05-02

Breaking Changes

Reworked autoresearch storage and protocol. State now lives in ~/.omp/autoresearch/<project>.db (SQLite) and per-run logs in ~/.omp/autoresearch/<project>/runs/<id>/benchmark.log. The repo-side artifacts autoresearch.md, autoresearch.sh, autoresearch.checks.sh, autoresearch.program.md, autoresearch.ideas.md, autoresearch.jsonl, .autoresearch/, and autoresearch.config.json are no longer read or written; they are deleted by /autoresearch clear. Any existing data is not migrated.
Removed the autoresearch edit guard. write/edit/ast_edit are no longer blocked based on scope. Scope/off-limits are now post-hoc accountability fields on log_experiment.
Replaced rigid init_experiment contract validation with a simpler schema: name, goal, primary_metric, metric_unit, direction, secondary_metrics, scope_paths, off_limits, constraints, max_iterations, new_segment. Removed from_autoresearch_md, abandon_unlogged_runs, force, and preferred_command flags — the harness ./autoresearch.sh is the canonical workload, edit it and bump segment when you need to change it.
run_experiment no longer accepts a command parameter. The tool always runs bash autoresearch.sh. To change the workload, edit the harness and call init_experiment new_segment: true. Removed force, checks_timeout_seconds, and the legacy autoresearch.checks.sh auto-execution; run validation through the regular bash tool.
Replaced log_experiment ASI requirements and force/skip_restore flags with justification (post-hoc explanation for scope deviations) and flag_runs (mark earlier runs suspect to exclude them from baseline math). ASI is now opaque metadata.
/autoresearch clear now resets the worktree to the session's recorded baseline commit (when on an autoresearch/* branch or with --reset-tree), closes the active session, and deletes any leftover legacy autoresearch repo artifacts.
/autoresearch now refuses on a dirty worktree with an explicit error instead of silently continuing on the current branch. Commit or stash before invoking — the session needs a clean baseline on a dedicated autoresearch/* branch.
Split /autoresearch into a two-phase protocol. Phase 1 (no session) prompts the agent to build the benchmark harness as ./autoresearch.sh (must exit 0 and print METRIC <name>=<value>). Calling init_experiment ends Phase 1: it requires ./autoresearch.sh to exist, auto-commits any pending harness changes on an autoresearch/* branch, then records that commit as the baseline. Phase 2 is the existing iteration loop.
Autoresearch sessions are now scoped to the git branch they were created on. Switching off the autoresearch/* branch hides the dashboard widget, detaches the experiment tools, and skips the autoresearch system prompt; switching back resumes seamlessly. /autoresearch on a fresh branch starts a fresh session instead of resurrecting a session bound to a different branch.
log_experiment discard no longer rewinds prior keep commits. On an autoresearch/* branch it now resets the worktree to HEAD (and git cleans untracked) instead of git reset --hard $baseline_commit. Discard reverts only the current iteration's uncommitted edits; previously kept improvements stay on the branch. /autoresearch clear continues to reset to the recorded baseline commit when explicitly requested.
Autoresearch SQLite storage is now created lazily on first init_experiment. Running omp in a project that never invokes /autoresearch no longer creates a per-folder DB.
Changed search, find, ast_grep, and ast_edit to accept paths: string[] instead of comma- or whitespace-delimited path strings.

Added

Added update_notes tool with body (replace) and append_idea (append a bullet under an ## Ideas section). Notes are injected into the system prompt every iteration and replace the file-based autoresearch.md / .program.md / .ideas.md ecosystem.

Changed

Updated log_experiment summary output to include the count of scope deviations detected for a run
Used the active session context in autoresearch resume instructions instead of referencing deleted repo-side files
Removed PI_STRICT_EDIT_MODE; model-specific edit mode fallbacks are no longer disableable by environment flag.

Fixed

Atom edit auto-rebase warning now dedupes by (originalLid, rebasedLine) pair. Previously, @Lid followed by N +TEXT lines emitted N identical "Auto-rebased anchor" warnings (one per cloned cursor anchor); now emits exactly one per distinct rebase.
Atom/hashline diff preview no longer renders deleted lines with a 2-space hash placeholder (-20 |old) that visually mimicked a Lid. Removed lines now use -- as the placeholder (-20--|old), making them unambiguously non-Lid.
Atom/hashline diff preview no longer folds size-mismatched -/+ runs into a confusing mix of * (paired modification) lines plus surplus -/+ lines. The * collapse now applies only to clean 1:1 line replacements (same number of dels and adds); range replaces with N→M (N≠M) render as plain unified-diff - then + runs.
Atom edits now warn when @Lid lands on a brace-opening line and the inserted content is at sibling indent (≤ anchor indent) — a foot-gun where the agent meant ^<nextSibling> but the inserts ended up as the first body element of the {...} block.
Atom auto-fix warning for adjacent-duplicate cleanup is now formatted as AUTO-FIX applied — verify the result. Removed ... instead of the easier-to-miss Auto-fixed: removed ..., and explains that {}/()/[] balance was the trigger.
Fixed multi-target search, ast-grep, and ast-edit path handling by running each resolved target separately under root-level path resolution
Fixed pagination and match/replacement summaries for multi-target AST and text searches so totals and affected file counts include all targets
Fixed returned file paths for multi-target search and ast-grep results by normalizing them to the original search scope
Fixed log_experiment keep silently dropping the iteration's diff on an autoresearch branch. The previous logic filtered out every path that was already dirty when run_experiment ran — but in the iteration cycle the agent's edits always land before run_experiment, so the entire iteration was filtered away and nothing was committed. On an autoresearch branch, keep now treats every currently-dirty path as the iteration's change and commits it.

[14.5.14] - 2026-05-01

Changed

Changed markdown conversion and archive tooling to defer loading heavy dependencies (Turndown, fflate, and browser agent content) until first use, reducing startup overhead for CLI startup and command initialization

Fixed

Fixed changelog state tracking by flushing lastChangelogVersion to settings immediately when showing new entries, so the updated version is persisted across restarts

[14.5.13] - 2026-05-01

Breaking Changes

Removed the built-in python tool in favor of eval, so tool allowlists and tool-call handlers referencing python need to migrate
Removed the python.toolMode setting and replaced mode control with separate eval.py and eval.js toggles
Changed the tool runtime config surface by migrating python execution timeout/export behavior to eval and replacing ./ipy/* internal exports with ./eval/* paths
Changed the eval tool wire format to a single input string composed of markdown fenced code blocks (with per-fence language, timeout, title, and reset metadata in the info string) instead of top-level cells, language, timeout, and reset fields

Added

Added a JavaScript backend to the eval tool with an in-process VM runtime and JS helper bridge (read, write, glob, etc.)
Added eval.py and eval.js settings so Python and JavaScript eval backends can be enabled or disabled independently
Added rename_file action to the Lsp tool to rename files and directories with LSP workspace/willRenameFiles and workspace/didRenameFiles flow, applying returned workspace edits before moving files
Added apply: false preview mode for rename_file so users can see planned LSP edits without performing filesystem changes
Added request action to invoke arbitrary LSP methods, with automatic textDocument/position parameter construction from file/line/symbol and support for explicit JSON payload
Added capabilities action to display language server capabilities (for a file or all configured servers) through the LSP tool

Changed

Changed AGENTS.md discovery to respect .gitignore files during project context collection so ignored context files are no longer loaded
Changed eval tool initialization to skip Python kernel preflight when the JavaScript backend is enabled, avoiding unnecessary startup checks
Changed model registry refresh flow to defer rebuilding the canonical model index until refresh operations complete, reducing refresh churn
Changed execution/tool discovery flow so exec maps to eval when any eval backend is enabled, while bash stays independently available
Changed eval dispatch to automatically fall back to JavaScript when Python is unavailable and JavaScript backend is enabled
Parallelized plugin root preloading with other startup initialization in runRootCommand to reduce startup latency
Parallelized session bootstrap work in createAgentSession, including AGENTS.md scanning, context discovery, prompt template loading, slash command loading, and skill discovery, to reduce time to first available session

Fixed

Fixed eval startup messaging to report eval as unavailable when Python is unreachable and JavaScript backend is disabled

[14.5.12] - 2026-04-30

Breaking Changes

Removed the legacy browser action verbs (goto, observe, click, type, fill, press, scroll, drag, wait_for_selector, extract_readable, and screenshot) in favor of invoking those workflows through run

Added

Added a browser tool open/run/close flow with a run action that executes async JavaScript and provides page, browser, tab, display, assert, and wait in scope
Added named tabs on open with default name main so browser state can be reused across run calls and subagents
Added support for app.path and app.cdp_url on open to launch/connect to CDP-capable desktop apps

Changed

Changed browser tool output rendering to display run calls as JavaScript code cells with status and output previews while showing open/close as compact status lines
Changed open to open or reuse named tabs and close to support all: true and kill-based process termination behavior
Changed app attachment behavior to reuse an existing CDP endpoint when available and avoid unnecessary respawn of matching app processes
Changed tab closing so closing a tab no longer implicitly affects unnamed sessions when multiple tabs are used
Changed browser export rendering to label outputs under the browser tool and include app metadata badges

Fixed

Fixed Electron/CDP attachment target selection to skip helper windows and pick the most likely user-visible page target
Fixed connection startup by waiting for the CDP endpoint and surfacing a timeout error when it does not become available
Fixed plan mode to auto-redirect write and edit calls targeting a bare PLAN.md (or any same-basename cwd-relative path) to the canonical local://PLAN.md plan artifact instead of rejecting them

[14.5.11] - 2026-04-30

Breaking Changes

todo_write: renamed replace op to init and reshaped its input to list: [{phase: string, items: string[]}]. Tasks no longer accept a status field; all start pending and the first auto-promotes to in_progress. The append op's items is now string[] (was {id, label}[])
todo_write: removed the synthetic task-N / phase-N ids — task identity is now its content and phase identity is its name. The task field on start/done/drop/note and the phase field on done/drop/rm/append take those values directly
todo_write: phase names no longer accept a numeric/roman prefix (I., 1., Phase 1:, …). The renderer numbers phases visually (Ⅰ. Ⅱ. Ⅲ. …) and the model-facing state stores the bare noun phrase

Changed

Changed /todo task and phase operations to target items by fuzzy content or phase name matching instead of numeric IDs
Changed initial todo markdown export template heading from # I. Todos to # Todos

Fixed

Fixed todo auto-clear scheduling to identify completed tasks by phase and content so only the matching task is cleared after delays

[14.5.10] - 2026-04-30

Breaking Changes

Removed the worktree parameter from github pr_checkout. Worktrees are now always written to ~/.omp/wt/<encoded-primary-repo>/pr-<number>/, derived from the primary repository path
Stopped reading the branch parameter for github pr_checkout. The local branch is now always pr-<number>; the branch schema field is still accepted by pr_push, repo_view, and run_watch

Added

Added checkouts summary entries to pr_checkout results, including each checkout's branch, worktree path, remote, and reuse status
Added combined summaries for pr_view and pr_diff when pr is an array, so multi-request responses now include all requested pull requests in one return
Added array support to the pr parameter on github pr_view, pr_diff, and pr_checkout so a single call can fetch, diff, or check out multiple pull requests in one batch
Added a per-repo serialization lock (withRepoLock) so concurrent pr_checkout calls against the same repository no longer race on git's internal .git/config.lock, commit-graph, and worktree lock files

Changed

Changed the diff preview shown after edits so changed lines are never collapsed: removed runs and the global preview budget no longer truncate, only unchanged context still collapses
Changed adjacent -/+ pairs in edit previews to fold into a single *<line><hash>|<new-content> modification line so 1:1 line replacements stay compact
Changed git.remote.add to be idempotent when the remote already exists with the same URL (instead of failing with remote ... already exists), and to surface a clear error when the existing URL differs
Changed pr_checkout to run gh pr view calls in parallel for batch invocations while serializing the in-repo git mutations to keep the operation race-free
Changed pr_checkout to auto-derive the worktree location and local branch name (see Breaking Changes), removing the per-call overrides that previously let callers pin a worktree path or local branch

Removed

Removed the ./hooks and ./hooks/* package export entries
Removed the Suspicious duplicate warning emitted after edits — it produced too many false positives (e.g. legitimate adjacent \t});\n\t});); the auto-fix path that uses bracket balance to safely de-duplicate is unchanged

Fixed

Fixed bash interceptor rules to also check the original command before cd normalization, so leading cd ... && wrappers no longer bypass interception
Fixed LSP client shutdown to properly await the language server's exit instead of fire-and-forget, preventing premature process termination on SIGINT and SIGTERM
Fixed concurrent bash commands being tracked independently so aborting one no longer silently drops tracking of others

[14.5.9] - 2026-04-30

Added

Added the /context slash command to display an estimated context-usage breakdown panel for the current session
Added -LidA..LidB syntax to delete inclusive line ranges in a single atom operation
Added LidA..LidB=TEXT range-replace syntax with \TEXT and \ continuation lines for multi-line replacement blocks
Added shorthand cursor+insert operations in atom edits, including ^Lid (insert before anchor), ^+TEXT, $+TEXT, and Lid+TEXT/@Lid+TEXT
Added standalone file-op fallback so !rm and !mv DEST inputs can be normalized into sections when using split input parsing

Changed

Changed token counting to use tokenizer-based estimates instead of a character-per-4 heuristic for context and compaction calculations
Changed hashline anchor auto-rebase tolerance from ±2 lines to ±5 lines for stale Lid recovery
Changed atom input handling so #-prefixed lines are treated as comments and ignored
Changed execution when all edits are no-op Lid=TEXT replacements to return success with a no-change explanation instead of throwing

Fixed

Fixed malformed range and unified-diff-like atom syntax by rejecting reversed ranges, mismatched range endpoint hashes, and forms like +Lid|TEXT, +Lid=TEXT, and -LidA..LidB|TEXT with explicit actionable errors
Fixed hash mismatch errors to include likely-shifted anchor hints when a unique matching line is found elsewhere in the file

[14.5.8] - 2026-04-29

Breaking Changes

Changed the task runner toggle from just.enabled to runCommand.enabled, so existing configurations using just.enabled must be migrated
Removed the legacy just tool and replaced it with run_command
Renamed the built-in tool API from just to run_command, so clients requesting/handling the old tool name must update

Added

Added a new run_command tool that runs project tasks via a single op argument, auto-detecting and supporting recipes from justfiles, package.json scripts (including workspace packages), Cargo bin/example/test targets, Makefiles, and Taskfiles
Added support for explicit runner-qualified tasks via run_command with runnerId:task syntax in the prompt guidance

Changed

Changed automatic tool availability so requesting bash can now auto-include run_command when a supported task runner manifest is detected in the working directory
Changed task resolution to disambiguate identical task names across multiple runners and show runner-aware command execution errors

Fixed

Fixed editor draft being erased when a user message queued during streaming was eventually submitted; the queue/steer path now preserves any new prompt the user has typed since queuing, matching the existing optimistic-send protection.

[14.5.7] - 2026-04-29

Fixed

Fixed hook editors to recognize Ctrl+Enter when terminals include NumLock or keypad Enter metadata.

[14.5.6] - 2026-04-29

Changed

Removed the atom edit mode's multi-anchor auto-rebase rejection so stale-but-uniquely-rebasable block edits apply with warnings instead of failing.

[14.5.5] - 2026-04-29

Breaking Changes

Rejected atom diffs with unrecognized operations (including lone '-' lines) by throwing parse errors instead of treating them as inserts

Added

Added duplicate-line post-edit detection that warns on newly introduced adjacent identical lines and auto-removes one duplicate when bracket-balance is restored
Added a warning when suspicious adjacent duplicates are introduced after edits so users can review potential stale-line issues

Changed

Changed anchor rebase handling to fail when multiple mutating anchors would need auto-rebase, preventing silent misapplied contiguous block rewrites

Fixed

Fixed bracket-corruption caused by botched block rewrites by automatically removing a newly introduced duplicate adjacent line when removing it restores the original {}, (), and [] balance and by warning when automatic removal is unsafe

[14.5.4] - 2026-04-28

Breaking Changes

Changed the atom edit mode from JSON { path, edits } calls to the compact file-oriented input patch language that was previously exposed as atomd; atomd is no longer a separate edit variant
Renamed MCP tool identifiers from the mcp_<server>_<tool> format to mcp__<server>_<tool> so custom tool names, active tool lists, and persisted MCP selections must be updated to the new prefix
Renamed the built-in content-search tool from grep to search, including SDK/tool event names and settings keys (search.enabled, search.contextBefore, search.contextAfter), so integrations using grep and grep.* references must be updated

Added

Added internal URL support to the search tool, allowing artifact://-style paths that resolve to local files to be searched directly
Added IRC relay observation in the main agent UI so every IRC exchange between agents is rendered in the main transcript, even when the main agent is not a direct participant
Added stateful href/hrefr prompt helpers that can reuse anchors remembered from prior hline helper calls

Changed

Changed file-path rendering across search, find, AST, LSP, and related edit outputs to display targets as cwd-relative paths when they resolve inside the working directory and keep absolute paths for files outside the cwd
Changed system prompt guidance so in-cwd tool paths must be passed as cwd-relative paths and absolute paths only for out-of-cwd targets or ~ expansion
Updated edit streaming diff previews for patch, replace, and hashline to produce a single request-level preview for the new single-file path mode
Bumped default read.defaultLimit from 300 to 500 lines, and scaled the read tool's byte budget with the line limit (max(50KB, lines * 512)) so the configured line count is no longer truncated by the shared 50KB cap

Fixed

Fixed atom edit streaming previews to use atom headers for file names instead of apply_patch parsing errors.
Fixed collapsed search result rendering so summary and truncation rows stay within the collapsed output budget
Updated search path handling to support path lists and internal file paths while preserving previous search behavior

[14.5.3] - 2026-04-27

Added

Added bracketed loc forms (anchor), [anchor], [anchor, (anchor, anchor], and anchor) to atom splice editing so a single anchor can target a block body, whole node, or partial node region
Added automatic block-delimiter inference for block splices using file extension, defaulting to { and using ( for Lisp-family files
Added optional pre/post arguments to the href prompt helper so hashline references can be wrapped as bracketed or parenthesized anchors
Added destination-aware indent handling for block replacements by detecting file indent style and reapplying tabs/spaces to spliced body text

Changed

Changed bracketed atom locators to be splice-only and reject pre, post, or sed on region locators
Changed applyAtomEdits to forbid mixing splice_block with other anchor-scoped edit verbs in one call
Changed splice_block resolution behavior to include selected block range and enclosing-count context in warning output
Changed balanced-block parsing to support kind selection ({, (, [), nesting depth, and safer same-line enclosing selection

Removed

Removed the sed F option for literal matching; sed now accepts only pat, rep, and optional g, with F-style literal matching no longer supported

Fixed

Fixed splice_block multi-line replacements to replace the exact target region and avoid duplicate braces or duplicated signature lines from bare-anchor splice attempts
Fixed false-positive “unbalanced” replacement-body warnings caused by braces in regex/string/comment text by skipping those constructs during block scanning
Fixed splice_block for same-line ( bodies so inline call sites like int(port) can be replaced correctly

[14.5.2] - 2026-04-26

Breaking Changes

Removed support for sed-style string expressions and required sed to be specified as an object with pat and rep (and optional g, F, i flags)

Changed

Changed atom sed replacements to be global by default and require g:false for first-match-only replacements
Changed anchor validation so multiple sed operations can target the same line and run sequentially
Changed cross-entry conflict resolution so del edits on an anchor are ignored when that line is also replaced by sed or splice in another edit entry

Fixed

Fixed zero-length regex sed patterns (for example (), ^, $) to fall back to literal substring matching instead of producing insertion-like replacements
Fixed sed chaining so each edit on the same anchor applies to the latest line state from prior replacements

[14.5.1] - 2026-04-26

Removed

Removed \t escaped-tab indentation autocorrect from hashline and atom edit modes (and the PI_HASHLINE_AUTOCORRECT_ESCAPED_TABS environment toggle); literal \t in edit content is now preserved verbatim
Removed the suspicious-\uDDDD warning preflight from hashline edits
Removed the hand-rolled JSON unescape fallback in the streaming edit-arg renderer; partial fragments that fail JSON.parse are now surfaced raw rather than partially decoded with a non-spec-compliant unescaper that mishandled lone surrogates

[14.4.3] - 2026-04-26

Added

Added irc tool for agent-to-agent messaging with list and send operations, including optional broadcast to all and optional suppression of reply waits
Added irc.enabled tool setting (default true) to toggle agent-to-agent messaging
Added live in-chat rendering of IRC incoming and auto-reply messages so peer messages now appear in the session transcript

Changed

Changed peer-aware prompts for subagents to include currently live agent peers and IRC usage guidance when available
Changed the /btw helper to use a session-side ephemeral turn path that preserves streaming-context handling and updates the existing request handling behavior
Merged the poll and cancel_job tools into a single job tool that accepts poll and cancel arrays; the renamed tool reuses the richer poll renderer for both polling and cancellation calls

Fixed

Fixed IRC messaging to use a background ephemeral turn path so a recipient can reply even while its main loop is busy
Fixed /btw handling of empty prompts and missing model configuration by rejecting invalid requests before starting a stream
Fixed /btw request replacement so issuing a new query cleanly aborts the previous active request

[14.4.2] - 2026-04-26

Breaking Changes

Changed /todo append from JSON payload input to /todo append [<phase>] <task...> with optional quoted tokens and automatic phase creation

Added

Added note to todo-write operations so you can append follow-up text notes to a task via op: "note" and text
Added markdown note-block support to /todo export and /todo import so task notes are written as blockquote lines and reloaded with the todo list
Added /todo export <path> to write the current todo list as Markdown to a file, defaulting to TODO.md when no path is provided
Added /todo import <path> to replace the current todo list from a Markdown file, defaulting to TODO.md when no path is provided
Added live poll progress updates so the UI now emits intermediate job state while waiting for jobs to finish
Added a dedicated TUI renderer for the poll tool that displays job status, counts, duration, and result/error previews
Added a /todo slash command to view and modify todos with edit, copy, start, done, drop, rm, append, and replace operations
Added /todo edit to open the current todo list in $EDITOR as Markdown and sync the edited checklist back into the session
Added /todo copy to copy the rendered Markdown todo list to the clipboard

Changed

Changed todo list rendering and summaries to show a +N note marker on task lines and display attached notes for tasks in progress
Changed /todo start, /todo done, /todo drop, and /todo rm to resolve task/phase targets by fuzzy id/name matching instead of strict identifiers
Removed /todo replace from supported slash commands
Changed todo list restoration to include user todo-edit custom session entries so slash-command and editor-based todo updates persist after reload
Restored sensible defaults for grep.contextBefore (1) and grep.contextAfter (3) so grep matches show context lines by default after the pre/post parameters were folded into settings

Removed

Removed the chunk edit mode, chunk-aware read selectors, chunk-aware grep rendering, and the omp read chunk CLI subcommand
Removed the read.prosechunks, read.explorechunks, and read.anchorstyle settings
Removed the underlying chunk native module and AST-based chunk schema generation from pi-natives

Fixed

Fixed poll final output to reflect live job data from the async job manager, improving status and result visibility
Fixed job duration and output reporting to use current job snapshots instead of initial poll input metadata
Fixed poll wait duration parsing to fall back to 30s when the provided value is an empty string

[14.4.1] - 2026-04-26

Breaking Changes

Replaced the legacy gh_repo_view, gh_issue_view, gh_pr_view, gh_pr_diff, gh_pr_checkout, gh_pr_push, gh_run_watch, gh_search_issues, and gh_search_prs tool names with only github, which requires updating existing callers that invoked the old gh_* tools

Added

Added a sed verb to the atom edit tool for line-local substitutions using sed-style syntax (s/pattern/replacement/) with g, i, and F flags and model-tolerant delimiter choices
Added the unified github tool with op-based dispatch for repository, issue, pull request, search, checkout, push, and Actions watch workflows
Added op routing so callers can select repo_view, issue_view, pr_view, pr_diff, pr_checkout, pr_push, search_issues, search_prs, or run_watch through a single tool entry point

Changed

Changed hashline-based read and match output formatting to use LINE+ID|content as the anchor/content separator, and updated match/context markers to > for matches and : for context
Updated GitHub CLI render output to show GitHub <op> for tool calls dispatched through github operations

Removed

Removed the built-in taplo Language Server entry from default LSP settings, so TOML files no longer have default TOML server startup

Fixed

Fixed atom loc parsing so path-qualified anchors like path:263ti| ... and single-anchor locs containing hyphens no longer mis-parse as ranges
Fixed hashline anchor handling in atom edits so a provided content hint after the anchor (| or : suffix) can rebond a stale hash to the intended line
Fixed atom sed execution to tolerate common model-emitted forms such as /pat/rep/, and to apply safe literal fallbacks for regex failures or metacharacter-heavy patterns while still erroring when no match is possible

[14.4.0] - 2026-04-26

Breaking Changes

Removed multi-pattern array input from ast_grep by changing pat to a single pattern string, so call sites using pat: [...] must be updated to send one query per invocation
Removed lang, glob, and sel options from ast_edit and ast_grep, and moved those behaviors into the required path argument
Required path for ast_edit and ast_grep, so invocations that relied on implicit repo-root searching are no longer valid
Changed todo_write from multi-field verb payloads to an ordered array of flat operations, while retaining replace for harness bootstrap compatibility
Renamed atom edit operations from before and after to pre and post, so existing atom payloads using the old operation keys must be updated
Changed the hashline anchor format from LINE#ID:content to LINEID:content (no # separator, colon between anchor and content, no padding on line numbers); expanded the bigram alphabet from 40 hand-picked English bigrams to the full 647 single-token 2-letter bigrams — invalidates every previously captured LINE#ID reference
Renamed the subagent completion contract from submit_result to yield, so subagent sessions must now finish with the yield tool and the requireYieldTool option; submit_result/requireSubmitResultTool and old completion calls are no longer recognized
Changed the hashline and chunk anchor ID format from the prior hex-like tokens to two-letter BPE bigrams (for example #th), which invalidates previously captured LINE#ID/chunk selectors and requires re-reading to refresh anchors

Added

Added inline file overrides in atom locators (loc: "a.ts:160sr") so cross-file edits can be written without a separate per-entry path field
Added openai to the providers.image options, allowing image generation to be explicitly routed through the active GPT Responses/Codex model
Added between atom edit operation to replace only the lines between two surviving anchors while preserving the boundary anchors
Added conflict detection for between atom edits to require non-overlapping regions and forbid edits targeting lines strictly inside those regions
Added atom edit mode to edit with single-anchor operations (set, before, after, del, sub, ins) for hashline-anchored line edits
Added support for request-level path defaults in patch, replace, and chunk edits so shared file paths no longer need to be repeated in every entry

Changed

Updated atom and hashline edit anchor validation to auto-rebase a stale anchor within ±2 lines when the same hash matches a unique nearby line, continuing the edit with a warning instead of immediate failure
Changed bash command output labels from [full result: artifact://…] to [raw output: artifact://…] for artifact references produced from large command output
Changed todo_write done, rm, and drop operations to target all tasks when neither task nor phase is provided, and made append create the target phase automatically when missing
Updated ast_edit and ast_grep to pass file-selection intent through path (including inline globs and comma/space-separated path lists) instead of separate glob filters
Changed ast_grep pagination API from offset to skip
Flattened todo_write operation arguments to { op, task?, phase?, items? }[] and removed task details from the persisted todo shape
Changed grep truncation output to report Result limit reached; narrow path. and label match/result caps as first N
Changed JSON tree output to truncate inline argument pairs by available width and add an ellipsis when values no longer fit in the display
Changed JSON tree rendering to hide harness-internal intent and __partialJson fields from top-level tool output
Simplified the grep tool schema by requiring path, folding glob and type filtering into path globs, auto-detecting multiline patterns, removing model-controlled context and limit options, and renaming result skipping to skip.
Changed atom edit request format to use a shared loc selector, including range ("160sr-9ab") and boundary ("^", "$") forms instead of per-operation anchor fields
Changed atom edit payload fields so set, pre, and post now require line-array values and sub now takes a [find, replace] tuple, with boundary deletion now expressed as set: []
Changed edit diff wrapping to preserve the active line-prefix separator (| or │) while keeping continuation lines aligned by line-number width
Changed Vim focus and viewport rendering to align cursor/selection markers and line numbers in a single gutter format
Changed auto image provider selection for providers.image=auto to try active GPT image generation before Antigravity, OpenRouter, and Gemini
Updated atom and hashline anchor validation to require the full line+suffix anchor format and report missing-line-number errors more clearly, including guidance when only a 2-letter suffix is provided
Changed read, grep, and ast-edit line-prefixed output to drop fixed-width line number padding, so anchors render in natural width without leading spaces
Updated terminal diff rendering to use a continuous │ gutter and hide repeated line numbers on adjacent diff lines
Updated subagent reminders, prompts, and rendered subagent output to reference yield completion and report missing/final results from yield tool data
Updated the edit workflow to treat atom mode like hashline mode for read output, so hashline anchors are shown when atom is selected
Adjusted patch/replace/chunk tooling to accept optional entry paths and to apply a top-level path default
Updated hashline/chunk selector parsing to the new stable bigram token set used for checksums
Renamed the image generation implementation module to image-gen and routed active GPT Responses/Codex models through OpenAI's hosted image_generation tool with WebP output

Removed

Removed line-range support from atom mode selectors, including loc values like 160sr-170ab, so edits must target a single anchor (160sr, ^, or $) per entry
Removed the atom del verb and now require anchored-line deletion to be requested with set: []
Removed todo_write task details and the add_notes operation

Fixed

Improved no-op edit diagnostics for atom and hashline operations so edits that leave content unchanged now fail with contextual details (edit index, locator, and reason), including guidance for replace_range no-op cases
Wrapped todo_write operations in an ops object so Codex/OpenAI function schemas always use a JSON Schema object.
Fixed JSON tree rendering for tool arguments by excluding injected internal keys from displayed root records
Printed assistant errorMessage text in print mode output to stderr so message-level errors are visible during non-interactive runs
Displayed assistant errorMessage text in the assistant message component for completed tool responses with non-terminal stop reasons
Fixed atom input handling to ignore null optional verb fields so entries with pre, set, post, or sub set to null remain valid
Fixed status-line Git branch rendering to degrade gracefully when the process hits ENFILE/EMFILE while reading optional Git refs
Changed hashline mismatch failure output to show a clean numbered context block with numbered gutter and full-anchor alignment guidance when edits are rejected after the file changed
Fixed atom mode to apply multiple edits on the same anchor line without index-shift artifacts, so mixed operations like before, after, set, sub, ins, and del now resolve consistently
Fixed atom mode append_file insertion to preserve a file’s trailing newline sentinel when appending content
Fixed read output for raw archive entries so hashline anchors, line numbers, and chunked formatting are not injected into raw content
Fixed hashline parsing so lines like # Note: or # TODO: are no longer misinterpreted and stripped as hashline prefixes
Adjusted patch and replace validation to report a clear missing-path error when neither an entry path nor a top-level path is provided

[14.3.0] - 2026-04-25

Added

Added Markdown pipe-table row_N chunk selectors for row-level table edits.
Added resolveToolAlias export so tool names in CLI and session setup are normalized to canonical names, including mapping legacy read references to open
Added new open and open-chunk tool prompt documentation pages to describe canonical open usage for local files/directories, chunk reads, and URLs
Added full-output retrieval metadata to minimized shell command output by appending an artifact://<id> footer with byte counts, allowing users to open the original unminimized command output
Added streaming preview API exports from the package (resolveEditMode, EDIT_MODE_STRATEGIES, and chunk preview helpers) so editors can reuse mode-aware edit preview logic programmatically
Added shellMinimizer configuration options (enabled, settingsPath, only, except, and maxCaptureBytes) so users can control shell output minimization behavior

Changed

Changed the canonical file/URL reader tool from read to open across default tool lists and routing, including system prompts, plan mode, cursor handlers, and runtime tool registration
Changed runtime and UI handling to render and track open tool calls as first-class (with read accepted as legacy alias), including ACP mapping, session observers, and streaming message groups
Changed chunk edit guidance to document parser-specific region behavior, including TypeScript decorator/JSDoc sibling chunks, Python docstrings as body content, Python opaque nested chunks, Markdown whole-chunk fallbacks, ID volatility, and indentation display differences
Changed chunk deletion in chunk edit mode to require explicit delete: true; write: null and bare { path } entries now fail with guidance instead of deleting content.
Changed chunk edit validation to reject entries with multiple operation fields instead of choosing one and ignoring the rest.
Changed chunk edit validation to reject write: "" as an accidental destructive empty replacement; use the open tool for inspection or delete: true for deletion.
Changed chunk edit responses to warn when appending or prepending to a container without ~, since that inserts outside the container rather than inside its body.
Changed fetch output logging so URL-fetch artifacts now use .open.log naming instead of .read.log
Changed Bash interception guidance and errors to recommend open in place of read for cat/head/tail-style commands
Changed exported SDK tool surface to expose OpenTool as canonical and keep ReadTool as a compatibility alias
Changed session list loading to use parallel workers and fixed-size prefix reads per session file, reducing latency when loading many or large sessions
Changed edit call rendering to use mode-aware streaming diff previews, including multi-file chunk edit previews grouped by file path while arguments are still streaming
Changed shell execution in both interactive and non-interactive modes to route command output through the configured shell output minimizer
Changed default behavior so shell output minimization can now be toggled from settings without code changes
Changed shell output minimization to leave compound and piped commands unchanged; only a single eligible whole command is captured and minimized after it exits

Removed

Removed the chunk edit read: true operation; use the open tool to inspect chunks without modifying files.
Removed the replace: { old, new } chunk edit operation. Use write or insert for chunk edits instead.

Fixed

Fixed startup crashes on Linux systems where Bun's os.cpus() fails on non-contiguous CPU numbering (#779)
Fixed gh_pr_push so branches without gh_pr_checkout metadata fail instead of falling back to the tracked merge branch, and updated the GitHub tool setting copy to stop calling the tool group read-only (#778)
Fixed session list metadata extraction to better populate session titles and first-user summaries from partial session data when full JSONL parsing is unavailable
Fixed shell execution output to replace raw streamed bash output with the minimizer’s rewritten text before final output while still preserving the full original output as artifact metadata
Fixed bash command minimization to save the full unminimized output as a bash-original artifact during AgentSession shell execution, enabling artifact:// access to complete command output
Fixed streaming chunk previews that could display an incomplete trailing edit as a deletion when partial JSON temporarily converted in-flight values to null
Fixed edit streaming preview updates to cancel obsolete in-flight computations and avoid rendering stale previews as args change
Fixed chunk edits to reject unsafe ^/~ writes on code leaf chunks instead of falling back to whole-chunk replacement and risking structural indentation corruption
Fixed chunk replace operations to dedent multiline replacement snippets before reapplying the matched source indentation, preventing Python nested replacements from compounding indentation on repeated edits.
Fixed Go chunk trees to classify package clauses separately from imports and to avoid duplicating method receivers in method summaries.
Fixed chunk path-not-found guidance so it recommends sel="?" without claiming the already-shown listing must be re-read.
Fixed Markdown chunk appends to preserve blank-line separators after line-oriented inserts such as table rows
Fixed Markdown section region-fallback warnings to call out child chunks that will be replaced by whole-section edits.
Fixed rejected chunk-edit errors to distinguish current file content from hypothetical post-edit parse-error previews and to state when a same-file batch was rolled back.
Fixed unsafe Python head-region edits by rejecting decorated Python ^ writes and Python ^ deletes that can orphan indented bodies while still parsing.
Fixed Markdown table-row appends so row-shaped content lands inside the table block instead of after the trailing blank-line separator.
Fixed Markdown root writes to preserve fenced-code indentation verbatim.
Fixed Rust enum-variant replacement matching so trailing commas are included consistently with whole-variant writes.
Fixed streaming edit call headers to keep showing the target file path while the edit arguments are still arriving
Fixed Mermaid fenced markdown rendering in assistant messages on terminals without image protocol support (#650)
Fixed chunk edit path parsing so plan-mode edits to section-addressed local://PLAN.md:<selector> paths are classified as writes to the plan file
Fixed SQLite read helper queries to reject where= clauses with SQL control syntax that could override the structured selector's pagination; raw SQL remains available through q=SELECT ...

[14.2.0] - 2026-04-23

Added

Added an apply_patch edit mode that accepts Codex *** Begin Patch envelopes, shares patch-mode execution and diagnostics, and renders streaming per-file diffs in the TUI.

Changed

Changed Spark models to default to apply_patch edit mode instead of replace.
Tightened the contract for SearchParams.recency in web/search/providers/base.ts: providers MUST interpret recency as a pure time filter and MUST NOT use it as an implicit signal to change topic scope, content domain, or ranking strategy.
Inline read tool previews are now optional via read.toolResultPreview and default to off

Fixed

Fixed apply_patch streaming previews to avoid showing the missing *** End Patch parse error while the patch body is still arriving.
Fixed diagnostics rendering to replace tabs before TUI output, preventing compiler messages from breaking tree alignment.
Fixed compiled omp binaries to ignore project-local bunfig.toml and .env autoloading at startup, preventing unrelated project config from crashing or preloading code into the CLI
Fixed edit tool diff and replace operations to report missing-file failures as File not found: <path> errors instead of raw filesystem ENOENT errors
Fixed local:// URL path leak on Linux where // collapsing to / produced local:/path forms that bypassed the internal protocol handler and leaked as filesystem paths, breaking plan mode file resolution
Fixed Darwin compiled binaries failing to start under Bun 1.3.12 by ad-hoc signing local and release binary builds after applying Bun's no-codesign workaround (#754)
Fixed Tavily web search silently returning off-topic news articles when --recency was set. The provider was unconditionally coupling topic: "news" to recency, which scoped Tavily's index to news publications and excluded documentation, release notes, GitHub, and all non-news technical content. Technical queries with --recency now return the correct corpus.
Fixed status-line sanitization to strip OSC, DCS, PM, APC, and 8-bit CSI escape sequences instead of leaving payload fragments in the UI
Fixed inline read tool previews to avoid rendering duplicate summary rows above the same code cell

[14.1.3] - 2026-04-17

Breaking Changes

Replaced the legacy todo_write ops-based API (replace, update, add_task, and remove_task) with direct top-level fields, requiring migration of any callers using the old request shape
Removed in-place updates to existing task content, details, and notes via todo_write; note changes now append through add_notes
Phased task definitions in todo_write now reject notes on initial creation, so notes must be added later with add_notes

Added

Added complete, start, abandon, remove, add_notes, and add_tasks parameters to todo_write so callers can complete, jump to, drop, and annotate tasks without op wrappers
Added direct add_phase support as a top-level argument for inserting a new phase in todo_write
Added task.simple with default, schema-free, and independent modes so the task tool can disable task-call schema and shared context inputs while preserving agent-defined and inherited subagent schemas

Changed

Changed add_tasks to insert tasks by phase name or ID and allow multiple tasks to be added in one call

Fixed

Fixed task calls in schema-free and independent modes to return clear mode-specific errors when disallowed context or schema inputs are provided
Fixed newly generated session IDs to use UUIDv7 for new, forked, and branched sessions while preserving resumed session IDs

[14.1.1] - 2026-04-14

Breaking Changes

Removed the standalone vim tool from built-in tool lists, so vim-style editing is now invoked through edit in vim mode
Removed the searchDb field from session and extension tool contexts, so custom tools and extensions no longer receive a shared native search DB handle from ToolSession, CustomToolContext, ExtensionContext, and CreateAgentSessionOptions
Changed the vim tool API to require either open: "path" or kbd: [...] per call and removed direct line/col cursor parameters from open, so callers must position the cursor via key sequences after opening
Changed the edit schemas for patch, replace, hashline, and chunk modes from top-level request fields to edits array entries, requiring path/mode details on each edit and breaking callers that send legacy top-level path, old_text, new_text, op, move, or delete payloads

Added

Added Vim ex aliases :del, :ya, :co, and :mo as shorthand for existing delete, yank, copy, and move commands
Added support for additional Vim ex command aliases :write/write!, :edit/edit!, and :update/:up in command parsing
Added support for vim :global and :vglobal// variants as :g/pattern/d and :v/pattern/d parsing and execution
Added support for extra Vim operations by treating x, X, s, S, C, and D as delete/change operator aliases
Added support for new Vim motions gE/ge, g_, g*, g#, and |
Added support for C-f and C-b page motions in vim mode
Added C-u and C-o in vim insert mode to clear to line start and execute a one-off normal-mode command before returning to insert
Added insert-mode visual operators J, u, U, p, and P to join lines, convert case, and replace the selected region with register content
Added normal-mode line motions +, -, and _ to move to line offsets at the first non-blank character
Added * and # normal-mode commands to search forward or backward for the word under the cursor
Added gJ to join a line range, gv to restore the last visual selection, and ZZ/ZQ shortcuts for save-and-exit or exit-without-save in vim mode
Added paragraph text object p for ip/ap-style paragraph selection
Added support for Vim ex line-address forms like ., $, +N/-N, destination addresses such as :t$, and ranged :global commands
Added Vim ex :join/:j and :join!/:j! support to join addressed lines with or without whitespace normalization
Added a warning when chunk edits write to the ~ selector with body lines that appear over-indented, instructing users to start top-level body text at column 0
Added validation feedback for suspect indentation in chunk-mode ~ body writes so users can align content with the tool's automatic base indentation
Added support for multi-file edit calls across replace, patch, hashline, and chunk modes by grouping edits entries by file path and returning combined per-file results
Added per-edit path support in chunk entries so each operation can target explicit files when submitting mixed edits in a single request
Added support for computeHashlineDiff to accept hashline edits with loc and content payloads without requiring pre-resolved op fields
Added /rename <title> slash command to set an explicit session name, updating the session header and terminal tab title (#658)
Added session_name status line segment: displays the session name in the status bar right side with a stable hash-derived accent color unique to each name; shown in all presets when a name is set

Changed

Changed vim path normalization to accept colon-prefixed path values instead of rejecting them as Vim commands
Changed default providers.openaiWebsockets setting to off when unset, so OpenAI websocket transport is now disabled unless explicitly enabled
Changed Vim ex :update/:up execution to skip writing unchanged buffers and report buffer unchanged status
Changed Vim page-scroll commands C-f, C-b, C-u, and C-d to move in viewport-height based increments instead of fixed constants
Changed z command behavior so zt, zb, and z. now align cursor movement to first non-blank in the line
Changed :g/:v global command handling to process matching lines safely by working in reverse order and preserving file structure
Changed vim tab breadcrumb rendering from → to → in the editor view
Changed custom tool and task execution contexts to no longer expose a shared searchDb accessor, removing direct access to native grep/glob/fuzzyFind search backends from extension callbacks
Changed the task tool schema field to require JSON-encoded JTD schema text instead of a schema object, matching prompt guidance and task-subagent invocation
Changed chunk edit payloads to encode selectors as path: "file:selector" and updated chunk tool guidance and examples to match
Updated edit call/result rendering to show per-file diff sections and append a (+N more) hint when edits target multiple files
Grouped chunk-mode grep results by directory, file, and chunk so directory searches now render as hierarchical sections (#/##) with per-chunk anchor lines
Updated chunk-mode grep output to include match lines under their containing chunk entries with consistent line-number alignment based on file length
Changed eager todo enforcement to only apply on the first user message of a conversation, skipping subsequent user turns that may correct, clarify, or redirect the prior task

Removed

Removed live in-progress Vim tool previews during streaming call execution, so the TUI now shows only the last completed file viewport until the call finishes

Fixed

Fixed vim-mode multi-step line edits by auto-reordering ascending line-positioned commands to descending order before execution
Fixed Vim viewport rendering to display the inline highlighted cursor character and keep long cursor lines centered around the cursor in tool previews
Fixed Vim :global command defaults to handle only supported subcommands and report unsupported ones explicitly
Fixed Vim ex execution so parsed :update, :yank, and :put commands now run instead of falling through
Fixed vim tool rendering so streamed calls preview the live target viewport and large insert payloads update incrementally instead of popping in all at once
Fixed session event delivery so streaming message_update/tool-call previews reach the TUI immediately instead of waiting for extension handlers to finish
Fixed HTML session export rendering so background-job wait calls render as poll instead of stale await, while still recognizing legacy exported sessions
Fixed OpenRouter model resolution to accept dated routed selectors such as openrouter/z-ai/glm-4.7-20251222:nitro, inheriting metadata from the base catalog model when the exact variant is not listed yet
Fixed pre-execution edit preview routing so replace/patch/hashline mode diffs are computed from the new structured edit entries
Adjusted chunk/hashline/prompt guidance and validation to align with the refactored per-entry schema
Fixed chunk streaming output detection to verify chunk edits with chunkToolEditSchema, preventing non-chunk edit payloads from being rendered as chunk diffs
Fixed tool execution output to return the original toolResult text content from tools instead of sanitizing it before sending completion messages
Fixed session accent rendering in the status line and editor to reset only foreground color (\x1b[39m) so applying a session color no longer clears other ANSI styles
Session name sanitization: strip C0/C1 control characters (including ANSI ESC) from session names at storage time and in status line rendering, preventing escape sequence injection into TUI output
Auto-generated session titles no longer overwrite a name set via /rename: setSessionName now tracks whether the name was set by the user or auto-generated and silently ignores auto titles once a user name is in place; terminal title follows the same guard
Session accent border color now applied on session resume and after auto-title generation, not only after an explicit /rename
Fixed retained Python kernel ownership so AgentSession.dispose() only shuts down kernels owned by that session, including warmup-created kernels

[14.1.0] - 2026-04-11

Added

Added richer tool rendering details in session export HTML, including metadata badges, argument formatting, and todo task tree styling for exported tool and workflow messages
Added a persistent js tool backed by node:vm, with cross-session highway KV/pubsub, tool calls from inside JS cells, and $ / $$ interactive JavaScript execution
Added SQLite database read support to the read tool for .sqlite, .sqlite3, .db, and .db3 files with table listing, schema + sample output, row lookup, paginated query filtering, and read-only q=SELECT mode
Added SQLite mutation support to the write tool so db.sqlite:table inserts JSON5 rows and db.sqlite:table:key updates or deletes rows via row key
Added rendering of usage report entries for accounts with no usage limits, including account label and optional plan type with a -- no limits indicator
Updated account label resolution to fall back to email or accountId so unlabeled unlimited-plan accounts display a meaningful name
Added canonical model equivalence and provider coalescing across models.yml, enabledModels, --models, /model, and --list-models
Added equivalence overrides/exclusions to models.yml and modelProviderOrder to config.yml for global canonical-provider preference

Changed

Enabled await and cancel_job to be available when bash.autoBackground.enabled is set, so auto-backgrounded bash jobs can be awaited or cancelled without enabling async.enabled
Updated bash auto-background behavior so short commands returned inline output when they completed before the configured threshold, while longer runs moved to background jobs automatically
Replaced the LLM-callable Python execution path with JavaScript execution in the shared VM context, including updated renderers, prompts, session messages, and extension events
Updated interactive and CLI model listings/selectors to work with canonical model ids while resolving them to concrete provider variants for actual execution
Updated role assignment persistence so selected model settings now store the selector used by users, including thinking-level suffixes, while runtime continues to run against the resolved concrete provider model
Updated model scope resolution to expand exact canonical model ids into all matching provider variants when filtering supported model sets
Changed the agent to avoid giving time estimates or task-duration predictions in user responses, focusing on required work instead
Changed generated code guidance to avoid speculative abstractions and extra compatibility scaffolding, favoring direct implementations that match current needs
Changed model role resolution so roles can store either canonical model ids or explicit provider/model selectors while sessions continue to record the concrete model actually used
Updated bash execution to optionally auto-background long-running commands through the existing background-job pipeline, with dedicated settings for enabling the behavior and adjusting the delay

Fixed

Fixed session export rendering so JavaScript execution messages now use jsExecution labels and content instead of pythonExecution, matching current tool behavior
Fixed JavaScript cell execution to auto-display returned values once and preserve persistent VM bindings across calls until reset
Fixed .db/.db3 reads to verify SQLite file headers and fall back to normal file reading when the extension matches but the content is not a SQLite database
Fixed SQLite selector parsing and resolution to correctly route requests to database operations at the file-extension boundary instead of misrouting through plain file/archive handlers
Fixed unsupported or unsafe selectors by rejecting missing tables, composite primary keys for row lookups, unknown query parameters, and row operations on non-existent tables
Fixed model resolution for commit message generation, title generation, memory consolidation, and image inspection when role strings use canonical ids instead of raw provider/model values
Fixed default-model updates so previously configured thinking levels were preserved when reassigning a role
Fixed model scope and selection handling in CLI/session startup paths that previously failed to resolve aliases consistently across features
Fixed short-lived git subprocesses to disable core.fsmonitor and core.untrackedCache, avoiding unnecessary repository watchers and cache work during agent git operations

Security

Blocked destructive SQL execution in read-mode SQLite access by using read-only connections and rejecting bound-parameter raw SQL

[14.0.5] - 2026-04-11

Added

Added designer model role for UI/UX design tasks with Gemini 3.1 Pro as default model
Added support for model role fallback lists — roles can now resolve to multiple model patterns with automatic fallback to next available model
Added extractReadableFromHtml utility function to extract readable content from HTML with Readability article extraction and CSS selector fallback
Added support for GFM (GitHub Flavored Markdown) features including tables, strikethrough, and task lists in HTML-to-markdown conversion
Added resolveDiagnosticTargets utility function to handle glob pattern resolution with fallback to literal file paths for bracket-style paths

Changed

Clarified fenced code block editing behavior in markdown — the tool now preserves literal indentation inside fenced blocks, with content written verbatim as supplied
Updated guidance for inserting content after markdown section headings to use after on the heading chunk rather than before/prepend on the section itself
Reduced default image resize limits to 1568px (from 2000px) and 500KB (from 4.5MB) to match Anthropic's internal downscaling threshold and reduce payload sizes in tool calls
Adjusted screenshot compression to use 1024px max dimensions and 150KB budget for more aggressive optimization of browser screenshots in LLM requests
Updated JPEG quality defaults from 80 to 75 and refined quality ladder steps (70, 60, 50, 40) for tighter byte budgets
Improved image resize fast-path to skip re-encoding when images are already within dimensions and at ≤25% of byte budget, avoiding unnecessary processing of small icons and diagrams
Clarified that chunk names are truncated and must be copied from read or ? output rather than constructed from source identifiers
Enhanced guidance for editing fenced code blocks in markdown to preserve exact whitespace using raw reads, as the tool normalizes tabs to spaces which can damage indentation-sensitive content
Updated designer agent to use pi/designer role alias instead of explicit model list
Refactored model role resolution to support multiple fallback patterns per role, improving model availability handling
Replaced regex-based HTML-to-markdown conversion with Turndown library and GFM plugin for more accurate formatting of complex HTML structures
Simplified no-changes response to omit redundant response text when chunk content already matches
Clarified region suffix behavior on leaf and compound statement chunks — ~ and ^ now fall back to whole-chunk replacement with explicit guidance to supply complete structural content
Updated CRC refresh guidance to direct users to use CRCs from edit responses or run read(path="file", sel="?")
Added clarification that region suffixes fall back to whole-chunk replacement for prose and data formats (markdown, YAML, JSON, fenced code blocks, frontmatter)
Documented L20 shorthand syntax for single-line reads extending to end-of-file, with L20-L20 for one-line windows
Refactored diagnostic target resolution to use new resolveDiagnosticTargets function, consolidating glob pattern detection and file matching logic
Updated chunk selector syntax from @region format to ~ (body) and ^ (head) suffixes for more concise region targeting
Simplified chunk edit documentation to use new ~ and ^ region syntax instead of @head, @body, @tail, @decl keywords
Replaced internal raceAbort function with imported raceWithAbort utility from pi-utils
Refactored cleanup timer to use async iterator pattern with timers.setInterval instead of setInterval
Made #cleanupIdleSessions synchronous and moved async cleanup loop logic to new #runCleanupLoop method
Replaced regex-based htmlToBasicMarkdown with a Turndown + GFM plugin pipeline (tables, strikethrough, task lists, nested lists now convert correctly). Added direct turndown and turndown-plugin-gfm dependencies

Fixed

Fixed chunk edit tool to report file-not-found error distinctly when attempting to use chunk selectors on non-existent files, with guidance to use write tool or verify the path
Fixed stale child selector reuse to correctly match chunks by checksum when multiple sibling chunks with the same name exist under the same parent
Fixed stale diagnostics being reused after unrelated file publishes by clearing cached diagnostics before refreshing file state
Fixed Codex search to use streamed answer text when final answer is an image placeholder or empty

[14.0.4] - 2026-04-10

Added

Added PI_CHUNK_AUTOINDENT environment variable to control whether chunk read/edit tools normalize indentation to canonical tabs or preserve literal file whitespace
Added dynamic chunk tool prompts that automatically adjust guidance based on PI_CHUNK_AUTOINDENT setting without exposing a tool parameter
Added <instruction-priority>, <output-contract>, <default-follow-through>, <tool-persistence>, and <completeness-contract> sections to system prompt for improved long-horizon agent workflows

Changed

Updated chunk edit tool to apply normalizeIndent setting during edit operations, enabling literal whitespace preservation when PI_CHUNK_AUTOINDENT=0
Refactored environment variable parsing to use $flag() and $envpos() utilities from pi-utils for consistent boolean and integer handling across codebase
Updated system prompt communication guidelines to emphasize conciseness and information density, and added guidance on avoiding repetition of user requests
Enhanced system prompt with explicit rules for design integrity, verification before yielding, and handling of missing context via tool-based retrieval
Added PI_CHUNK_AUTOINDENT to control whether chunk read/edit tools normalize indentation, and updated chunk prompts to switch guidance automatically based on that setting
Refined the default system prompt with explicit instruction-priority, output-contract, tool-persistence, completeness, and verification rules for long-horizon GPT-5.4-style agent workflows

Fixed

Fixed typo in system prompt: 'backwards compatibiltity' → 'backwards compatibility'

[14.0.3] - 2026-04-09

Fixed

Fixed cached Ollama discovery rows so upgraded installs switch to the OpenAI Responses transport instead of staying on the old completions transport

[14.0.2] - 2026-04-09

Added

Added /force slash command to force the next agent turn to use a specific tool
Added ToolChoiceQueue for managing tool-choice directives with lifecycle callbacks and requeue semantics
Added setForcedToolChoice() method to AgentSession to programmatically force tool invocations
Added toolChoiceQueue property to AgentSession for direct queue access
Added peekQueueInvoker() method to AgentSession to retrieve in-flight tool invocation handlers
Added queueResolveHandler() function as the canonical entry point for preview/apply workflows
Added buildToolChoice() and steer() methods to ToolSession for tool-choice queue integration
Added getToolChoiceQueue() method to ToolSession for accessing the tool-choice queue
Added support for embedded URL selectors (:raw and :L#-L# line ranges) in read command paths
Added parseReadUrlTarget function to parse and validate URL read targets with line range support
Added decl region to chunk selector for targeting declarations without leading trivia
Exported hooks subpath for extensibility API access
Added build script for compiling binary artifacts

Changed

Refactored pending action handling from PendingActionStore to ToolChoiceQueue with generator-based directives
Changed tool-choice override mechanism from simple override to a queue-based system with callbacks
Updated ResolveTool to dispatch to in-flight queue invokers instead of popping from a pending action store
Updated custom tool loader to accept pushPendingAction callback instead of PendingActionStore instance
Updated AstEditTool to use queueResolveHandler() for preview/apply semantics
Changed eager-todo prelude to use the tool-choice queue instead of simple override
Updated todo reminder suppression to check for user-forced directives via consumeLastServedLabel()
Made model-specific edit mode defaults conditional on PI_STRICT_EDIT_MODE environment variable for greater flexibility in edit mode selection
Updated slash command handlers to support returning remaining text as prompt input instead of consuming input entirely
Enhanced slash command parser to recognize both whitespace and colon (:) as command argument separators
Updated indentation guidance for chunk edit content to use single leading spaces per indent level instead of tabs
Updated read CLI to delegate URL inputs through the read tool pipeline instead of treating them as local file paths
Updated chunk edit documentation to clarify region semantics and emphasize using the narrowest region for edits
Improved chunk selector guidance with visual diagram showing region boundaries
Renamed build:binary script to build
Refactored check script to include linting via Biome and type checking
Added check:types, lint, fmt, and fix scripts for improved developer workflow
Simplified TypeScript configuration by extending workspace-level config

Removed

Removed PendingActionStore class and related pending-action module
Removed pendingActionStore parameter from AgentSession config
Removed pendingActionStore from ToolSession interface
Removed consumeNextToolChoiceOverride() method from AgentSession (replaced by nextToolChoice())

Fixed

Fixed tool-choice queue cleanup on agent loop abort to prevent orphaned in-flight directives
Fixed requeue semantics to preserve onInvoked and onRejected callbacks across multiple abort cycles

[14.0.1] - 2026-04-08

Changed

Improved auto-generated file detection to gracefully handle ENOENT errors when peeking file content, preventing unnecessary abort failures
Optimized context emission by skipping message cloning when no extensions have context handlers
Improved message cloning resilience by falling back to shallow array clone when structured cloning fails due to non-cloneable objects
Made assertEditableFileContent synchronous instead of async for improved performance in streaming edit checks
Enhanced streaming edit abort detection to check for auto-generated files as soon as the file path is available, rather than waiting for the full diff
Improved file prefix reading in session storage to use peekFile utility from @oh-my-pi/pi-utils for better efficiency
Moved image metadata detection to @oh-my-pi/pi-utils package for shared use across projects
Simplified image loading API by removing redundant metadata parameters and consolidating image utilities
Updated imports to use readImageMetadata and parseImageMetadata from @oh-my-pi/pi-utils instead of local implementations

Removed

Removed image-input.ts utility module; functionality consolidated into image-loading.ts
Removed mime.ts utility module; MIME detection moved to @oh-my-pi/pi-utils
Removed ImageMetadata interface and ReadImageMetadataOptions from local codebase

Fixed

Fixed streaming edit abort for auto-generated files by adding LRU caching and early path-based detection to prevent unnecessary edits

[14.0.0] - 2026-04-08

Breaking Changes

Simplified chunk edit operations: removed append_child, prepend_child, append_sibling, prepend_sibling, and replace_body ops in favor of unified replace, before, after, prepend, and append with region targeting (@head, @body, @tail)
Chunk edit target format changed: now accepts selector#CRC@region for mutations and selector@region for insertions; removed separate crc and anchor fields from edit operations
Removed checksum requirement from insert operations (before, after, prepend, append); only replace requires #CRC suffix

Added

Auto QA tool (report_tool_issue) for automated tracking of unexpected tool behavior; enabled via PI_AUTO_QA=1 environment variable or dev.autoqa setting
dev.autoqa setting to enable automated tool issue reporting for all agents
System prompt guidance when report_tool_issue tool is available, encouraging agents to report tool behavior discrepancies
LSP server discovery at startup via discoverStartupLspServers() to detect configured language servers without blocking initialization
LSP startup event channel (lsp:startup) for asynchronous server warmup notifications with completion or failure status
LspStartupServerInfo type for tracking LSP server status including connecting, ready, and error states
LSP server status display in /info command showing connecting, ready, and error states with color-coded indicators
Multi-session support in ACP mode: agents can now manage multiple concurrent sessions with independent state, models, and configurations
Session forking in ACP mode: unstable_forkSession creates a new session from an existing one's history
Session resumption in ACP mode: unstable_resumeSession reloads a previously saved session
Session closure in ACP mode: unstable_closeSession cleanly shuts down a session and releases resources
Model state reporting in ACP mode: SessionModelState with available models and current selection in session responses
Direct model setting in ACP mode: unstable_setSessionModel RPC command for changing the active model
Turn-level usage tracking in ACP mode: prompt responses now include usage with input/output/cached token counts
Message ID tracking in ACP mode: stable message IDs for assistant chunks enabling client-side message correlation
Settings cloning: Settings.cloneForCwd() method to create isolated settings instances for different working directories
Extension flag value retrieval: ExtensionRunner.getFlagValues() to inspect current flag state
Exported autoresearch module and submodules via ./autoresearch and ./autoresearch/* package paths
Exported autoresearch tools via ./autoresearch/tools/* package path
Exported CLI commands via ./cli/commands/* package path
Exported DAP module and submodules via ./dap and ./dap/* package paths
Exported edit module and submodules via ./edit, ./edit/*, and ./edit/modes/* package paths
Exported bundled ci-green custom command via ./extensibility/custom-commands/bundled/ci-green package path
Exported extensibility plugins marketplace via ./extensibility/plugins/marketplace and ./extensibility/plugins/marketplace/* package paths
Exported ACP mode via ./modes/acp and ./modes/acp/* package paths
Exported web utilities via ./web/* package path
Exported line-hash utilities from edit module via ./edit/line-hash
Host-owned custom tools support: RPC clients can now register custom tools via setCustomTools() and the RPC server will invoke them over the transport with host_tool_call requests
RPC host tool framework: RpcHostToolBridge for managing host tool execution, RpcHostToolDefinition for tool metadata, and bidirectional host_tool_call, host_tool_cancel, host_tool_update, and host_tool_result frames
RPC client tool API: defineRpcClientTool() helper, RpcClientCustomTool interface, and RpcClientToolContext for implementing host-side tool execution with update streaming and abort support
set_host_tools RPC command to replace the active set of host-owned tools before the next model call
refreshRpcHostTools() method on AgentSession to integrate host tools into the active tool registry with conflict detection and auto-activation of non-hidden tools
Instruction breakpoints support: set_instruction_breakpoint and remove_instruction_breakpoint debug actions for setting breakpoints at specific instruction addresses
Data breakpoints support: data_breakpoint_info, set_data_breakpoint, and remove_data_breakpoint debug actions for monitoring variable/memory access
Memory introspection: read_memory and write_memory debug actions for inspecting and modifying debugger memory
Disassembly support: disassemble debug action for viewing assembly instructions with symbol resolution
Module and source introspection: modules and loaded_sources debug actions for querying loaded modules and source files
Custom DAP requests: custom_request debug action for sending arbitrary Debug Adapter Protocol commands
Reverse request handling in DAP client: onReverseRequest() method for responding to adapter-initiated requests like runInTerminal and startDebugging
DAP reverse request support: adapters can now request terminal spawning and child debug sessions via runInTerminal and startDebugging reverse requests
Instruction pointer reference in debug snapshots: instructionPointerReference field in session summaries for low-level debugging
Hit condition support for instruction and data breakpoints: hit_condition parameter for conditional breakpoint triggering
RPC set_todos command and todoPhases in get_state, allowing hosts to pre-seed and inspect session todo state over the protocol
Deferred diagnostics support in LSP writethrough: onDeferredDiagnostics callback and deferredSignal in WritethroughOptions allow callers to receive diagnostics that arrive after the main 5-second timeout
Language detection for .pm (Perl modules), .astro (Astro framework), and special filenames containerfile and justfile
Workspace-scoped diagnostics and reload actions via * file parameter; diagnostics action now supports * for workspace-wide diagnostics across all configured servers
Socket-mode DAP adapter support for debuggers like dlv that communicate via network sockets instead of stdio; Linux uses unix domain sockets, macOS/other platforms use TCP with client-addr dialing
Improved extensionless binary debugging: native debuggers (gdb, lldb-dap) and adapters with root markers are now preferred over unrelated adapters like debugpy
Debug tool with DAP (Debug Adapter Protocol) support for launching and attaching debuggers, setting breakpoints, stepping through execution, inspecting threads/stack/variables, and evaluating expressions
Debug adapter configuration for gdb, lldb-dap, debugpy, and dlv with language/file-type matching and root marker detection
Debug session management with support for source and function breakpoints, conditional breakpoints, stack trace inspection, scope/variable exploration, and program output capture
debug.enabled setting to control debug tool availability
Chunk read formatting: anchorStyle (full / kind / bare), read.anchorstyle setting, and chunked flag on file display mode
read.prosechunks and read.explorechunks settings for prose chunk trees and checksum-free explore trees
Handlebars helpers anchor and sel (with template anchorStyle context) for chunk examples in prompts
Chunk-mode grep lines as path:selector>LINE|content; unified grep tool template behind IS_CHUNK_MODE
lru-cache for chunk tree caching
Chunk-mode read output: recursive rendering with $XXXX checksum suffixes, inline large-chunk previews, and normalized #path$XXXX selectors between read and edit
Autoresearch: init_experiment options new_segment, from_autoresearch_md, abandon_unlogged_runs; log_experiment options skip_restore and broader force; run_experiment force (with warnings); pre-run dirty-path tracking; abandonUnloggedAutoresearchRuns and abandonedAt on runs
LSP: diagnostic versioning (versionSupport, stored document version per diagnostic set), and waitForDiagnostics / getDiagnosticsForFile options (expectedDocumentVersion, allowUnversioned)

Changed

Extracted working directory formatting logic into formatToolWorkingDirectory() utility for consistent path display across tools
Bash command rendering now sanitizes tabs and shortens home directory paths in command previews
Chunk edit tool schema: renamed target parameter to sel for consistency with read tool terminology
Chunk edit tool: op parameter is now required (previously optional with replace default)
Chunk edit documentation: updated all region references from @inner to @body for clearer semantics
Chunk edit documentation: expanded with comprehensive real-world examples showing full read output, operation effects, and indentation rules
Chunk edit documentation: simplified indentation guidance to write content at indent-level 0 with automatic re-indentation by the tool
Chunk edit documentation: clarified that @region only works on container chunks, not leaf chunks
Chunk edit documentation: emphasized that CRCs change after every edit and must be refreshed from latest responses
Read chunk tool documentation: updated selector examples to use @body instead of @inner
Chunk edit region terminology updated: @inner renamed to @body for clearer semantics in container chunks
Chunk edit documentation restructured with comprehensive examples showing full read output, operation effects, and indentation rules
Chunk edit indentation guidance simplified: content should be written at indent-level 0 and the tool automatically applies correct base indentation
Chunk edit examples expanded with realistic TypeScript code samples demonstrating replace, insert, prepend, append, and delete operations
Python tool description now dynamically reflects prelude documentation availability instead of static text
Python tool now automatically warms the environment on first execution if prelude helpers are unavailable, ensuring documentation is loaded before use
Tool creation now auto-injects report_tool_issue when auto QA is enabled, regardless of requested tool list
Chunk edit region names standardized to @head, @body, and @tail for clearer semantics
Chunk edit documentation clarified: region defaults to full chunk when omitted; leaf chunks no longer support region targeting
Chunk read documentation updated: selector examples now use region-specific selectors based on @head, @body, and @tail
LSP server connecting status in welcome banner now uses muted pending symbol instead of warning symbol for clearer visual distinction
Codex websocket prewarm now runs asynchronously in the background instead of blocking session creation, allowing faster startup
Codex websocket status updates now display in interactive mode when prewarm completes or fails
LSP server warmup now runs asynchronously in the background instead of blocking session creation, allowing faster startup
LSP servers returned from createAgentSession() now include connecting status during initial warmup phase
Interactive mode now subscribes to LSP startup events and displays status updates and error messages to the user
LSP server status in /info command now distinguishes between connecting (yellow), ready (green), and error (red) states
ACP agent now manages multiple sessions instead of a single session; session lifecycle and configuration are now per-session
ACP session creation now uses a factory function to support creating new sessions for different working directories
ACP event mapping now accepts optional getMessageId callback for stable message ID assignment to assistant chunks

Removed

Deleted src/utils/prompt-format.ts module; prompt formatting logic moved to pi-utils
Deleted src/utils/frontmatter.ts module; frontmatter parsing logic moved to pi-utils
Removed waitForChildProcess utility (child process termination now handled by native killTree from pi-natives)
grep-chunk.md (folded into unified grep template)
startMacAppearanceObserver export (use MacAppearanceObserver.start())
copyToClipboard export from pi-natives
PI_CHUNK_SPLICES env and chunkSplicesEnabled()
Autoresearch segmentFingerprint and related config hashing

Fixed

Chunk edit parameter validation: corrected detection of chunk edit operations to check for sel field instead of target
Chunk edit streaming previews: updated to reference sel parameter instead of target
Python prelude introspection now respects execution timeout and signal options, preventing hangs during environment warmup
Welcome banner LSP server status now updates in real-time when background startup warmup completes, eliminating stale connecting status displays
Welcome banner LSP startup rows now re-render when background warmup finishes, use the pending status symbol while servers are still connecting, and no longer add a redundant LSP ready status line on successful startup
ACP session initialization now registers connection cleanup handlers to dispose all sessions on disconnect
Reorganized package.json exports: moved ./edit exports before ./plan-mode for better logical grouping
Notebook conversion logic now checks for raw read mode or non-chunk mode before converting via markit, allowing chunk-mode reads of .ipynb files to use chunk parsing instead of conversion
Go receiver methods now render as top-level siblings instead of nested under their receiver type in chunk read output
Moved prompt formatting and rendering utilities from coding-agent to pi-utils package; renderPromptTemplate() and formatPromptContent() now accessed via prompt.render() and prompt.format() from @oh-my-pi/pi-utils
Moved parseFrontmatter() utility from coding-agent to pi-utils package; now imported from @oh-my-pi/pi-utils instead of local utils
Consolidated prompt template handling: TemplateContext type now available as prompt.TemplateContext from @oh-my-pi/pi-utils
DAP initialization now advertises support for runInTerminal and startDebugging reverse requests, and memory references
Debug tool schema expanded with new parameters for instruction/data breakpoints, memory operations, and custom requests
DAP session state now tracks instruction and data breakpoints separately from source breakpoints
Replaced Bun.which() with $which() from pi-utils for command resolution
Chunk edit tool documentation restructured: replaced operation-specific examples with region-based guidance and canonical indentation rules
Chunk read documentation updated: selectors now support region syntax (e.g., class_Foo.fn_bar#ABCD@body) and canonical target listings show supported regions per chunk
Chunk edit schema simplified: target description now documents region format; op and content descriptions clarified for region-aware operations
Chunk edit streaming previews updated: labels now reflect region-aware operations (e.g., append instead of append child, insert after without anchor reference)
Removed CRC parsing from parseChunkSelector() and parseChunkReadPath(): selectors no longer extract embedded checksums
Chunk edit normalization simplified: no longer requires async checksum resolution or context-dependent operation mapping
RPC mode now automatically disables session title generation by default; hosts can opt in with PI_RPC_EMIT_TITLE=1 environment variable to receive title updates
RPC mode now resets workflow-altering todo.*, task.*, and async.* settings to built-in defaults instead of inheriting user overrides
RPC mode now disables automatic session title generation by default and suppresses setTitle extension UI requests unless hosts opt in with PI_RPC_EMIT_TITLE=1
Reorganized edit tool implementation from patch/ to edit/ directory structure with dedicated mode subdirectories (edit/modes/chunk.ts, edit/modes/hashline.ts, edit/modes/patch.ts, edit/modes/replace.ts)
Updated package.json exports to use ./edit path instead of ./patch for edit tool and related utilities
Chunk edit tool documentation simplified: removed line-based edit examples, clarified target format with full path and CRC suffix, added guidance for replace_body operation to preserve declarations
LSP diagnostics timeout reduced from 10 seconds to 5 seconds for faster feedback; slow diagnostics now fetch in background via deferred mechanism
Diagnostics action error messaging clarified: requires file parameter or * for workspace scope; improved guidance in error responses
Workspace symbols and reload actions now accept * to operate across all configured servers instead of requiring a file path
DAP session initialization now subscribes to stop events before launching/attaching to avoid missing stopOnEntry events
Stack frame fetching moved outside the event dispatch loop to prevent deadlocks and improve responsiveness
Evaluate requests now default to the top stopped frame when frameId is not explicitly provided
Eager todo enforcement now skips prompts ending with question marks or exclamation marks, treating them as queries or commands rather than statements requiring task planning
Chunk read output now displays fully-qualified anchor paths (e.g., [class_Worker.fn_run#CRC]) instead of bare names, making targets unambiguous for edits
Chunk edit tool documentation clarified: target must be the fully-qualified path with #CRC suffix; added guidance to run read(path="file", sel="?") for canonical target listings when anchor style is unclear
Chunk read tool documentation updated: sel parameter now documents the ? selector for canonical target listings, and clarifies that default output shows full paths
Chunk edit schema and tool contract: explicit op (replace, append, prepend, after, before); use replace with empty content to remove a chunk (no separate delete op); sibling inserts use anchor instead of separate after/before target fields; insert ops omit CRC where appropriate, mutations require checksum on target
Chunk path handling: parse selector and CRC separately, sanitize selectors (strip filename prefixes, uppercase checksums), accept embedded #CRC on targets, auto-accept stale CRC for later ops in the same batch on the same chunk
Chunk UX: streaming and final edit previews show chunk edits next to hashline edits with op-specific labels; prompt docs shortened with rules table, … in examples, and helper-based path/anchor samples
log_experiment only reverts files modified by the run; prompts and errors document that pre-existing dirty files are preserved; richer pending-run error context; init_experiment no-ops when the contract matches unless new_segment; secondary metrics informational only (no force for drift)
Autoresearch: log_experiment reloads benchmark/scope/constraints from autoresearch.md after resolving a pending run; /autoresearch without autoresearch.md follows /plan-style toggle and message flow; setup moved into autoresearch system prompt (removed command-initialize.md)
Native/shell alignment: GrepOutputMode from pi-natives; shell and getDiagnosticsForFile callbacks use error-first (err, chunk); getDiagnosticsForFile takes an options object
Clipboard: copyToClipboard / readImageFromClipboard live in utils/clipboard.ts (OSC 52 and Termux)
macOS: session-wide power assertion while the agent runs; MacAppearanceObserver.start() with error-first callback; detectMacOSAppearance() returns enum values
ACP session cleanup now properly cancels in-flight prompts and disposes resources when sessions are closed or connection aborts
Removed unused _createErrorToolResult helper function from RPC host-tools module
Fixed Go receiver method indentation in append operations to preserve relative indentation from the anchor chunk
Fixed Go type chunk line counts to report only the type body lines instead of including grouped receiver methods
Fixed enum variant insertion to avoid adding extra blank lines between variants
Chunk read output now correctly preserves embedded CRC in selectors (e.g., class_Foo.fn_bar#ZZPM) instead of stripping them during path parsing
Chunk edit error messages now consistently report checksum mismatches with format Checksum mismatch instead of variable phrasing
Chunk-mode read output now correctly displays scoped response trees showing only touched chunks and adjacent siblings, preventing unrelated distant chunks from appearing in responses
DAP stopped event handling no longer blocks the message reader, preventing potential deadlocks during rapid event sequences
Chunk-mode whole-chunk replaces now preserve attached leading comments and docblocks when replacement content starts at the declaration, preventing accidental comment loss during agent edits
Chunk edit error messages now consistently report checksum mismatches with the format did not match checksum "XXXX" instead of variable phrasing
Chunk selector validation for edits now rejects non-canonical selectors (suffix-only like fn_run or prefix-stripped like run), requiring fully-qualified paths to prevent ambiguity
Plan review previews now re-append at the chat tail on refresh, keeping them adjacent to the active selector instead of updating off-screen
log_experiment validates and reverts run-scoped file changes without clobbering unrelated dirty worktree state
Chunk edit targets that embed CRC in the selector (e.g. fn_foo#ABCD) parse correctly
Shell paths check errors before consuming chunk output (bash executor, config resolution)
/autoresearch toggles like /plan when empty; slash completion no longer suggests off/clear on an empty prefix after the command
Chunk-mode read/edit edge cases (zero-width gap replaces, stale batch diagnostics, grouped Go receivers, line-count headers, parse error locations)

[13.19.0] - 2026-04-05

Added

Added idle auto-compaction settings and scheduling so sessions can compact after inactive turns without auto-continuing.
Added onExternalEditor callback to extension UI dialog options for handling external editor shortcut in select dialogs
Added external editor shortcut support in plan review selector, allowing users to open and edit the plan in their configured editor
Added matchesAppExternalEditor keybinding matcher to detect external editor shortcut (Ctrl+G or configured binding)
Added trimTrailingNewline option to openInEditor function to preserve trailing newlines when editing files
Added GitHub CLI utilities to git module (utils/git.github) with available(), run(), json(), and text() methods for GitHub CLI operations
Exported git utilities from main package entry point for use by extensions
Added comprehensive git utility module (utils/git) with organized namespaces for common git operations (branch, commit, diff, log, patch, ref, stage, status, head, repository)

Changed

Changed idle compaction settings (compaction.idleThresholdTokens and compaction.idleTimeoutSeconds) from enum to numeric type for flexible configuration
Modified secret obfuscation to deobfuscate restored session messages for local display while keeping outbound LLM messages obfuscated
Updated stash pop operation to preserve staged changes with --index flag when restoring after task branch merges
Changed secret placeholders to deterministic hash-style redaction tokens and deobfuscated assistant output for local display.
Updated hook editor and hook selector components to use matchesAppExternalEditor matcher for consistent external editor keybinding detection
Modified plan review flow to read the latest plan content from disk before approval, allowing changes made in external editor to be reflected
Enhanced plan review help text to dynamically display the configured external editor keybinding
Refactored git operations to use centralized utility module instead of ControlledGit class throughout codebase
Replaced ControlledGit dependency injection pattern with direct cwd parameter in commit agent tools
Migrated git HEAD resolution in footer and status-line components to use new synchronous and asynchronous utilities
Updated git status summary calculation in status-line component to use new git utility API
Simplified git branch operations in task execution and cleanup to use new utility functions
Refactored patch application logic in task worktree to use new git patch utilities

Removed

Removed gh-cli.ts module; GitHub CLI functionality now available via utils/git.github
Removed ControlledGit class and associated git wrapper infrastructure from commit/git module
Removed mergeStdoutStderr helper function from autoresearch git utilities
Removed findGitHeadPathAsync and findGitHeadPathSync from modes/shared module (replaced by git utilities)
Removed ./commit/git export from package.json (internal diff parsing still available via ./commit/git/*)

Fixed

Fixed idle compaction timer to properly cancel when event controller is disposed, preventing memory leaks
Fixed session resumption to preserve the last non-empty session when starting a fresh session
Fixed stash detection to use git ref resolution instead of output parsing for reliable stash state tracking
Fixed isolated task merge-back to preserve task outputs on merge failure and stash dirty worktrees before cherry-pick.
Fixed web search source rendering to truncate long title, metadata, and URL lines before they overflow the UI.
Fixed PR checkout tool to resolve symlinks in worktree paths, ensuring consistent path references in results and metadata
Fixed read output for file-backed internal URLs like local://... to include hashline prefixes in hashline edit mode, preserving usable line refs for follow-up edits
Fixed the plan review selector to support the external editor shortcut for opening and updating the current plan from the approval screen

[13.18.0] - 2026-04-02

Breaking Changes

Removed standalone fetch tool; URL fetching is now integrated into the read tool

Added

Added URL reading capability to read tool with support for web pages, GitHub issues, Stack Overflow, Wikipedia, Reddit, NPM, arXiv, technical blogs, RSS/Atom feeds, and JSON endpoints
Added offset and limit parameter support for paginating cached URL fetch results
Added URL caching mechanism to avoid redundant network requests when reading the same URL multiple times

Changed

Renamed fetch.enabled setting to Read URLs with updated description to reflect integration with read tool
Updated read tool to accept timeout and raw parameters for URL handling
Updated read tool to support file:// URLs for local file paths
Removed fetch tool from agent tool lists (explore, librarian, oracle, plan, reviewer agents)

Fixed

Fixed read tool to properly handle file:// URL scheme by converting to filesystem paths

[13.17.5] - 2026-04-01

Added

Added support for writing to ZIP archives using fflate library for cross-platform compatibility

Changed

Modified archive writing to detect and handle ZIP files separately from Bun archives

Removed

Removed GhPrPushTool test case

[13.17.4] - 2026-04-01

Added

Support for writing to archive entries in .tar, .tar.gz, .tgz, and .zip files using archive.ext:path/inside/archive syntax
Ability to create new archives when writing to archive subpaths that don't yet exist

[13.17.3] - 2026-04-01

Added

Added support for converting Jupyter notebooks (.ipynb) to markdown via markit
Added markit-ai npm package for native document and notebook conversion
Added support for reading files from .tar, .tar.gz, .tgz, and .zip archives using virtual subpaths like archive.ext:path/to/file
Added ability to list archive contents and navigate subdirectories within supported archive formats
Added archive-aware read support for .tar, .tar.gz, .tgz, and .zip, including virtual subpaths like archive.ext:path/to/file
Added /tools slash command to show the tools currently visible to the agent in the interactive session

Changed

Replaced Python-based markitdown CLI tool with native markit-ai library for document conversion
Updated document conversion to use markit library instead of external markitdown command
Removed markitdown from Python tools manager (no longer needed as external dependency)
Updated read tool documentation to reflect archive support and usage patterns

[13.17.2] - 2026-04-01

Added

Added /marketplace help command to display usage guide for all marketplace operations
Added dedicated gh-renderer.ts module for rich terminal rendering of GitHub Actions workflow runs with live status snapshots and job details
Added gh_pr_checkout tool to check out GitHub pull requests into dedicated git worktrees with contributor push metadata
Added gh_pr_push tool to push checked-out pull request branches back to their source branches
Added gh_repo_view tool to read GitHub repository metadata using the local GitHub CLI
Added gh_issue_view tool to read GitHub issues with optional comment context
Added gh_pr_view tool to read GitHub pull requests with optional comment context
Added gh_pr_diff tool to read GitHub pull request diffs with optional file filtering
Added gh_search_issues tool to search GitHub issues with repository scoping
Added gh_search_prs tool to search GitHub pull requests with repository scoping
Added gh_run_watch tool to watch GitHub Actions workflow runs, fast-fail on job failures, and stream tailed logs for failed jobs
Added github.enabled setting to enable read-only gh_* GitHub CLI tools for repository, issue, pull request, diff, and search workflows
Added bundled /green command to generate iterative CI fix prompts with optional tag instructions when HEAD is tagged
Added github.enabled setting to enable read-only gh_* GitHub CLI tools for repository, issue, pull request, diff, and search workflows
Added gh_repo_view tool to read GitHub repository metadata using the local GitHub CLI
Added gh_issue_view tool to read GitHub issues with optional comment context
Added gh_pr_view tool to read GitHub pull requests with optional comment context
Added gh_pr_diff tool to read GitHub pull request diffs with optional file filtering
Added gh_search_issues tool to search GitHub issues with repository scoping
Added gh_search_prs tool to search GitHub pull requests with repository scoping
Added gh_run_watch tool to watch GitHub Actions workflow runs, fast-fail on job failures, and stream tailed logs for failed jobs
Added bundled /green command to generate iterative CI fix prompts with optional tag instructions when HEAD is tagged
Project-scoped marketplace plugin installs: omp plugin install --scope project name@marketplace and /marketplace install --scope project name@marketplace install plugins into the nearest .omp/ or .git-rooted project directory instead of the user directory (#581)
--scope user|project flag added to /marketplace uninstall, /marketplace upgrade, /plugins enable, and /plugins disable to disambiguate when a plugin is installed in both scopes
omp plugin upgrade --scope project with no plugin ID warns that --scope is ignored for bulk upgrades
Added opt-in gh_* GitHub CLI tools behind the github.enabled setting for repository, issue, pull request, diff, and search workflows
Added opt-in gh_run_watch to fast-fail GitHub Actions runs, stream job status snapshots, and return tailed logs for failed jobs
Added bundled /green command to generate the iterative “fix CI until green” prompt, with final tag instructions included only when HEAD already has a tag

Changed

Improved marketplace catalog parsing to skip invalid plugin entries with warnings instead of failing the entire catalog load
Enhanced /marketplace discover command to suggest adding the official marketplace when no plugins are available
Improved /marketplace command messaging with clearer guidance for first-time setup and available commands
Enhanced gh_run_watch tool call rendering to display animated spinner status and target description (run ID, branch, or current HEAD) with improved visual hierarchy
Enhanced gh_pr_view tool to include inline review comments alongside pull request reviews for improved discussion context
Improved gh_run_watch tool output rendering with dedicated visual component for streaming run snapshots and job status updates

Fixed

Fixed marketplace error messages to display error details instead of object stringification
Fixed artifact storage for non-persistent sessions to use in-memory fallback instead of returning undefined, enabling proper spill truncation for all session types
Fixed prompt file formatting to include trailing newlines at EOF for consistency across all prompt markdown files
Fixed gh_pr_diff to preserve raw patch content instead of normalizing tabs and whitespace
Fixed gh_pr_view to include inline review comments alongside pull request reviews and issue-style comments for discussion context
Fixed gh_run_watch to resolve explicit branch watches against the selected branch head instead of local HEAD
Fixed gh_run_watch to hide repo and polling internals from the tool schema and save full failed-job logs as session artifacts alongside the inline tailed output
Fixed gh_* tool outputs to spill full large results to artifacts instead of pre-truncating the head with unusable offset= guidance
Fixed bundled /green to watch the workflow runs for the current HEAD commit instead of whichever branch run was newest
Fixed OpenAI Responses session rehydration to strip stale assistant replay payloads before resumed requests (#594 by @daandden)
Fixed inline image rendering to cap image height and preserve multiplexer scrollback during terminal resizes (#587 by @smileynet)

[13.17.1] - 2026-04-01

Removed

Removed code_search tool for code snippet and documentation search

Fixed

Fixed edit tool diff rendering to wrap long diff lines with continuation gutters instead of truncating them at terminal width (#578)
Fixed --list-models and /model provider filtering to hide models from disabled providers (#588)
Fixed edit tool diffstats to use diff-specific add/remove theme colors instead of success/error status colors (#589)

[13.17.0] - 2026-03-30

Added

Added marketplace.autoUpdate setting (off/notify/auto, default notify) for automatic plugin update checking on startup
Added background marketplace catalog refresh on startup when catalogs are stale (>24h)
Added /marketplace upgrade [name@marketplace] slash command to upgrade outdated plugins
Added omp plugin upgrade [name@marketplace] CLI command for plugin upgrades
Added checkForUpdates(), upgradePlugin(), upgradeAllPlugins(), and refreshStaleMarketplaces() to MarketplaceManager
Added marketplace plugin system: registry types, ID helpers, atomic read/write for marketplaces.json and installed_plugins.json (Claude Code-compatible format)
Added MarketplaceManager orchestrator for marketplace and plugin lifecycle (add/remove/update marketplaces, install/uninstall/enable plugins)
Added marketplace fetcher with source classification (GitHub, git, URL, local) and catalog validation
Added plugin source resolver with pathIsWithin containment checks and versioned cache manager
Added CLI commands: omp plugin marketplace add|remove|update|list, omp plugin discover [marketplace]
Added classifyInstallTarget() to distinguish name@marketplace from npm install targets
Extended listClaudePluginRoots() to read OMP's installed plugins registry alongside Claude Code's, with OMP as authoritative for duplicate plugin IDs
Added --plugin-dir <path> repeatable CLI flag for loading plugins from local directories
Added /reload-plugins slash command that invalidates fs content cache and plugin roots cache
Added printPluginHelp() entries for marketplace and discover commands
Added MCP server loading from marketplace plugin .mcp.json files with ${CLAUDE_PLUGIN_ROOT} variable substitution
Added skill and command namespacing for marketplace plugins (plugin-name:skill-name)
Added LSP config loading from marketplace plugin roots via getPreloadedPluginRoots()
Wired --plugin-dir runtime injection into plugin roots at session startup with highest precedence
Added git (GitHub, SSH, HTTPS) and HTTP URL marketplace source fetching
Added /marketplace TUI slash command with subcommands: add, remove, update, list, discover, install, uninstall, installed
Added /plugins TUI slash command to view all installed plugins (npm + marketplace) and enable/disable marketplace plugins

Changed

Changed marketplace clone promotion to occur after duplicate and drift checks, improving safety of concurrent marketplace operations

Removed

Removed grep.app code search provider support; code search now uses Exa exclusively
Removed providers.codeSearch setting and related configuration options

Fixed

Fixed git-subdir plugin source resolution to properly clean up temporary clone directories on path validation errors
Fixed LSP config loading to use correct filenames variable when scanning plugin roots
Fixed plugin selector UI to request render on cancel, preventing stale display state
Fixed marketplace install command error handling to display user-friendly error messages instead of crashing
Fixed MCP tools from newly added servers not being activated after /mcp add — refreshMCPTools preserves prior MCP tool selections, so brand-new servers had their tools registered in the registry but never passed to the agent; tools are now explicitly activated on successful connection
Fixed skill:// URI resolver to handle namespaced skills via longest-prefix matching against registered skill names

[13.16.5] - 2026-03-29

Fixed

Fixed --model provider/id resolving to wrong provider when model ID exists in multiple catalogs (#560)

[13.16.4] - 2026-03-28

Changed

Renamed hashline helper functions from hlineref/hlinefull to href/hline for brevity
Simplified hashline edit location API: replaced separate line and block properties with unified range property accepting { pos, end } for all range-based edits
Updated hashline prompt documentation to reflect new range syntax and clarified editing guidelines

Fixed

Added detection for kysely-codegen generated files in auto-generated file guard

[13.16.1] - 2026-03-27

Added

Added searchDb parameter to PromptActionAutocompleteProvider constructor for native search database integration in autocomplete workflows
Added searchDb parameter to enable native search database integration for grep and find operations
Exported SearchDb type from tools module for type-safe search database usage

Changed

Updated grep tool to accept and utilize searchDb parameter for improved search performance
Updated find tool to pass searchDb parameter to underlying search operations
Updated grep tool description to remove ripgrep-specific implementation detail

[13.16.0] - 2026-03-27

Added

Implemented root path alias: bare / in tool inputs now resolves to the session working directory instead of the filesystem root
Added browser.screenshotDir setting to configure screenshot save directory with path expansion

Changed

Improved hashline tool documentation with clearer guidance on block boundary handling and closing delimiter duplication prevention
Updated screenshot path resolution to use resolveToCwd for consistent workspace-relative path handling
Updated hook editor hint text to include ctrl+g external editor option when using prompt style
Refactored question result formatting to consistently include question ID in output

[13.15.3] - 2026-03-26

Added

Added configurable app.model.selectTemporary keybinding for temporary model selection.

[13.15.0] - 2026-03-23

Breaking Changes

Changed hashline edit schema from flat op/pos/end/lines fields to structured loc/content format with location-specific objects
Renamed hashline edit operations: replace_line → { line: anchor }, replace_range → { block: { pos, end } }, append_at → { append: anchor }, prepend_at → { prepend: anchor }, append_file → "append", prepend_file → "prepend"
Changed lines parameter to content in hashline edit entries
Renamed hashline edit operation types: append → append_at, prepend → prepend_at, append_eof → append_file, prepend_bof → prepend_file
Changed hashline edit operation types from replace (with optional end) to explicit replace_line and replace_range operations
Added required append_eof and prepend_bof operations for file-level edits; append and prepend now require an anchor position
Made pos parameter required for replace_line, append, and prepend operations; append_eof and prepend_bof no longer accept anchors

Added

Added prompt for tradeoff metrics during autoresearch setup to collect secondary metrics alongside primary metric
Added validation of contract path specifications to reject absolute paths and parent directory references
Added stricter benchmark command validation in isAutoresearchShCommand() to reject chained commands, pipes, and redirects
Added protection against prototype pollution in ASI data and metric cloning by filtering __proto__, constructor, and prototype keys
Added autoResumeArmed flag to track when autoresearch should automatically resume pending runs
Added lastAutoResumePendingRunNumber to prevent duplicate auto-resume prompts for the same pending run
Added git clean -X invocation during failed experiment rollback to remove ignored build artifacts
Added validation to reject init_experiment when a previous run is still pending and unlogged
Added autoresearch contract system for validating benchmark commands, metrics, scope paths, off-limits paths, and constraints with fingerprint tracking to detect configuration drift
Added autoresearch.program.md support for repo-local playbook overlays that guide session strategy while preserving autoresearch.md as source of truth
Added pending run artifact tracking and recovery to resume incomplete experiments from .autoresearch/runs/ directory with run numbers and benchmark logs
Added run directory organization with numbered run artifacts, benchmark logs, and optional checks logs for experiment traceability
Added segment fingerprinting to detect when benchmark configuration changes between runs and warn about potential incomparability
Added support for secondary metrics tracking alongside primary metric with configurable direction (lower/higher is better)
Added getCurrentAutoresearchBranch() helper to detect and validate existing autoresearch branches for session resumption
Added PendingRunSummary type to track unlogged run state including parsed metrics, ASI data, and pass/fail status
Added hidden next-turn message delivery via deliverAs: 'nextTurn' with optional triggerTurn to queue context for next LLM call without exposing in editable queue
Added #queueHiddenNextTurnMessage() and #promptQueuedHiddenNextTurnMessages() to AgentSession for autonomous tool reactions
Added resume context support in command-resume.md template for user-provided guidance when resuming sessions
Added current segment snapshot display in autoresearch prompt showing recent runs, baseline metrics, and best results
Added pending run indicator in autoresearch prompt to guide users to complete unlogged experiments before starting new benchmarks
Added local playbook section in autoresearch prompt when autoresearch.program.md exists
Added tab replacement in dashboard and tool output rendering to prevent display corruption from shell commands with tabs
Added boundary duplication warning when replace_range or replace_line operations include a last inserted line that matches the next surviving line, helping detect off-by-one range errors
Added git branch isolation for autoresearch sessions via ensureAutoresearchBranch() to safely revert failed experiments
Added branch status line to autoresearch initialization and resume prompts showing created or reused branch name
Added Files in Scope, Off Limits, and Constraints sections to autoresearch.md template for explicit scope definition
Added validation of ASI metadata requirements in log_experiment tool, requiring hypothesis for all runs and rollback context for failed runs
Added keybinding matcher utilities matchesAppInterrupt() and matchesSelectCancel() for consistent escape key handling across components
Added support for customizable app.interrupt and tui.select.cancel keybindings in interactive components
Added defaultInactive property to ToolDefinition to allow tools to be registered but excluded from the initial active set, with extension responsibility for activation/deactivation
Added dynamic tool activation/deactivation in autoresearch mode via setActiveTools() API
Added separate initialization and resume workflows for autoresearch with command-initialize.md and command-resume.md prompts
Added intent dialog to prompt users for autoresearch optimization goals when starting fresh
Added automatic detection of existing autoresearch.md to resume from previous sessions without re-prompting for intent
Added autoresearch extension with autonomous experiment loop capabilities
Added init_experiment tool to initialize and reset autoresearch sessions with configurable metrics
Added log_experiment tool to record experiment results with metric parsing and confidence tracking
Added run_experiment tool to execute commands and capture metrics with timeout and crash detection
Added autoresearch dashboard controller for displaying experiment results and optimization progress
Added support for secondary metrics tracking alongside primary metric
Added ExtensionWidgetContent and ExtensionUiComponentFactory types for flexible widget configuration
Added ExtensionWidgetOptions interface with placement parameter to position widgets above or below editor
Added WidgetPlacement type supporting 'aboveEditor' and 'belowEditor' placement options
Added hookWidgetContainerAbove and hookWidgetContainerBelow containers to InteractiveMode for separate widget management
Added autoresearch mode for autonomous experiment loops with init_experiment, log_experiment, and run_experiment tools
Added autoresearch dashboard widget displaying experiment results, metrics, and optimization progress
Added support for metric tracking with configurable direction (lower/higher is better) and secondary metrics
Added widget placement options to position extensions above or below the editor via placement parameter
Added ExtensionWidgetContent and ExtensionWidgetOptions types for flexible widget configuration
Added ACP (Agent Client Protocol) mode for headless agent operation via --mode acp
Added support for Agent Client Protocol SDK integration with session management, MCP server configuration, and streaming communication
Added ensureOnDisk() method to SessionManager to persist sessions immediately for ACP discovery

Changed

Changed autoresearch initialization to collect and validate benchmark command, metric definition, scope paths, off-limits list, and constraints before init_experiment
Changed init_experiment to require exact benchmark command, metric definition, scope, off-limits, and constraints matching collected contract
Changed log_experiment to record run number, benchmark command, scope paths, off-limits list, constraints, and segment fingerprint with each result
Changed run_experiment to organize output in numbered run directories with separate benchmark and checks logs for artifact preservation
Changed autoresearch dashboard to show pending run indicator when unlogged experiment exists
Changed autoresearch resume workflow to detect and offer recovery of pending run artifacts before continuing experiment loop
Changed ExperimentResult to include runNumber, benchmarkCommand, scopePaths, offLimits, constraints, and segmentFingerprint fields
Changed RunningExperiment to track runDirectory and runNumber for artifact organization
Changed AutoresearchRuntime to include lastRunArtifactDir, lastRunNumber, lastRunSummary, benchmarkCommand, secondaryMetrics, scopePaths, offLimits, constraints, and segmentFingerprint
Changed autoresearch prompts to emphasize autoresearch.md as source of truth for benchmark, scope, and constraints
Changed command-initialize.md to display collected setup (benchmark command, metric, direction, scope, off-limits, constraints) before initialization
Changed resume-message.md to reference pending run artifacts and guide completion of unlogged experiments
Changed sendMessage() API documentation to clarify deliverAs: 'nextTurn' behavior for hidden context delivery
Changed SendMessageHandler type documentation to explain hidden next-turn message queuing during prompt teardown
Changed autoresearch startup to create or reuse a dedicated autoresearch/... git branch before enabling the experiment loop
Changed autoresearch to refuse startup when unrelated worktree changes would make auto-reverts unsafe
Changed autoresearch prompts to emphasize scope and constraints as source of truth for session direction
Changed component escape key handling to use keybinding manager for app.interrupt and tui.select.cancel with fallback to raw Escape matching
Updated autoresearch prompt guidance to require explicit files in scope, off-limits paths, and session constraints
Changed autoresearch command to use intent-based initialization instead of goal parameter, with user input dialog for new sessions
Changed autoresearch startup to create or reuse a dedicated autoresearch/... git branch before enabling the experiment loop, and to refuse startup when unrelated worktree changes would make auto-reverts unsafe
Changed autoresearch startup to activate experiment tools (init_experiment, run_experiment, log_experiment) only when autoresearch mode is enabled
Changed autoresearch shutdown to deactivate experiment tools when mode is disabled or cleared
Changed autoresearch session rehydration to dynamically manage experiment tool activation based on session state
Changed autoresearch prompts and notes guidance to require explicit files in scope, off-limits paths, and session constraints
Refactored hashline edit validation to enforce stricter anchor requirements per operation type
Updated edit application logic to handle explicit file-level operations (append_eof, prepend_bof) separately from anchor-based operations
Changed setWidget API to accept ExtensionWidgetOptions parameter for placement control
Changed widget placement logic to manage widgets above and below editor separately
Changed hashline edit application to preserve duplicated boundary lines exactly as provided instead of auto-correcting them
Updated RPC mode to support widget placement option in setWidget requests
Changed hashline edit application to preserve duplicated boundary lines exactly as provided instead of auto-correcting them
Changed widget API to support placement options and component factories in addition to string arrays
Updated extension UI controller to manage widgets above and below the editor separately
Updated ask tool rendering to support markdown formatting in questions and option labels
Refactored hook input and selector components to render titles as markdown for richer text formatting
Changed session collection to include sessions with zero messages, enabling ACP mode to create discoverable sessions immediately
Changed session persistence logic to use atomic file rewrite when flushing unflushed sessions to prevent duplication
Removed hashline edit autocorrection for duplicated boundary lines; escaped-tab autocorrection remains available for leading \\t sequences

Removed

Removed command-start.md prompt template in favor of separate initialize and resume workflows
Removed auto-correction of off-by-one range edits that duplicated closing braces or boundary lines
Removed shouldAutocorrect function and related boundary line deduplication logic from hashline editor
Removed auto-correction of off-by-one range edits that duplicated closing braces or boundary lines

Fixed

Fixed autoresearch resume to detect and recover pending run artifacts that were left unlogged from previous sessions
Fixed dashboard overlay to display when running experiment even with zero completed results
Fixed tab character rendering in dashboard command display and tool output summaries
Fixed autoresearch logging to require durable ASI metadata (hypothesis, rollback_reason, next_action_hint) for every run including rollback context for discarded, crashed, and checks-failed experiments
Fixed autoresearch logging to require durable ASI metadata for every run, including rollback context for discarded, crashed, and checks-failed experiments

[13.14.0] - 2026-03-20

Added

Auto-reconnect MCP servers on connection loss with proactive SSE stream monitoring and retry backoff
Tool-level reconnect: retriable connection errors (ECONNREFUSED, ECONNRESET, stale session 404/502/503) trigger automatic reconnection and single retry
/mcp reconnect <name> command for manual server recovery after extended outages

Changed

Extended transport reconnect handling to all transport types (not just HTTP/SSE), ensuring stdio and other transports trigger automatic reconnection on connection loss
Improved reconnect robustness by aborting retry attempts when MCP server configuration changes during reconnection sequence
Updated explore agent thinking level from off to med for improved reasoning
Simplified explore agent output schema: consolidated file references into single ref field with optional line ranges instead of separate path, line_start, line_end fields
Removed code section from explore agent output (critical code excerpts no longer extracted)
Removed dependencies section from explore agent output
Removed risks section from explore agent output
Removed start_here section from explore agent output

Fixed

Fixed reconnect retry loop continuing after configuration changes by checking epoch before each reconnection attempt
roots/list timeout on MCP server initialization: connectToServer now always installs a default handler for ping and roots/list
Fixed resumed GitHub Copilot conversations that could fail with 401 input item does not belong to this connection on the first follow-up after process restart (#488)
Fixed STT Alt+H mic cursor rendering to measure the actual microphone glyph width, preventing one-column TUI overflow crashes when the active symbol preset uses a wide icon (#484)

[13.13.2] - 2026-03-18

Added

Added automatic stripping of hashline display prefixes (LINE#ID:) from write tool content when hashline edit mode is enabled, preventing the model from accidentally copying display markers into files
Added mcpServerName and mcpToolName optional properties to custom tools for MCP server discovery and search metadata

[13.13.1] - 2026-03-18

Added

Automatic deduplication of identical context files by content, keeping the closest (lowest depth) copy when duplicates are discovered

[13.13.0] - 2026-03-18

Added

Added edit.blockAutoGenerated setting to control whether auto-generated file detection is enforced (enabled by default)
Improved auto-generated file detection to use language-specific comment parsing instead of broad regex patterns, reducing false positives
Added auto-generated file detection to prevent accidental modification of generated code (protoc, sqlc, buf, swagger, etc.)
Added validation in Edit and Write tools to block modifications to files with auto-generated markers or naming patterns

Changed

Enhanced auto-generated marker detection to only scan leading header comments rather than entire file prefix, improving accuracy for files with generated markers in code

[13.12.10] - 2026-03-17

Added

Added args field to ShellResult to capture the executed command
Added exit_code property to ShellResult as an alias for returncode
Added check_returncode() method to ShellResult to raise CalledProcessError on non-zero exit codes

Changed

Renamed code field to returncode in ShellResult (accessible via code property for backward compatibility)
Updated run() command documentation to clarify available ShellResult fields

[13.12.9] - 2026-03-17

Added

Added /session delete command to delete current session with confirmation and return to session selector
Added session deletion in session selector via Delete key with confirmation dialog

Changed

Changed session deletion callback to return a boolean indicating success, allowing callers to distinguish between failed deletions and upstream cancellations

Fixed

Fixed OAuth redirect URI validation to preserve exact configured values without adding trailing slashes
Fixed session deletion error handling to display error messages in the session selector UI instead of silently failing
Added oauth.redirectUri, oauth.clientSecret, and oauth.callbackPath support for MCP server OAuth config so providers can use exact registered redirect URIs while preserving local callback listener settings (#445)

[13.12.8] - 2026-03-16

Breaking Changes

Changed SessionManager.create() to require explicit sessionDir parameter instead of optional—callers must now pass SessionManager.getDefaultSessionDir(cwd) to use default behavior
Changed SessionManager.continueRecent() to require explicit sessionDir parameter instead of optional—callers must now pass SessionManager.getDefaultSessionDir(cwd) to use default behavior
Changed SessionManager.forkFrom() to require explicit sessionDir parameter instead of optional—callers must now pass SessionManager.getDefaultSessionDir(cwd) to use default behavior
Changed SessionManager.list() signature to accept only sessionDir parameter instead of cwd and optional sessionDir—callers must now compute and pass the session directory explicitly

Added

Added SessionManager.getDefaultSessionDir() static method to explicitly resolve the canonical default session directory for a working directory
Added support for quoted paths in grep, ast_grep, and find tools to handle directory names with spaces
Added normalizePathLikeInput utility function to consistently handle quoted and whitespace-trimmed path inputs

Changed

Made sessionDir parameter optional in SessionManager.create(), SessionManager.continueRecent(), and SessionManager.forkFrom()—callers can now omit it to use the default session directory
Changed SessionManager.list() signature to accept cwd as the first parameter instead of requiring an explicit sessionDir—callers can now omit sessionDir to use the default for the given working directory
Updated SessionManager.getDefaultSessionDir() to accept optional agentDir parameter for computing session directories within a custom agent root
Improved status line path display to strip display roots using canonical path resolution, correctly handling symlink aliases to home and Projects directories
Improved error messaging in ast_grep when no matches are found with parse errors, now suggests narrowing path/glob or setting lang to resolve mis-scoped queries

Fixed

Fixed SDK-created default sessions to honor the configured agentDir for session storage, preventing tests from writing stray session directories into the real ~/.omp/agent/sessions root
Fixed session directory resolution to correctly handle symlink-equivalent paths, ensuring aliased home and temp directories resolve to the same session storage location as their real targets

[13.12.7] - 2026-03-16

Changed

Modified getSelectedMCPToolNames() to return only active MCP tools in non-discovery sessions, filtering by tool registry availability
Updated search_tool_bm25 tool instantiation to conditionally create the tool only when MCP discovery mode is enabled and execution hooks are available
Changed search results to exclude already-selected MCP tools before applying the limit parameter, allowing discovery of additional tools in subsequent searches

Fixed

Fixed MCP tool selection tracking to properly distinguish between discovery-enabled and non-discovery sessions, preventing orphaned tool selections after manual deactivation

[13.12.6] - 2026-03-15

Changed

Updated llama.cpp model discovery to read context window from the /props endpoint's default_generation_settings.n_ctx field instead of using hardcoded 128000 default
Updated llama.cpp model discovery to detect vision capabilities from the /props endpoint's modalities.vision field instead of defaulting to text-only input
Changed llama.cpp maxTokens calculation to respect discovered context window limits, capping at 8192 or the server's context window, whichever is smaller

Fixed

Fixed llama.cpp auto-discovery to read context window and vision support from the native /props endpoint instead of relying on hardcoded defaults

[13.12.5] - 2026-03-15

Added

Automatic discovery of Ollama model context window from model metadata, enabling accurate token limit configuration
Added attribution option to PromptOptions to explicitly control billing/initiator attribution for prompts
Added automatic clearing of completed and abandoned todo tasks after ~1 minute

Changed

Ollama model registration now uses discovered context window instead of hardcoded 128000 token default
Ollama model maxTokens now respects discovered context window constraints
Improved session directory migration to handle legacy absolute paths with double-dash format, automatically relocating them to new canonical locations
Enhanced session directory encoding to use -tmp- prefix for temporary directories instead of legacy double-dash format for better clarity
Updated SessionManager.create() to require both cwd and sessionDir parameters for explicit session directory control
Improved session directory naming for temporary working directories using -tmp- prefix instead of legacy -- format
Made cwd and sessionDir fields mutable in SessionManager to support session relocation without type casting
Changed subagent prompts to explicitly set attribution: "agent" for accurate billing attribution
Strip already-completed tasks when restoring session from branch history

Fixed

Fixed automatic migration of legacy session directories to new -tmp- prefixed naming scheme for temp-root sessions

[13.12.4] - 2026-03-15

Added

Exposed settings instance in CustomToolContext for session-specific configuration access

Changed

Improved artifact spill configuration to use session settings with schema defaults as fallback
Refactored type annotations for better type safety in tool result handling

[13.12.2] - 2026-03-15

Added

Added compaction.thresholdTokens setting as a fixed token limit alternative to percentage-based compaction threshold
Added more artifact spill threshold options (1 KB to 1 MB) with size descriptions
Added more artifact tail bytes and tail lines options with descriptions
Added toExtensionId capability method to enable granular disabling of individual capabilities by ID
Added support for disabling specific capabilities (skills, tools, hooks, rules, prompts, instructions, slash commands, MCP servers, extension modules, and context files) via disabledExtensions setting
Added includeDisabled and disabledExtensions options to LoadOptions for capability loading
Added plugin manifest support for extensions entry points to allow plugins to contribute extension modules
Added extensions field to plugin features for feature-specific extension entry points
Added automatic discovery of extension modules from installed plugins during extension loading
Added disabledExtensions setting to allow disabling specific extensions and skills by ID
Added support for filtering skills by disabled extension IDs with skill: prefix

Changed

Changed capability loading to filter out disabled items based on extension IDs before returning results
Changed plugin loader to support extensions as a manifest entry type alongside tools, hooks, and commands
Changed extension discovery to include extension entry points from all enabled plugins
Changed context file path handling to use path.basename() for consistent cross-platform filename extraction

Fixed

Fixed skill loading to properly respect disabled skill names when loading from custom directories

[13.12.1] - 2026-03-15

Added

Support for move-only operations that preserve exact bytes including binary files

Fixed

Fixed handling of file moves when no edits are specified, now correctly preserves binary content
Fixed validation to reject move operations where source and destination paths are identical

[13.12.0] - 2026-03-14

Added

Added per-rule TTSR interrupt mode override via interruptMode field in rule frontmatter to allow fine-grained control over when TTSR interrupts stream processing
Added task model role to allow configuring a dedicated model for subtask execution via modelRoles.task setting
Added moveCursorToMessageEnd and moveCursorToMessageStart prompt actions to navigate to the beginning and end of the entire message
Added support for provider-level compat configuration to apply OpenAI compatibility settings across all models from a provider
Added reasoningEffortMap configuration option to map reasoning effort levels to provider-specific values
Added support for supportsUsageInStreaming, requiresToolResultName, requiresAssistantAfterToolResult, requiresThinkingAsText, thinkingFormat, and supportsStrictMode OpenAI compatibility options
Added support for provider-configurable OpenAICompat.extraBody to inject request-body fields for custom gateway/proxy routing
Added close() method to SessionManager for properly closing persistent writers after flushing pending data
Added omp config init-xdg command to initialize XDG Base Directory structure on Linux
Added getHistoryDbPath(), getModelDbPath(), getMemoriesDir(), getTerminalSessionsDir() path helpers

Changed

Path resolution on Linux redirects to XDG locations when XDG_DATA_HOME / XDG_STATE_HOME / XDG_CACHE_HOME environment variables are set
Changed TTSR interrupt logic to respect per-rule interruptMode settings, falling back to global ttsr.interruptMode when rule-level override is not specified
Reorganized settings tabs from 12 tabs (display, agent, input, tools, config, services, bash, lsp, ttsr, status) to 8 focused tabs (appearance, model, interaction, context, editing, tools, tasks, providers) for improved discoverability
Consolidated status line settings into the Appearance tab instead of a separate Status tab
Reorganized sampling parameters (temperature, topP, topK, minP, presencePenalty, repetitionPenalty) into the Model tab
Moved edit tool settings (mode, fuzzyMatch, fuzzyThreshold, streamingAbort) to the Editing tab
Moved read tool settings (readLineNumbers, readHashLines, read.defaultLimit) to the Editing tab
Moved LSP settings (lsp.enabled, lsp.formatOnWrite, lsp.diagnosticsOnWrite, lsp.diagnosticsOnEdit) to the Editing tab
Moved bash interceptor settings to the Editing tab
Moved Python settings (python.toolMode, python.kernelMode, python.sharedGateway) to the Editing tab
Moved task delegation settings (task.isolation.*, task.eager, task.maxConcurrency, task.maxRecursionDepth) to the Tasks tab
Moved skill and command settings to the Tasks tab
Moved provider selection settings (providers.webSearch, providers.codeSearch, providers.image, etc.) to the Providers tab
Moved Exa settings to the Providers tab
Moved secret handling settings to the Providers tab
Moved speech-to-text settings to the Interaction tab
Moved context promotion, compaction, branch summary, memories, and TTSR settings to the Context tab
Updated tab icon symbols across unicode, nerd, and ASCII presets to match new tab structure
Changed default agent model from default to pi/task to enable independent model configuration for subtasks
Changed agent model resolution to support single-pattern inheritance fallback, allowing pi/task agents to inherit the active session model when the task role is unconfigured
Changed system prompt to use ISO 8601 date format (YYYY-MM-DD) instead of locale-specific formatting
Changed system prompt template to use {{date}} instead of {{dateTime}} for current date display
Changed tool download timeout from 15 seconds to 120 seconds to accommodate slower network conditions
Changed working directory paths in system prompt to use forward slashes for consistency across platforms
Modified bash executor to fall back to one-shot shell execution after a persistent session hard timeout, preventing subsequent commands from hanging

Removed

Removed bash executor hard timeout recovery test file (functionality already documented in existing entries)

Fixed

Fixed bash execution to fall back to one-shot shell runs after a persistent session hard timeout, preventing later commands from hanging until restart
Fixed timeout handling in RpcClient to properly clear timeouts and prevent resource leaks
Fixed AgentSession disposal to call SessionManager's close() method when available, ensuring proper cleanup of persistent writers
Removed redundant path.join() call wrapping getHistoryDbPath() in history-storage.ts

[13.11.1] - 2026-03-13

Added

Added llama.cpp as local provider
Added code_search tool supporting both Exa and grep.app providers for code snippet and documentation search
Added providers.codeSearch setting to configure code search provider (exa or grep)
Added grep.app integration for public code search with result ranking by context relevance

Changed

Updated compact diff preview to include line hashes for visibility and integrity verification of unchanged and added lines
Modified compact diff preview to track line number synchronization between old and new files when processing insertions and deletions
Simplified web search tools: removed web_search_deep, web_search_crawl, web_search_linkedin, and web_search_company tools
Removed exa.enableLinkedin and exa.enableCompany settings; LinkedIn and company research are no longer available
Refactored code search to use pluggable provider system instead of Exa-only implementation

Removed

Removed Exa LinkedIn search tool (exa_linkedin)
Removed Exa company research tool (exa_company)
Removed Exa deep search tool (exa_search_deep)
Removed Exa URL crawl tool (exa_crawl)

Fixed

Fixed line number parsing in compact diff preview to handle variable-width line number fields with leading whitespace

[13.11.0] - 2026-03-12

Added

Added Parallel as a web search provider with support for fast and research modes
Added Parallel extract API integration for URL content fetching and YouTube video extraction
Added providers.parallelFetch setting to enable/disable Parallel extract for URL fetching
Added /login parallel command support for Parallel API authentication
Added subcommands to /copy command: code (copy last code block), all (copy all code blocks), cmd (copy last bash/python command), and last (copy full message)
Added support for copying last executed bash or python command via /copy cmd subcommand
Added assignment field to task progress and result objects to track the raw per-task assignment text separately from the full templated task
Added details field to todo items for storing implementation specifics, file paths, and edge cases (shown only when task is active)
Added support for multi-line details in todo items with automatic indentation in interactive and reminder displays
Added todo.eager setting to automatically create a comprehensive todo list after the first user message
Added buildNamedToolChoice utility function to build provider-aware tool choice constraints for named tools
Support for comma/space-separated path lists in find, grep, ast_grep, and ast_edit tools (e.g., apps/,packages/,phases/ or apps/ packages/ phases/)
New resolveMultiSearchPath and resolveMultiFindPattern functions to handle multi-path search inputs with automatic common base path detection

Changed

Updated HTML-to-text rendering to prefer Parallel extract when credentials are available, before falling back to jina, trafilatura, or lynx
Updated YouTube scraper to prefer Parallel extract when credentials are available, before falling back to yt-dlp
Updated web search provider priority order to include Parallel between Exa and Kagi
Updated hashline tool documentation with explicit guidance on replace operation semantics, clarifying that lines must not extend past end to avoid unintended line duplication
Improved diagnostic message formatting to group errors by file path with indented details for better readability
Modified eager todo prelude to use hidden custom message type instead of visible developer message, preventing duplicate prompt text in session history
Updated eager todo prompt to remove dynamic user request injection, simplifying the template and preventing request repetition in displayed messages
Modified eager todo enforcement to prepend the todo reminder to the first user turn instead of executing it as a separate synthetic turn, reducing unnecessary prompt calls
Updated task rendering to display assignment text instead of full task template when available, reducing noise in progress and result displays
Modified task section rendering to show trimmed assignment text without stripping context blocks, simplifying the display logic
Updated todo item display to show details field indented below active tasks in both interactive mode and todo reminder component
Modified tool choice resolution to support per-turn tool choice overrides via consumeNextToolChoiceOverride()
Updated tool documentation to clarify that path parameter accepts files, directories, glob patterns, or comma/space-separated path lists
Refactored path resolution logic in find, grep, ast_grep, and ast_edit tools to use unified multi-path handling

Fixed

Fixed hashline line normalization to trim trailing whitespace and strip carriage returns instead of removing all whitespace, preserving intentional spacing in code
Fixed noop detection in hashline replace operations to check array length equality before comparing lines, preventing false noop classification when single-line replacements expand to multiple lines
Fixed path resolution to accept bare directory names without trailing slashes in comma/space-separated path lists (e.g., apps packages phases)
Per-role modelRoles thinking selectors now propagate through commit/title helper model selection, legacy commit analysis, and agentic commit sessions while preserving default thinking inheritance when no role override is configured

[13.10.1] - 2026-03-10

Added

Exported submitInteractiveInput() function for programmatic submission of user input in interactive mode
Added proactive OAuth token refresh for MCP server connections with 5-minute expiry buffer
Added reactive 401/403 retry with automatic token refresh on HTTP MCP transports
Added refreshMCPOAuthToken() for standard OAuth 2.0 refresh_token grants
Persisted tokenUrl, clientId, and clientSecret in MCP auth config for cross-session token refresh

[13.10.0] - 2026-03-10

Fixed

Preserved text signature metadata (id and phase) when building OpenAI native history during session compaction

[13.9.16] - 2026-03-10

Breaking Changes

Web search tool no longer accepts provider parameter in tool calls; use internal provider resolution instead
Removed no_fallback option from search parameters

Added

Added before_provider_request extension event to intercept and modify provider request payloads before sending
Added emitBeforeProviderRequest() method to ExtensionRunner for chaining payload transformations across extensions
Added refreshInBackground() method to ModelRegistry for non-blocking model discovery
Added refreshProvider() method to refresh models for a specific provider on demand
Added getDiscoverableProviders() method to list all configured discoverable providers
Added getProviderDiscoveryState() method to inspect provider discovery status, cache age, and errors
Added provider discovery state tracking with status indicators (idle, ok, cached, unavailable, unauthenticated)
Added model caching with 24-hour TTL to preserve discovered models across sessions
Added provider-specific empty state messages in model selector showing cache age and discovery status
Added live provider refresh when switching provider tabs in model selector

Changed

Changed model discovery to load cached models immediately before attempting live refresh, improving startup performance
Changed model selector to refresh offline by default when reloading config, deferring live discovery to background
Changed model discovery timeout from 3000ms to 250ms for faster failure detection
Changed model discovery error handling to preserve cached models when live refresh fails
Changed refresh() strategy parameter to support 'offline' mode for config-only reloads
Changed main.ts to defer model refresh until needed (--list-models or background refresh)
Changed SDK session creation to use background refresh instead of blocking on model discovery
Removed provider parameter from web search tool schema; provider selection now handled internally
Removed no_fallback parameter from web search parameters; fallback behavior now automatic based on provider availability
Renamed SearchParams type to SearchToolParams for tool execution; introduced SearchQueryParams for CLI queries with optional provider selection

Fixed

Fixed model discovery to continue using cached models when provider is temporarily unavailable
Fixed unauthenticated provider discovery to preserve cached models instead of discarding them
Fixed model selector to show discovery status messages when provider has no models

[13.9.15] - 2026-03-10

Added

Added ensureLoadingAnimation() method to manage loading animation lifecycle and prevent duplicate spinners

Changed

Refactored loading animation initialization to use centralized ensureLoadingAnimation() method in event and input controllers
Updated showError() to properly clean up loading animation state when errors occur

[13.9.12] - 2026-03-09

Added

Added Tavily as a supported web search provider with TAVILY_API_KEY credential discovery and provider fallback support
Added #-triggered prompt action suggestions in the editor, with keybinding hints for line navigation and prompt copy actions
Added Tavily as a supported web search provider with TAVILY_API_KEY credential discovery and provider fallback support (#313)

Removed

Removed Kagi Universal Summarizer integration from fetch tool—HTML rendering now uses jina, trafilatura, and lynx only
Removed fetch.useKagiSummarizer setting
Removed Kagi summarization from YouTube video handling

Fixed

Canonicalized bash executor working directories before handing them to brush so pwd stays aligned with canonical Git worktree paths in symlinked workspaces

[13.9.10] - 2026-03-08

Added

Added env parameter to bash tool to pass environment variables safely without shell re-parsing, preventing quote and special character bugs with multiline or untrusted values
Added support for rendering partial env assignments in command preview while tool arguments are still streaming
Added env support to the bash tool so commands can reference safe shell variables without inline quoting bugs for multiline or quote-heavy values

Changed

Changed bash tool to display environment variable assignments in command preview when env parameter is used

[13.9.8] - 2026-03-08

Added

Added docs.rs scraper for extracting Rust crate documentation from rustdoc JSON, including support for modules, functions, structs, traits, enums, and other Rust items with caching

[13.9.7] - 2026-03-08

Added

Added skipPostPromptRecoveryWait option to handoff operations to defer recovery work until after handoff completion
Added deferred auto-compaction scheduling to allow threshold-triggered handoffs to complete while the original prompt is still unwinding

Changed

Extracted handoff document template to dedicated prompt file for improved maintainability and template variable support
Changed handoff prompt generation to use template rendering with support for custom focus instructions
Refactored internal prompt-in-flight tracking from boolean flag to counter to properly handle nested prompt operations
Moved llms.txt endpoint discovery to fallback strategy when rendered page content is low quality, prioritizing page-specific content over site-wide files
Enhanced llms.txt endpoint detection to scope candidates to the requested URL path, searching section-specific files before site-wide ones

[13.9.6] - 2026-03-08

Added

Added glob parameter to ast_grep and ast_edit tools for additional glob filtering relative to the path parameter
Added combineSearchGlobs utility function to merge glob patterns from path and glob parameters

Changed

Renamed patterns parameter to pat in ast_grep tool for consistency
Renamed selector parameter to sel in ast_grep and ast_edit tools for brevity
Updated tool documentation with expanded guidance on AST pattern syntax, metavariable usage, and contextual matching strategies
Updated grep tool to combine glob patterns from path and glob parameters instead of throwing an error when both are provided

[13.9.4] - 2026-03-07

Added

Automatic detection of Ollama model capabilities including reasoning/thinking support and vision input via the /api/show endpoint
Improved Kagi API error handling with extraction of detailed error messages from JSON and plain text responses

Changed

Updated Kagi provider description to clarify requirement for Kagi Search API beta access

[13.9.3] - 2026-03-07

Breaking Changes

Changed ThinkingLevel type to be imported from @oh-my-pi/pi-agent-core instead of @oh-my-pi/pi-ai
Changed thinking level representation from string literals to Effort enum values (e.g., Effort.High instead of "high")
Changed getThinkingLevel() return type to ThinkingLevel | undefined to support models without thinking support
Changed model reasoning property to thinking property with ThinkingConfig for explicit effort level configuration
Changed thinkingLevel in session context to be optional (ThinkingLevel | undefined) instead of always present

Added

Added thinking.ts module with getThinkingLevelMetadata() and resolveThinkingLevelForModel() utilities for thinking level handling
Added ThinkingConfig support to model definitions for specifying supported thinking effort levels per model
Added enrichModelThinking() function to apply thinking configuration to models during registry initialization
Added clampThinkingLevelForModel() function to constrain thinking levels to model-supported ranges
Added getSupportedEfforts() function to retrieve available thinking efforts for a model
Added Effort enum import from @oh-my-pi/pi-ai for type-safe thinking level representation
Added /fast slash command to toggle OpenAI service tier priority mode for faster response processing
Added serviceTier setting to control OpenAI processing priority (none, auto, default, flex, scale, priority)
Added compaction.remoteEnabled setting to control use of remote compaction endpoints
Added remote compaction support for OpenAI and OpenAI Codex models with encrypted reasoning preservation
Added fast mode indicator (⚡) to model segment in status line when priority service tier is active
Added context usage threshold levels (normal, warning, purple, error) with token-aware thresholds for better context awareness
Added isFastModeEnabled(), setFastMode(), and toggleFastMode() methods to AgentSession for fast mode control

Changed

Changed credential deletion to disable credentials with persisted cause instead of permanent deletion
Added disabledCause parameter to credential deletion methods to track reason for disabling
Changed thinking level parsing to use parseEffort() from local thinking module instead of parseThinkingLevel() from pi-ai
Changed model list display to show supported thinking efforts (e.g., "low,medium,high") instead of yes/no reasoning indicator
Changed footer and status line to check model.thinking instead of model.reasoning for thinking level display
Changed thinking selector to work with Effort type instead of ThinkingLevel for available levels
Changed model resolver to return undefined for thinking level instead of "off" when no thinking is specified
Changed compaction reasoning parameters to use Effort enum values instead of string literals
Changed RPC types to use Effort for cycling thinking levels and ThinkingLevel | undefined for session state
Changed theme thinking border color function to accept both ThinkingLevel and Effort types
Changed context usage coloring in footer and status line to use token-aware thresholds instead of fixed percentages
Changed compaction to preserve OpenAI remote compaction state and encrypted reasoning across sessions
Changed compaction to skip emitting kept messages when using OpenAI remote compaction with preserved history
Changed session context to include serviceTier field for tracking active service tier across session branches
Changed compact() function to accept remoteInstructions option for custom remote compaction prompts
Changed model registry to apply hardcoded policies (gpt-5.4 context window) consistently across all model loading paths

Fixed

Fixed OpenAI remote compaction to correctly append incremental responses instead of replacing entire history
Fixed thinking level display logic in main.ts to correctly check for undefined instead of "off"
Fixed model registry to preserve explicit thinking configuration on runtime-registered models
Fixed usage limit reset time calculation to use absolute resetsAt timestamps instead of deprecated resetInMs field
Fixed compaction summary message creation to no longer be automatically added to chat during compaction (now handled by session manager)

[13.9.2] - 2026-03-05

Added

Support for Python code execution messages with output display and error handling
Support for mode change entries in session exports
Support for TTSR injection and session initialization entries in tree filtering

Changed

Updated label lookup to use targetId field instead of parentId for label references
Changed model change entry display to use model field instead of separate provider and modelId fields
Simplified model change rendering by removing OpenAI Codex bridge prompt display
Updated searchable text extraction to include Python code from pythonExecution messages

Removed

Removed codexInjectionInfo from session data destructuring
Removed OpenAI Codex-specific bridge prompt UI from model change entries

Fixed

Auto-corrected off-by-one range start errors in hashline edits that would duplicate preceding lines

[13.9.0] - 2026-03-05

Added

Added read.defaultLimit setting to configure default number of lines returned by read tool when no limit is specified (default: 300 lines)
Added preset options for read default limit (200, 300, 500, 1000, 5000 lines) in settings UI

Changed

Updated read tool prompt to distinguish between default limit and maximum limit per call
Moved ThinkingLevel type from @oh-my-pi/pi-agent-core to @oh-my-pi/pi-ai for centralized thinking level definitions
Replaced local thinking level validation with parseThinkingLevel() and ALL_THINKING_LEVELS from @oh-my-pi/pi-ai
Updated thinking level option providers to use THINKING_MODE_DESCRIPTIONS from @oh-my-pi/pi-ai for consistent descriptions
Renamed RoleThinkingMode type to ThinkingMode and changed default value from 'default' to 'inherit' for clarity
Replaced formatThinkingEffortLabel() utility with formatThinking() from @oh-my-pi/pi-ai
Renamed extractExplicitThinkingLevel() to extractExplicitThinkingSelector() in model resolver
Updated thinking level clamping to use getAvailableThinkingLevel() from @oh-my-pi/pi-ai

Removed

Removed thinking-effort-label.ts utility file (functionality moved to @oh-my-pi/pi-ai)
Removed local VALID_THINKING_LEVELS constant definitions across multiple files
Removed isValidThinkingLevel() function (replaced by parseThinkingLevel() from @oh-my-pi/pi-ai)
Removed parseThinkingLevel() helper from discovery module (now uses centralized version from @oh-my-pi/pi-ai)

Fixed

Fixed provider session state not being cleared when branching or navigating tree history, preventing resource leaks with codex provider sessions

[13.8.0] - 2026-03-04

Added

Added buildCompactHashlineDiffPreview() function to generate compact diff previews for model-visible tool responses, collapsing long unchanged runs and consecutive additions/removals to show edit shape without full file content
Added project-level discovery for .agent/ and .agents/ directories, walking up from cwd to repo root (matching behavior of other providers like .omp, .claude, .codex). Applies to skills, rules, prompts, commands, context files (AGENTS.md), and system prompts (SYSTEM.md)

Changed

Changed edit tool response to include diff summary with line counts (+added -removed) and a compact diff preview instead of warnings-only output
Limited auto context promotion to models with explicit contextPromotionTarget; models without a configured target now compact on overflow instead of switching to arbitrary larger models (#282)

Fixed

Fixed :thinking suffix in modelRoles config values silently breaking model resolution (e.g., slow: anthropic/claude-opus-4-6:high) and being stripped on Ctrl+P role cycling

[13.7.6] - 2026-03-04

Added

Exported dedupeParseErrors utility function to deduplicate parse error messages while preserving order

Fixed

Reduced duplicate parse error messages when multiple patterns fail on the same file
Normalized parse error output in ast-grep to remove pattern-specific prefixes and show only file-level errors

[13.7.4] - 2026-03-04

Added

Added fetch.useKagiSummarizer setting to toggle Kagi Universal Summarizer usage in the fetch tool.

Fixed

Fixed incorrect message history reference in session title generation that could cause missing or stale titles on first message
Added startup check requiring Bun 1.3.7+ for JSONL session parsing (Bun.JSONL.parseChunk) and clear upgrade guidance so /resume and --resume do not silently report missing sessions on older Bun runtimes

[13.7.3] - 2026-03-04

Added

Added Kagi Universal Summarizer integration for URL summarization, now prioritized before Jina and other methods
Added Kagi Universal Summarizer support for YouTube video summaries when credentials are available
Exported searchWithKagi and summarizeUrlWithKagi functions from new web/kagi module for direct API access
Added KagiApiError exception class for Kagi API-specific error handling

Changed

Updated hashline prompt documentation with clearer operation syntax and improved examples showing full edit structure with path and edits array
Refactored hlineref Handlebars helper to return JSON-quoted strings for safer embedding in JSON blocks within prompts
Improved hashlineParseText to correctly preserve blank lines and trailing empty strings in array input while stripping trailing newlines from string input
Optimized duplicate line detection in range replacements to use trimmed comparison, reducing false positives from whitespace differences
Refactored Kagi search provider to use shared Kagi API utilities from web/kagi module
Changed HTML-to-text rendering priority order to try Kagi first, then Jina, Trafilatura, and Lynx

Fixed

Fixed isEscapedTabAutocorrectEnabled environment variable parsing to use switch statement for clearer logic and consistent default behavior

[13.7.2] - 2026-03-04

Added

Added support for direct OAuth provider login via /login <provider> command (e.g., /login kagi)
Added optional providerId parameter to showOAuthSelector() to enable direct provider selection without UI selector

Changed

Simplified web search result formatting to omit empty sections and metadata when not present

[13.7.0] - 2026-03-03

Fixed

Fixed ask timeout handling to auto-select the recommended option instead of aborting the turn, while preserving explicit user-cancel abort behavior (#266)

[13.6.2] - 2026-03-03

Fixed

Fixed LM Studio API key retrieval to use configured provider name instead of hardcoded 'lm-studio'
Fixed resource content handling to properly check for empty text values (null/undefined)
Fixed resource refresh tracking to prevent stale promise reuse when server connection changes
Fixed update target resolution to properly handle cases where binary path cannot be resolved

[13.6.1] - 2026-03-03

Fixed

Fixed omp update silently succeeding without actually updating the binary when the update channel (bun global vs compiled binary) doesn't match the installation method (#247)
Added post-update verification that checks the resolved omp binary reports the expected version, with actionable warnings on mismatch
omp update now detects when the omp in PATH is not managed by bun and falls back to binary replacement instead of updating the wrong location

[13.6.0] - 2026-03-03

Added

Added mcp:// internal URL protocol for reading MCP server resources directly via the read tool (e.g., read(path="mcp://resource-uri"))
Added LM Studio integration to the model registry and discovery flow.
Added support for authenticating with LM Studio using the /login lm-studio command.
Added fuse-projfs task isolation mode for Windows ProjFS-backed overlays.
Added /mcp registry search <keyword> integration with Smithery, including interactive result selection, editable server naming before deploy, Smithery configSchema prompts, and immediate runtime reload so selected MCP tools are available without restarting
Added OAuth failure fallback in /mcp registry search deploy flow to prompt for manual bearer tokens and validate them before saving configuration
Added Smithery auth support for /mcp registry search with cached API key login (/mcp registry login, /mcp registry logout) and automatic login prompt/retry on auth or rate-limit responses

Changed

Updated MCP resource update notifications to recommend using read(path="mcp://<uri>") instead of the deprecated read_resource tool
Updated Anthropic Foundry environment variable documentation and CLI help text to the canonical names: CLAUDE_CODE_USE_FOUNDRY, CLAUDE_CODE_CLIENT_CERT, and CLAUDE_CODE_CLIENT_KEY
Documented Foundry-specific Anthropic runtime configuration (FOUNDRY_BASE_URL, ANTHROPIC_FOUNDRY_API_KEY, ANTHROPIC_CUSTOM_HEADERS, NODE_EXTRA_CA_CERTS) in environment variable reference docs
fuse-overlay task isolation now targets fuse-overlayfs on Unix hosts only; on Windows it falls back to worktree with a <system-notification> suggesting fuse-projfs.
fuse-projfs now performs Windows ProjFS preflight checks and falls back to worktree when host or repository prerequisites are unavailable.
Cross-repo patch capture now uses the platform null device (NUL on Windows, /dev/null elsewhere) for git diff --no-index.

Removed

Removed read_resource tool; MCP resource reading is now integrated into the read tool via mcp:// URLs

Fixed

Fixed MCP resource subscription handling to prevent unsubscribing when notifications are re-enabled after being disabled
Fixed LM Studio base URL validation to preserve invalid configured URLs instead of silently falling back to localhost
Fixed URI template matching to correctly handle expressions that expand to empty strings

[13.5.6] - 2026-03-01

Changed

Updated OAuth client name from 'oh-my-pi MCP' to 'Codex' for dynamic client registration

Fixed

Fixed exit_plan_mode handler to abort active agent turn before opening plan approval selector, ensuring proper session cleanup

[13.5.5] - 2026-03-01

Added

Added Kagi web search provider (Search API v0) with related searches support and automatic KAGI_API_KEY detection

[13.5.4] - 2026-03-01

Added

Added authServerUrl field to AuthDetectionResult to capture OAuth server metadata from Mcp-Auth-Server headers
Added extractMcpAuthServerUrl() function to parse and validate Mcp-Auth-Server URLs from error messages
Added support for /.well-known/oauth-protected-resource discovery endpoint to resolve authorization servers
Added recursive auth server discovery to follow authorization_servers references when discovering OAuth endpoints
Added omp agents unpack CLI subcommand to export bundled subagent definitions to ~/.omp/agent/agents by default, with --project support for ./.omp/agents

Changed

Enhanced discoverOAuthEndpoints() to accept optional authServerUrl parameter and query both auth server and resource server for OAuth metadata
Improved OAuth metadata extraction to handle additional field name variations (clientId, default_client_id, public_client_id)
Refactored OAuth endpoint discovery logic into reusable findEndpoints() helper for consistent metadata parsing across multiple sources
Task subagents now strip inherited AGENTS.md context files and the task tool prompt no longer warns against repeating AGENTS guidance, aligning subagent context with explicit task inputs (#233)

Fixed

Fixed MCP OAuth discovery to honor Mcp-Auth-Server metadata and resolve authorization endpoints from the declared auth server, restoring Figma MCP login URLs with client_id (#235)

[13.5.3] - 2026-03-01

Added

Auto-include ast_grep and ast_edit tools when their text-based counterparts (grep, edit) are requested and the AST tools are enabled
Enforced tool decision in plan mode—agent now requires calling either ask or exit_plan_mode when a turn ends without a required tool call
Auto-correction of escaped tab indentation in edits (enabled by default, controllable via PI_HASHLINE_AUTOCORRECT_ESCAPED_TABS environment variable)
Warning when suspicious Unicode escape placeholder \uDDDD is detected in edit content

Changed

Updated bash tool description to conditionally show ast_grep and ast_edit guidance based on tool availability in the session
Replaced timeout-based cancellation with AbortSignal-based cancellation in the ask tool for more reliable user interaction handling
Updated ask tool to distinguish between user-initiated cancellation and timeout-driven auto-selection, with only user cancellation aborting the turn
Updated hashline documentation to clarify that \t in JSON represents a real tab character, not a literal backslash-t sequence

Fixed

Fixed race condition in dialog overlay handling where multiple concurrent resolutions could occur
Cancelling the ask tool now aborts the current turn instead of returning a normal cancelled selection, while timeout-driven auto-cancel still returns without aborting

[13.5.2] - 2026-03-01

Added

Added checkpoint tool to create context checkpoints before exploratory work, allowing you to investigate with many intermediate tool calls and minimize context cost afterward
Added rewind tool to end an active checkpoint and replace intermediate exploration messages with a concise investigation report
Added checkpoint.enabled setting to control availability of the checkpoint and rewind tools
Added render_mermaid tool to convert Mermaid graph source into ASCII diagram output
Added renderMermaid.enabled setting to control availability of the render_mermaid tool

Changed

Changed Mermaid rendering from PNG images to ASCII diagrams in theme rendering
Changed prerenderMermaid() function to synchronously render ASCII instead of asynchronously rendering PNG

[13.5.0] - 2026-03-01

Added

Added hlinejsonref Handlebars helper for embedding hashline references inside JSON blocks in prompts
Added librarian agent for researching external libraries and APIs by reading source code
Added oracle agent for deep reasoning on debugging, architecture decisions, and technical advice
Added dependencies and risks output fields to explore agent for better context handoff
Added support for lsp, fetch, web_search, and ast_grep tools to explore, plan, and reviewer agents

Changed

Enhanced hashline tool documentation with explicit prohibition on formatting-only edits
Added mandatory rule requiring indentation in lines to match surrounding context exactly from read output
Changed explore agent output field query to summary with expanded description for findings and conclusions

[13.4.1] - 2026-03-01

Fixed

Pending resolve reminders now trigger as soon as a preview action is queued, before the next assistant turn, with regression coverage in agent-session-resolve-reminder tests

[13.4.0] - 2026-03-01

Breaking Changes

ast_grep parameter pattern (string) replaced by patterns (string[])
ast_edit parameters pattern + rewrite replaced by ops: Array<{ pat: string; out: string }>

Added

Added resolve tool to apply or discard pending preview actions with required reasoning
AST edit now registers pending actions after preview, allowing explicit apply/discard workflow via resolve tool
Custom tools can register pending actions via pushPendingAction(action) in CustomToolAPI, enabling the resolve workflow for custom preview-apply flows
deferrable?: boolean field added to AgentTool, CustomTool, and ToolDefinition interfaces; tools that set it signal they may stage pending actions
HIDDEN_TOOLS and ResolveTool exported from @oh-my-pi/pi-coding-agent SDK for manual tool composition
PendingActionStore now uses a LIFO stack (push/peek/pop); multiple deferrable tools can stage actions that resolve in reverse order of registration
Added gemini, codex, and synthetic as supported values for the providers.webSearch setting
ast_grep tool now accepts a patterns array (replaces single pattern); multiple patterns run in one native pass and results are merged before offset/limit
ast_edit tool now accepts an ops array of { pat, out } entries (replaces pattern + rewrite); duplicate patterns are rejected upfront
AST find output now uses >> prefix on match-start lines and pads line numbers; directory-tree grouping with # dir / ## └─ file headers for directory-scoped searches
AST replace output now renders diff-style (-before / +after) change previews grouped by directory
Both AST tools now report scopePath, files, and per-file match/replacement counts in tool details
Task item id max length raised from 32 to 48 characters
Anthropic web search provider now uses buildAnthropicSearchHeaders (dedicated search header builder separate from inference headers)
Gemini web search provider: endpoint fallback (daily → sandbox) with retry on 429/5xx
Gemini web search now injects Antigravity system instruction and aligned request metadata (requestType, userAgent, requestId) for Antigravity credentials
buildGeminiRequestTools() helper for composable Gemini tool configuration (googleSearch, codeExecution, urlContext)
Web search schema exposes max_tokens, temperature, and num_search_results as tool parameters
Web search provider fallback: when an explicit provider is unavailable, resolves the auto chain instead of returning empty results

Changed

Simplified resolve tool output rendering to use inline highlighted format instead of boxed layout
Updated resolve tool to parse source tool name from label using colon separator for cleaner display
resolve tool is now conditionally injected: included only when at least one active tool has deferrable: true (previously always included)
discoverAndLoadCustomTools / loadCustomTools accept an optional pendingActionStore parameter to wire pushPendingAction for custom tools
AST edit tool no longer accepts preview parameter; all AST edit calls now return previews by default
AST edit workflow changed: preview is always shown, then use resolve tool to apply or discard changes
Agent now suggests calling resolve tool after AST edit preview with system reminder
ast_grep: include_meta parameter removed; metavariable captures are now always included in output
ast_edit: dry_run renamed to preview; max_files removed from schema and capped globally via $PI_MAX_AST_FILES (default 1000); max_replacements renamed to limit
ast_grep and ast_edit: parse errors in tool output are now capped at PARSE_ERRORS_LIMIT (20); excess errors are summarised as N / total parse issues rather than flooding the context
Updated ast_grep and ast_edit tool prompt examples to use concise, idiomatic patterns

Removed

Removed normativeRewrite setting that rewrote tool call arguments to normalized format in session history
Removed buildNormativeUpdateInput() helper and normative patch transformation logic

Fixed

ast_edit no longer rejects empty out values; an empty string now deletes matched nodes
ast_edit no longer trims pat and out values, preserving intentional whitespace
gemini_image tool: corrected responseModalities values from 'Image'/'Text' to uppercase 'IMAGE'/'TEXT' matching the API enum

[13.3.14] - 2026-02-28

Added

Expanded AST tool language support from 7 to all 25 ast-grep tree-sitter languages (Bash, C, C++, C#, CSS, Elixir, Go, Haskell, HCL, HTML, Java, JavaScript, JSON, Kotlin, Lua, Nix, PHP, Python, Ruby, Rust, Scala, Solidity, Swift, TSX, TypeScript, YAML)
AST find now emits all lines of multiline matches with hashline tags (LINE#HASH:content) consistent with read/grep output
Added AST pattern syntax reference (metavariables, wildcards, variadics) to system prompt
Added examples and scoping guidance to ast-find and ast-replace tool prompts
Added provider-schema-compatibility.test.ts: integration test that instantiates every builtin and hidden tool, runs their parameter schemas through adaptSchemaForStrict, sanitizeSchemaForGoogle, and prepareSchemaForCCA, and asserts zero violations against each provider's compatibility rules

Fixed

Non-code files (.md, .zip, .bin, .gitignore, etc.) are now silently skipped by AST tools instead of producing misleading parse errors
Fixed grep path wildcard handling so file patterns passed via path (for example schema-review-*.test.ts) are resolved as glob filters instead of failing path existence checks

[13.3.11] - 2026-02-28

Fixed

Restored inline rendering for read tool image results in assistant transcript components, including streaming and rebuilt session history paths.
Fixed shell-escaped read paths (for example, pasted \ -escaped screenshot filenames) by resolving unescaped fallback candidates before macOS filename normalization variants.

[13.3.8] - 2026-02-28

Added

Added ast_find tool for structural code search using AST matching via ast-grep, enabling syntax-aware pattern discovery across codebases
Added ast_replace tool for structural AST-aware rewrites via ast-grep, enabling safe syntax-level codemods without text-based fragility
Added astFind.enabled and astReplace.enabled settings to control availability of AST tools
Added system prompt guidance to prefer AST tools over bash text manipulation (grep/sed/awk/perl) for syntax-aware operations
Extracted prompt formatting logic into reusable formatPromptContent() utility with configurable render phases and formatting options
Added type_definition action to navigate to symbol type definitions with source context
Added implementation action to find concrete implementations of symbols with source context
Added code_actions action to list and apply language server code fixes, refactors, and import suggestions
Added symbol parameter to automatically resolve column position by searching for substring on target line
Added occurrence parameter to disambiguate repeated symbol matches on the same line
Added source code context display (3 lines) for definition, type definition, and implementation results
Added context display for first 50 references with remaining references shown location-only to balance detail and performance
Added support for glob patterns in file parameter for diagnostics action (e.g., src/**/*.ts)
Added waitForIdle() method to ensure prompt completion waits for all deferred recovery work (TTSR continuations, context promotions, compaction retries) to fully settle
Added getLastAssistantMessage() method to retrieve the most recent assistant message from session state without manual array indexing
Implemented TTSR resume gate to ensure prompt() blocks until TTSR interrupt continuations complete, preventing race conditions between TTSR injections and subsequent prompts
Added tools.maxTimeout setting to enforce a global timeout ceiling across all tool calls

Changed

Replaced globSync from glob package with native Bun.Glob API for glob pattern matching
Replaced fileTypeFromBuffer from file-type package with inline MIME type detection for JPEG, PNG, GIF, and WebP formats
Reduced MIME type sniffing buffer size from 4100 bytes to 12 bytes for improved performance
Changed mermaid cache key type from string to bigint for more efficient hashing
Replaced smol-toml dependency with native Bun.TOML.parse() for TOML parsing, reducing external dependencies
Replaced node-html-parser dependency with linkedom for HTML parsing, improving performance and reducing bundle size
Updated HTML parsing API calls from node-html-parser to linkedom across all web scrapers (arXiv, IACR, Go pkg, Read the Docs, Twitter, Wikipedia)
Changed element text extraction from .text property to .textContent property for compatibility with linkedom DOM API
Optimized document link extraction to use regex-based parsing with deduplication and a 20-link limit instead of full DOM traversal
Unified path parameter in ast_find and ast_replace tools to accept files, directories, or glob patterns directly, eliminating the separate glob parameter
Removed strictness parameter from ast_find and ast_replace tools
Removed fail_on_parse_error parameter from ast_replace tool (now always false)
Updated ast_find and ast_replace prompt guidance to clarify that path accepts glob patterns and no longer requires separate glob specification
Refactored prompt template rendering to use unified formatPromptContent() function with phase-aware formatting (pre-render vs post-render)
Updated format-prompts.ts script to use centralized prompt formatting utility instead of inline implementation
Replaced column parameter with symbol parameter for more intuitive position specification
Removed files parameter; use glob patterns in file parameter instead
Removed end_line and end_character parameters; range operations now use single position
Changed include_declaration parameter to always be true for references (removed from API)
Updated LSP client capabilities to advertise support for typeDefinition and implementation requests
Improved definition results to include source context alongside location information
Refactored deferred continuation scheduling to use centralized post-prompt task tracking instead of raw setTimeout() calls, improving reliability of concurrent recovery operations
Updated subagent executor to explicitly await waitForIdle() after each prompt and reminder, ensuring terminal assistant state is determined only after all background work completes
Replaced #waitForRetry() with #waitForPostPromptRecovery() to handle both retry and TTSR resume gates, ensuring prompt completion waits for all post-prompt recovery operations
Introduced structured post-prompt recovery task tracking in AgentSession and added explicit session completion APIs (waitForIdle(), getLastAssistantMessage()) for callers that need deterministic turn finalization
Updated intent field parameter name from agent__intent to _i for cleaner tool call contracts
Refined intent parameter guidance to require concise 2-6 word sentences in present participle form
Centralized per-tool timeout constants and clamping into tool-timeouts.ts

Removed

Removed file-type dependency, reducing external dependencies
Removed glob dependency in favor of native Bun.Glob API
Removed ignore dependency and ignore file handling utilities
Removed marked dependency
Removed zod dependency
Removed ms and @types/ms dev dependencies
Removed rootDir and ignoreMatcher parameters from loadFilesFromDir() (kept for API compatibility)
Removed smol-toml dependency from package.json
Removed node-html-parser dependency from package.json
Removed files array parameter for batch file operations
Removed column, end_line, and end_character parameters in favor of symbol-based positioning
Removed include_declaration parameter from references action

Fixed

Fixed TTSR violations during subagent execution aborting the entire subagent run; #waitForPostPromptRecovery() now also awaits agent idle after TTSR/retry gates resolve, preventing prompt() from returning while a fire-and-forget agent.continue() is still streaming
Fixed deferred TTSR/context-promotion continuations still racing prompt() completion by tracking compaction checks and deferred agent.continue() tasks under a shared post-prompt recovery orchestrator
Fixed subagent reminder/finalization sequencing to await session-level idle recovery between prompts before determining terminal assistant stop state
Fixed code_actions apply mode to execute command-based actions via workspace/executeCommand
Fixed diagnostics glob detection to recognize bracket character class patterns (e.g., src/[ab].ts)
Fixed LSP render metadata sanitization for symbol values to prevent tab/newline layout breakage
Fixed LSP diagnostics glob requests that appeared stuck by capping glob expansion and shortening per-file diagnostic waits in batch mode
Fixed workspace symbol search to query all configured LSP servers and filter out non-matching results
Fixed references/rename/hover symbol targeting to error when symbol is missing on the line or occurrence is out of bounds
Fixed reload without a file to reload all active configured language servers instead of only the first server
Fixed todo_write task normalization to auto-activate the first remaining task and include explicit remaining-items output in tool results, removing the need for an immediate follow-up start update

[13.3.7] - 2026-02-27

Breaking Changes

Removed preloadedSkills option from CreateAgentSessionOptions; skills are no longer inlined into system prompts
Removed skills field from Task schema; subagents now always inherit the session skill set instead of per-task skill selection
Removed Task tool per-task tasks[].skills support; subagents now always inherit the session skill set
Removed preloadedSkills system prompt plumbing and template sections; skills are no longer inlined as a separate preloaded block

Changed

Refactored schema reference resolution to inline all $ref definitions instead of preserving them at the root level, eliminating unresolved references in tool parameters
Added lenientArgValidation flag to SubmitResultTool to allow the agent loop to bypass strict argument validation errors
Modified schema validation to allow non-conforming output on second validation failure, enabling recovery from strict schema constraints after initial rejection
Updated JTD-to-TypeScript conversion to gracefully fall back to 'unknown' type when conversion fails, preventing template rendering errors
Changed JTD-to-JSON Schema conversion to normalize nested JTD fragments within JSON Schema nodes, enabling mixed schema definitions
Changed output schema validation to gracefully fall back to unconstrained object when schema is invalid, instead of rejecting submissions
Changed schema sanitization to remove strict-mode incompatible constraints (minLength, pattern, etc.) from tool parameters while preserving them for runtime validation
Simplified task execution to always pass available session skills to subagents instead of resolving per-task skill lists
Added KILO_API_KEY to CLI environment variable help text for Kilo Gateway provider setup (#193)

Removed

Removed preloaded skills section from system prompt templates; skills are now referenced only as available resources

Fixed

Fixed schema compilation validation by adding explicit AJV compilation check to catch unresolved $ref references and other schema errors before tool execution
Fixed handling of circular and deeply nested output schemas to prevent stack overflow and enable successful result submission with fallback unconstrained schema
Fixed processing of non-object output schemas (arrays, primitives, booleans) to accept valid result submissions without blocking
Fixed handling of mixed JTD and JSON Schema output definitions to properly convert all nested JTD elements (e.g., elements → items, int32 → integer)
Fixed strict schema generation for output schemas with only required fields, enabling proper Claude API compatibility
Fixed handling of union type schemas (e.g., object|null) to normalize them into strict-mode compatible variants

[13.3.6] - 2026-02-26

Breaking Changes

Changed submit_result tool parameter structure from top-level data or error fields to nested result object containing either result.data or result.error

[13.3.5] - 2026-02-26

Added

Added support for setting array and record configuration values using JSON syntax

Changed

Increased default async max jobs limit from 15 to 100 for improved concurrent task handling

Fixed

Improved config display formatting to properly render arrays and objects as JSON instead of [object Object]
Enhanced type display in config list output to show correct type indicators for number, array, and record settings

[13.3.3] - 2026-02-26

Added

Support for move parameter in computeHashlineDiff to enable file move operations alongside content edits

Changed

Modified no-op detection logic to allow move-only operations when file content remains unchanged

[13.3.1] - 2026-02-26

Added

Added topP setting to control nucleus sampling cutoff for model output diversity
Added topK setting to sample from top-K tokens for controlled generation
Added minP setting to enforce minimum probability threshold for token selection
Added presencePenalty setting to penalize introduction of already-present tokens
Added repetitionPenalty setting to penalize repeated tokens in model output

Fixed

Fixed skill discovery to continue loading project skills when user skills directory is missing

[13.3.0] - 2026-02-26

Breaking Changes

Renamed task.isolation.enabled (boolean) setting to task.isolation.mode (enum: none, worktree, fuse-overlay). Existing true/false values are auto-migrated to worktree/none.

Added

Added PERPLEXITY_COOKIES env var for Perplexity web search via session cookies extracted from desktop app
Added fuse-overlay isolation mode for subagents using fuse-overlayfs (copy-on-write overlay, no baseline patch apply needed)
Added task.isolation.merge setting (patch or branch) to control how isolated task changes are integrated back. branch mode commits each task to a temp branch and cherry-picks for clean commit history
Added task.isolation.commits setting (generic or ai) for commit messages on isolated task branches and nested repos. ai mode uses a smol model to generate conventional commit messages from diffs
Nested non-submodule git repos are now discovered and handled during task isolation (changes captured and applied independently from parent repo)
Added task.eager setting to encourage the agent to delegate work to subagents by default
Added manual OAuth login flow that lets users paste redirect URLs with /login for callback-server providers and prevents overlapping logins

Fixed

Fixed nested repo changes being lost when tasks commit inside the isolation (baseline state is now committed before task runs, so delta correctly excludes it)
Fixed nested repo patches conflicting when multiple tasks contribute to the same repo (baseline untracked files no longer leak into patches)
Nested repo changes are now committed after patch application (previously left as untracked files)
Failed tasks no longer create stale branches or capture garbage patches (gated on exit code)
Merge failures (e.g. conflicting patches) are now non-fatal — agent output is preserved with merge failed status instead of failed
Stale branches are cleaned up when commitToBranch fails
Commit message generator filters lock files from diffs before AI summarization

[13.2.1] - 2026-02-24

Fixed

Fixed changelog tools to enforce category-specific arrays and reuse the shared category list for generation
Non-interactive environment variables (pager, editor, prompt suppression) were not applied to non-PTY bash execution, causing commands to potentially block on pagers or prompts

Changed

Extracted non-interactive environment config from bash-interactive.ts into shared non-interactive-env.ts module, applied consistently to all bash execution paths

[13.2.0] - 2026-02-23

Breaking Changes

Made description field required in CustomTool interface

[13.1.2] - 2026-02-23

Breaking Changes

Removed timeout parameter from await tool—tool now waits indefinitely until jobs complete or the call is aborted
Renamed job_ids parameter to jobs in await tool schema
Removed timedOut field from await tool result details

[13.1.1] - 2026-02-23

Fixed

Fixed bash internal URL expansion to resolve local:// targets to concrete filesystem paths, including newly created destination files for commands like mv src.json local://dest.json
Fixed bash local URL resolution to create missing parent directories under the session local root before command execution, preventing mv destination failures for new paths

[13.1.0] - 2026-02-23

Breaking Changes

Renamed file parameter to path in replace, patch, and hashline edit operations

Added

Added clarification in hashline edit documentation that the end tag must include closing braces/brackets when replacing blocks to prevent syntax errors

Changed

Restructured task tool documentation for clarity, moving parameter definitions into a dedicated section and consolidating guidance on context, assignments, and parallelization
Reformatted system prompt template to use markdown headings instead of XML tags for skills, preloaded skills, and rules sections
Renamed deviceScaleFactor parameter to device_scale_factor in browser viewport configuration for consistency with snake_case naming convention
Moved intent field documentation from per-tool JSON schema descriptions into a single system prompt block, reducing token overhead proportional to tool count

[13.0.1] - 2026-02-22

Changed

Simplified hashline edit schema to use unified first/last anchor fields instead of operation-specific field names (tag, before, after)
Improved resilience of anchor resolution to degrade gracefully when anchors are missing or invalid, allowing edits to proceed with available anchors
Updated hashline tool documentation to reflect new unified anchor syntax across all operations (replace, append, prepend, insert)

[13.0.0] - 2026-02-22

Added

Added getTodoPhases() and setTodoPhases() methods to ToolSession API for managing todo state programmatically
Added getLatestTodoPhasesFromEntries() export to retrieve todo phases from session history
Added local:// protocol for session-scoped scratch space to store large intermediate artifacts, subagent handoffs, and reusable planning artifacts
Added title parameter to exit_plan_mode tool to specify the final plan artifact name when approving a plan
Added LocalProtocolHandler for resolving local:// URLs to session-scoped file storage
Added renameApprovedPlanFile function to finalize approved plans with user-specified titles

Changed

Changed todo state management from file-based (todos.json) to in-memory session cache for improved performance and consistency
Changed todo phases to sync from session branch history when branching or rewriting entries
Changed TodoWriteTool to update session cache instead of writing to disk, with automatic persistence through session entries
Changed XML tag from <swarm-context> to <context> in subagent prompts and task rendering
Changed system reminder XML tags from underscore to kebab-case format (<system-reminder>)
Changed plan storage from plan:// protocol to local://PLAN.md for draft plans and local://<title>.md for finalized approved plans
Changed plan mode to use session artifacts directory for plan storage instead of separate plans directory
Updated system prompt to document local:// protocol and internal URL expansion behavior
Updated exit_plan_mode tool documentation to require title parameter and explain plan finalization workflow
Updated write tool documentation to recommend local:// for large temporary artifacts and subagent handoffs
Updated task tool documentation to recommend using local:// for large intermediate outputs in subagent context
Replaced docs:// protocol with pi:// for accessing embedded documentation files
Renamed DocsProtocolHandler to PiProtocolHandler for internal documentation URL resolution
Removed artifactsDir parameter from Python executor options; artifact storage now uses artifactPath only
Renamed prompt file from read_path.md to read-path.md for consistency
Updated system prompt XML tags to use kebab-case (e.g., system-reminder, system-interrupt) for consistency
Refactored bash tool to use NO_PAGER_ENV constant for environment variable management
Updated internal URL expansion to support optional noEscape parameter for unescaped path resolution

Removed

Removed plan:// protocol handler and related plan directory resolution logic
Removed PlanProtocolHandler and resolvePlanUrlToPath exports from internal URLs module

Fixed

Fixed todo reminder XML tags from underscore to kebab-case format (system-reminder)

[12.19.3] - 2026-02-22

Added

Added pty parameter to bash tool to enable PTY mode for commands requiring a real terminal (e.g., sudo, ssh, top, less)

Changed

Changed bash tool to use per-command PTY control instead of global virtual terminal setting

Removed

Removed bash.virtualTerminal setting; use the pty parameter on individual bash commands instead

[12.19.1] - 2026-02-22

Removed

Removed replaceText edit operation from hashline mode (substring-based text replacement)
Removed autocorrect heuristics that attempted to detect and fix line merges and formatting rewrites in hashline edits

[12.19.0] - 2026-02-22

Added

Added poll_jobs tool to block until background jobs complete, providing an alternative to polling read jobs:// in loops
Added task.maxConcurrency setting to limit the number of concurrently executing subagent tasks
Added support for rendering markdown output from Python cells with proper formatting and theme styling
Added async background job execution for bash commands and tasks with async: true parameter
Added cancel_job tool to cancel running background jobs
Added jobs:// internal protocol to inspect background job status and results
Added /jobs slash command to display running and recent background jobs in interactive mode
Added async.enabled and async.maxJobs settings to control background job execution
Added background job status indicator in status line showing count of running jobs
Added support for GitLab Duo authentication provider
Added clearer truncation notices across tools with consistent line/size context and continuation hints

Changed

Updated bash and task tool guidance to recommend poll_jobs instead of polling read jobs:// in loops when waiting for async results
Improved parallel task execution to schedule multiple background jobs independently instead of batching all tasks into a single job, enabling true concurrent execution
Enhanced task progress tracking to report per-task status (pending, running, completed, failed, aborted) with individual timing and token metrics for each background task
Updated background task messaging to provide real-time progress counts (e.g., '2/5 finished') and distinguish between single and multiple task jobs
Hid internal agent__intent parameter from tool argument displays in UI and logs to reduce visual clutter
Updated Python tool to detect and handle markdown display output separately from plain text
Updated bash tool to support async execution mode with streaming progress updates
Updated task tool to support async execution mode for parallel subagent execution
Modified subagent settings to disable async execution in child agents to prevent nesting
Updated tool execution component to handle background async task state without spinner animation
Changed event controller to keep background tool calls pending until async completion
Updated status line width calculation to accommodate background job indicator
Updated the system prompt pipeline to reduce injected environment noise and make instructions more focused on execution quality
Updated system prompt/workflow guidance to emphasize root-cause fixes, code quality, and explicit handoff/testing expectations
Changed default value of todo.reminders setting from false to true to enable todo reminders by default
Improved truncation/output handling for large command results to reduce memory pressure and keep previews responsive
Updated internal artifact handling so tool output artifacts stay consistent across session switches and resumes

Removed

Removed git context (branch, status, commit history) from system prompt — version control information is no longer injected into agent instructions

Fixed

Fixed task progress display to hide tool count and token metrics when zero, reducing visual clutter in status output
Fixed Lobsters scraper to correctly parse API responses where user fields are strings instead of objects, resolving undefined user display in story listings
Fixed artifact manager caching to properly invalidate when session file changes, preventing stale artifact references
Fixed truncation behavior around UTF-8 boundaries and chunked output accounting
Fixed submit_result schema generation to use valid JSON Schema when no explicit output schema is provided

[12.18.1] - 2026-02-21

Added

Added Buffer.toBase64() polyfill for Bun compatibility to enable base64 encoding of buffers

[12.18.0] - 2026-02-21

Added

Added overlay option to custom UI hooks to display components as bottom-centered overlays instead of replacing the editor
Added automatic chat transcript rebuild when returning from custom or debug UI to prevent message duplication

Changed

Changed custom UI hook cleanup to conditionally restore editor state only when not using overlay mode
Extracted environment variable configuration for non-interactive bash execution into reusable NO_PAGER_ENV constant
Replaced custom timing instrumentation with logger.timeAsync() and logger.time() from pi-utils for consistent startup profiling
Removed PI_DEBUG_STARTUP environment variable in favor of logger.debug() for conditional debug output
Consolidated timing calls throughout initialization pipeline to use unified logger-based timing system

Removed

Deleted utils/timings.ts module - timing functionality now provided by pi-utils logger

Fixed

Fixed potential race condition in bash interactive component where output could be appended after the component was closed

[12.17.2] - 2026-02-21

Changed

Modified bash command normalization to only apply explicit head/tail parameters from tool input, removing automatic extraction from command pipes
Updated shell snapshot creation to use explicit timeout and kill signal configuration for more reliable process termination

Fixed

Fixed persistent shell session state not being reset after command abort or hard timeout, preventing stale environment variables from affecting subsequent commands
Fixed hard timeout handling to properly interrupt long-running commands that exceed the grace period beyond the configured timeout

[12.17.1] - 2026-02-21

Added

Added filterBrowser option to filter out browser automation MCP servers when builtin browser tool is enabled
Added isBrowserMCPServer() function to detect browser automation MCP servers by name, URL, or command patterns
Added filterBrowserMCPServers() function to remove browser MCP servers from loaded configurations
Added BrowserFilterResult type for browser MCP server filtering results

[12.17.0] - 2026-02-21

Added

Added timeout protection (5 seconds) for system prompt preparation with graceful fallback to minimal context on timeout

Changed

Replaced glob-based AGENTS.md discovery with depth-limited directory traversal (depth 1-4) for improved performance and control
Refactored system prompt preparation to parallelize file loading operations with a 5-second timeout to prevent startup hangs
Unified renderCall signatures to (args, options, theme) across all tool renderers and extension types

[12.16.0] - 2026-02-21

Added

Added peekApiKey method to AuthStorage for non-blocking API key retrieval during model discovery without triggering OAuth token refresh
Exported finalizeSubprocessOutput function to handle subprocess output finalization with submit_result validation
Exported SubmitResultItem interface for type-safe submit_result tool data extraction
Added automatic reminders when subagent stops without calling submit_result tool (up to 3 reminders before aborting)
Added system warnings when subagent calls submit_result with null/undefined data or exits without calling submit_result after reminders

Changed

Changed model refresh behavior to support configurable strategies: uses 'online' mode when listing models and 'online-if-uncached' mode otherwise for improved performance
Changed default thinking level from 'off' to 'high' for improved reasoning and planning
Changed model discovery to use non-blocking API key peek instead of full key retrieval, improving performance by avoiding unnecessary OAuth token refreshes
Simplified submit_result termination logic to immediately abort on successful tool execution instead of waiting for message_end event
Updated submit_result tool to only terminate on successful execution (when isError is false), allowing retries on tool errors
Refactored subprocess output finalization logic into dedicated finalizeSubprocessOutput function for better testability and maintainability
Improved handling of missing submit_result calls by automatically aborting with exit code 1 after 3 reminder prompts

Fixed

Fixed submit_result retry behavior to properly handle tool execution errors and allow the subagent to retry before aborting
Fixed submit_result tool extraction to properly validate status field and only accept 'success' or 'aborted' results

[12.15.1] - 2026-02-20

Changed

Replaced nerd font pie-chart spinner with clock-outline icons for smoother looping
Moved status icon to front of code-cell headers in formatHeader

Fixed

Fixed ReadToolGroupComponent to show status icon before title instead of trailing
Fixed bash-interactive status badge to dim only bracket characters, not the enclosed text

[12.15.0] - 2026-02-20

Added

Added includeDisabled parameter to listAuthCredentials() to optionally retrieve disabled credentials
Added disableAuthCredential() method for soft-deleting auth credentials while preserving database records

Changed

Changed auth credential removal to use soft-delete (disable) instead of hard-delete when OAuth refresh fails, keeping credentials in database for audit purposes
Changed default value of tools.intentTracing setting from false to true

[12.14.1] - 2026-02-19

Fixed

Fixed omp stats failing on npm/bun installs by including required stats build files in published @oh-my-pi/omp-stats package (#113 by @masonc15)

[12.14.0] - 2026-02-19

Added

Support for docs:// internal URL protocol to access embedded documentation files (e.g., docs://sdk.md)
Added generate-docs-index npm script to automatically index and embed documentation files at build time
Support for executable tool files (.ts, .js, .sh, .bash, .py) in custom tools discovery alongside markdown files
Display streamed tool intent in working message during agent execution
Added tools.intentTracing setting to enable intent tracing, which asks the agent to describe the intent of each tool call before executing it
Support for file deletion in hashline edit mode via delete: true parameter
Support for file renaming/moving in hashline edit mode via rename parameter
Optional content-replace edit variant in hashline mode (enabled via PI_HL_REPLACETXT=1 environment variable)
Support for grepping internal URLs (artifact://) by resolving them to their backing files

Changed

System prompt now identifies agent as operating inside Oh My Pi harness and instructs reading docs:// URLs for omp/pi topics
Tool discovery now accepts executable script extensions (.ts, .js, .sh, .bash, .py) in addition to .json and .md files
Updated bash and read tool documentation to reference docs:// URL support
Hashline format separator changed from pipe (|) to colon (:) for improved readability (e.g., LINE#ID:content instead of LINE#ID|content)
Hashline hash representation changed from 4-character base36 to 2-character hexadecimal for more compact line references
Hashline edit API: renamed delete parameter to rm for consistency with standard file operations
Hashline edit API: renamed rename parameter to mv for consistency with standard file operations
Hashline edit API: content-replace operations now require explicit op: "replaceText" field to distinguish from other edit types
Hashline documentation terminology updated: references to 'anchors' replaced with 'tags' for clearer semantics
Intent tracing now uses _intent field name in tool schemas
Hashline edit API: renamed set operation to target/new_content for clearer semantics
Hashline edit API: renamed set_range operation to first/last/new_content
Hashline edit API: renamed insert operation fields from body to inserted_lines and made inserted_lines required non-empty
Hashline edit API: flattened replace operation to top-level fields (old_text, new_text, all) when enabled
Hashline edit validation now provides more specific error messages indicating which variant is expected

Fixed

Grep tool now properly handles internal URL resolution when searching artifact paths
Working message intent updates now fall back to tool execution events when streamed tool arguments omit the intent field

[12.13.0] - 2026-02-19

Breaking Changes

Removed automatic line relocation when hash references become stale; edits with mismatched line hashes now fail with an error instead of silently relocating to matching lines elsewhere in the file

Added

Added ssh command for managing SSH host configurations (add, list, remove)
Added /ssh slash command in interactive mode to manage SSH hosts with subcommands
Added support for SSH host configuration at project and user scopes (.omp/ssh.json and ~/.omp/agent/ssh.json)
Added --host, --user, --port, --key, --desc, --compat, and --scope flags for SSH host configuration
Added discovery of SSH hosts from project configuration files alongside manually configured hosts
Added NanoGPT as a login provider (/login nanogpt) with API key prompt flow linking to https://nano-gpt.com/api (#111)

Changed

Updated hashline reference format from LINE:HASH to LINE#ID throughout the codebase for improved clarity
Renamed hashline edit operations: set_line → set, replace_lines → set_range, insert_after → insert with support for before and between anchors
Changed hashline edit body field from string to array of strings for clearer multiline handling
Updated handlebars helpers: renamed hashline to hlineref and added hlinefull for formatted line output
Improved insert operation to support before, after, and between (both anchors) positioning modes
Made autocorrect heuristics (boundary echo stripping, indent restoration) conditional on PI_HL_AUTOCORRECT environment variable
Updated SSH host discovery to load from managed omp config paths (.omp/ssh.json and ~/.omp/agent/ssh.json) in addition to legacy root-level ssh.json and .ssh.json files
Improved terminal output handling in interactive bash sessions to ensure all queued writes complete before returning results

Fixed

Fixed insert-between operation to properly validate adjacent anchor lines and strip boundary echoes from both sides
Fixed terminal output handling to properly queue and serialize writes, preventing dropped or corrupted output in interactive bash sessions

[12.12.1] - 2026-02-19

Added

Added Kimi (Moonshot) as a web search provider with OAuth and API key support (#110 by @oglassdev)

Changed

Changed web search auto-resolve priority to prefer Perplexity first

Fixed

Fixed Mermaid pre-render failures from repeatedly re-triggering background renders (freeze loop) and restored resilient rendering when diagram conversion/callbacks fail (#109).

[12.12.0] - 2026-02-19

Added

Display streaming text preview during agent specification generation to show real-time progress
Added onRequestRender callback to agent dashboard for triggering UI updates during async operations
Added agent creation flow (press N in dashboard) to generate custom agents from natural language descriptions
Added ability to save generated agents to project or user scope with automatic identifier and system prompt generation
Added scope toggle (Tab) during agent creation to choose between project-level and user-level agent storage
Added agent regeneration (R key) to refine generated specifications without restarting the creation flow
Added model suggestions in model override editor to help users discover available models
Added success notices to confirm agent creation and model override updates

Changed

Updated agent creation flow to show review screen before generation completes, improving UX feedback
Changed generation status hint to display "Generating..." while specification is being created
Improved system prompt preview formatting with text wrapping and line truncation indicators

Fixed

Fixed interactive-mode editor height to stay bounded and resize-aware, preventing off-screen cursor drift during long prompt/history navigation (#99).

[12.11.3] - 2026-02-19

Fixed

Fixed model selector search initialization to apply the latest live query after asynchronous model loading.
Fixed Codex provider session lifecycle on model switches and history rewrites to clear stale session metadata before continuing the conversation.

[12.11.0] - 2026-02-19

Added

Support for Synthetic model provider in web search command
Model sorting by priority field and version number in model selector for improved model ranking
Support for Synthetic model provider with API key authentication
Support for Hugging Face model provider with API key authentication
Support for NVIDIA model provider with API key authentication
Support for Ollama model provider with optional API key authentication
Support for Cloudflare AI Gateway model provider with API key authentication
Support for Qwen Portal model provider with API key authentication
Support for LiteLLM model provider with API key authentication
Support for Moonshot model provider with API key authentication
Support for Qianfan model provider with API key authentication
Support for Together model provider with API key authentication
Support for Venice model provider with API key authentication
Support for vLLM model provider with API key authentication
Support for Xiaomi model provider with API key authentication

Changed

Refactored custom model building logic into reusable buildCustomModel function for consistency across provider configurations
Replaced generic error with AgentBusyError when attempting to send messages while agent is processing
Added automatic retry logic with idle waiting when agent is busy during prompt operations, with 30-second timeout

Fixed

Fixed model discovery to use default refresh mode instead of explicit 'online' parameter

[12.10.1] - 2026-02-18

Added

Added /login support for Cerebras and Synthetic API-key providers

[12.10.0] - 2026-02-18

Breaking Changes

Changed keyless provider auth sentinel from "<no-auth>" to kNoAuth ("N/A") for ModelRegistry.getApiKey() and ModelRegistry.getApiKeyForProvider()

Added

Added --no-rules CLI flag to disable rules discovery and loading
Added sessionDir option to RpcClientOptions for specifying agent session directory
Added Symbol.dispose method to RpcClient for resource cleanup support
Added rules option to CreateAgentSessionOptions for explicit rule configuration
Added sessionDir option to RpcClientOptions for specifying agent session directory
Added Symbol.dispose method to RpcClient for resource cleanup support
Added condition and scope fields to rule frontmatter for advanced TTSR matching and stream filtering
Added ttsr.interruptMode setting to control when TTSR rules interrupt mid-stream vs inject warnings after completion
Added support for loading rules, prompts, commands, context files (AGENTS.md), and system prompts (SYSTEM.md) from ~/.agent/ directory (with fallback to ~/.agents/)
Added scoped stream buffering for TTSR matching to isolate prose, thinking, and tool argument streams
Added file-path-aware TTSR scope matching for tool calls with glob patterns (e.g., tool:edit(*.ts))
Added legacy field support: ttsr_trigger and ttsrTrigger are accepted as fallback for condition

Changed

Changed TTSR injection tracking to record all turns where rules were injected (instead of only the last turn) to support repeat-after-gap mode across resumed sessions
Changed TTSR injection messages to use custom message type with metadata instead of synthetic user messages for better session tracking
Changed TTSR rule injection to persist injected rule names in session state for restoration when resuming sessions
Changed model discovery to automatically discover built-in provider models (Anthropic, OpenAI, Groq, Cerebras, Xai, Mistral, OpenCode, OpenRouter, Vercel AI Gateway, Kimi Code, GitHub Copilot, Google, Cursor, Google Antigravity, Google Gemini CLI, OpenAI Codex) when credentials are configured
Changed getModel() and getModels() imports to getBundledModel() and getBundledModels() across test utilities
Changed TTSR rule matching from single ttsrTrigger regex to multiple condition patterns with scope filtering
Changed TTSR buffer management to use per-stream-key buffers instead of a single global buffer
Changed rule discovery to use unified buildRuleFromMarkdown helper across all providers (builtin, cline, cursor, windsurf, agents)
Changed TTSR injection to defer warnings until stream completion when interruptMode is not always
Changed TtsrManager.addRule() to return boolean indicating successful registration instead of void

Fixed

Fixed TTSR repeat-after-gap mode to correctly calculate gaps when rules are restored from previous sessions
Fixed TTSR matching to respect tool-specific scope filters, preventing cross-tool rule contamination
Fixed path normalization in TTSR glob matching to handle both relative and absolute path variants

[12.9.0] - 2026-02-17

Added

Added OpenCode discovery provider to load configuration from ~/.config/opencode/ and .opencode/ directories
Added support for loading MCP servers from opencode.json mcp key
Added support for loading skills from ~/.config/opencode/skills/ and .opencode/skills/
Added support for loading slash commands from ~/.config/opencode/commands/ and .opencode/commands/
Added support for loading extension modules (plugins) from ~/.config/opencode/plugins/ and .opencode/plugins/
Added support for loading context files (AGENTS.md) from ~/.config/opencode/
Added support for loading settings from opencode.json configuration files

Changed

Improved path display in status line to strip both /work/ and ~/Projects/ prefixes when abbreviating paths
Refactored session directory naming to use single-dash format for home-relative paths and double-dash format for absolute paths, with automatic migration of legacy session directories on first access

[12.8.2] - 2026-02-17

Changed

Changed system environment context to use built-in os values for distro, kernel, and CPU model instead of native system-info data
Changed environment info generation to stop including unavailable native system detail fallbacks

Removed

Removed the Disk field from generated environment information

[12.8.0] - 2026-02-16

Changed

Improved /changelog performance by displaying only the most recent 3 versions by default, with a --full flag for the complete history (#85 by @tctev)
Centralized builtin slash command definitions and handlers into a shared registry, replacing the large input-controller if-chain dispatch

[12.7.5] - 2026-02-16

Changed

Updated SMOL_MODEL_PRIORITY to include additional model variants (gpt-5.3-spark, cerebras/zai-glm-4.7, haiku-4-5, haiku-4.5) for improved fast model discovery
Updated SLOW_MODEL_PRIORITY to include additional model variants (gpt-5.3-codex, gpt-5.3, gpt-5.1-codex, gpt-5.1, opus-4.6, opus-4-6, opus-4.5, opus-4-5, opus-4.1, opus-4-1) for improved comprehensive model discovery

[12.7.3] - 2026-02-16

Fixed

Suppressed hashline output for subagents without the edit tool (e.g. explore agents)

[12.7.1] - 2026-02-16

Changed

Restructured execution procedure guidance to emphasize scope assessment and continuous tool-driven progress over planning-first approach
Updated task tracking instructions to prohibit metadata-only turns and require immediate completion marking of todo items
Clarified parallel execution guidance to distinguish between genuine parallelization and sequential work, with explicit criteria for Task tool usage
Modified output style requirements to mandate stating intent before tool calls and reordered style constraints for clarity
Reinforced requirement that every turn must include at least one tool call advancing the deliverable, with explicit failure condition for planning-only turns

[12.7.0] - 2026-02-16

Added

Added Z.AI web search provider support via remote MCP endpoint (webSearchPrime)
Added zai as a selectable web search provider option in settings
Added Z.AI to automatic provider fallback chain for web search

[12.6.0] - 2026-02-16

Added

Added runtime tests covering extension provider registration and deferred model pattern resolution behavior.

Changed

Improved welcome screen responsiveness to dynamically show or hide the right column based on available terminal width

Fixed

Fixed welcome screen layout to gracefully handle small terminal widths and prevent rendering errors on narrow displays
Fixed welcome screen title truncation to prevent overflow when content exceeds available width
Fixed deferred --model resolution so extension-provided models are matched before fallback selection and unresolved explicit patterns no longer silently fallback.

[12.5.1] - 2026-02-15

Added

Added repeatToolDescriptions setting to render full tool descriptions in the system prompt instead of a tool name list

[12.5.0] - 2026-02-15

Breaking Changes

Replaced theme setting with theme.dark and theme.light (auto-migrated)

Added

Added previewTheme() function for non-destructive theme preview during settings browsing
Added animated microphone icon with color cycling during voice recording
Added support for discovering skills via symbolic links in skill directories
Added abort_and_prompt RPC command for atomic abort-and-reprompt without race conditions (#357)
Added automatic dark/light theme switching via SIGWINCH with separate theme.dark/theme.light settings, replacing the single theme setting (#65)
Added speech-to-text (STT) feature with Alt+H keybinding and /stt slash command
Added cross-platform audio recording: SoX, FFmpeg, arecord (Linux), PowerShell mciSendString (Windows fallback)
Added recording tool fallback chain — automatically tries each available tool in order
Added Python openai-whisper integration for transcription with automatic pip install
Added custom WAV-to-numpy pipeline in transcribe.py bypassing ffmpeg dependency
Added STT settings: stt.enabled, stt.language, stt.modelName
Added STT status line segment showing recording/transcribing state
Added /stt command with on, off, status, setup subcommands
Added auto-download of recording tools (best-effort FFmpeg via winget on Windows)
Added interactive debug log viewer with selection, copy, and expand/collapse controls
Added inline filtering and count display to the debug log viewer
Added pid filter toggle and load-older pagination controls to the debug log viewer
Enabled loading older debug logs from archived files in viewer
Added file hyperlinks for debug report paths in viewer

Changed

Changed theme preview to support asynchronous theme loading with request deduplication to prevent race conditions
Enhanced theme preview cancellation to restore the previously active theme instead of the last selected value
Refactored file discovery to use native glob with gitignore support instead of manual directory traversal, improving performance and consistency
Updated dependencies: glob to ^13.0.3, marked to ^17.0.2, puppeteer to ^24.37.3
Optimized skill and file discovery using native glob (Rust ignore crate) — reduces startup time by ~80% (1254ms → 6ms for skills)
Enhanced hashline reference parsing to handle prefixes like >>> and >> in line references
Strengthened type safety in hashline edit formatting with defensive null checks for incomplete edits
Changed STT status messages to display via state change callbacks instead of explicit status calls
Changed cursor visibility behavior during voice recording to hide hardware and terminal cursors

Removed

Removed dedicated STT status line segment in favor of animated cursor-based feedback

Fixed

Fixed theme preview updates being applied out-of-order when rapidly browsing theme options
Fixed skill discovery to correctly extract skill names from directory paths when frontmatter name is missing
Fixed session.abort() not clearing promptInFlight flag due to microtask ordering, which blocked subsequent prompts
Sanitized debug log display to strip control codes, normalize tabs, and trim width

[12.4.0] - 2026-02-14

Changed

Moved sanitizeText function from @oh-my-pi/pi-utils to @oh-my-pi/pi-natives for better code organization
Replaced internal #normalizeOutput methods with sanitizeText utility function in bash and Python execution components
Added line length clamping (4000 characters) to bash and Python execution output to prevent display of excessively long lines
Modified memory storage to isolate memories by project working directory, preventing cross-project memory contamination

Fixed

Fixed bash interactive tool to gracefully handle malformed output chunks by normalizing them before display
Fixed fetch tool incorrectly treating HTML content as plain text or markdown
Fixed output truncation notice displaying incorrect byte limit when maxBytes differs from outputBytes
Fixed Cloudflare returning corrupted bytes when compression is negotiated in web scraper requests

[12.3.0] - 2026-02-14

Added

Added autonomous memory extraction and consolidation system with configurable settings
Added /memory slash command with subcommands: view, clear, reset, enqueue, rebuild
Added memory injection payload that automatically includes learned context in system prompts
Added two-phase memory pipeline: Stage 1 extracts durable knowledge from session history, Phase 2 consolidates into reusable skills and guidance
Added memory storage layer with SQLite-backed job queue for distributed memory processing
Added configurable memory settings: concurrency limits, lease timeouts, token budgets, and rollout age constraints

Changed

Modified system prompt building to inject memory guidance when memories are enabled
Changed resolvePromptInput to handle multiline input and improve error handling for file reads

[12.2.0] - 2026-02-13

Added

Added providerSessionState property to AgentSession for managing provider-scoped transport and session caches
Added automatic cleanup of provider session state resources on session disposal
Added providers.openaiWebsockets setting to prefer websocket transport for OpenAI Codex models
Added provider details display in session info showing authentication mode, transport, and connection settings
Added automatic prewarm of OpenAI Codex websocket connections on session creation for improved performance
Added real-time authentication validation in OAuth provider selector with visual status indicators (checking, valid, invalid)
Added validateAuth and requestRender options to OAuthSelectorComponent for custom authentication validation and UI refresh callbacks

Changed

Changed providers.openaiWebsockets setting from boolean to enum with values "auto", "off", "on" for more granular websocket policy control (auto uses model defaults, on forces websocket, off disables it)
Enhanced provider details display to include live provider session state information
Enhanced session info output to display active provider configuration and authentication details
Replaced process.cwd() with getProjectDir() throughout codebase for improved project directory detection and handling
Made SessionManager.list() async to support asynchronous session discovery operations
Preserved internal whitespace and indentation in bash command normalization to support heredocs and indentation-sensitive scripts
Improved git context loading performance with configurable timeouts and parallel status/commit queries
Enhanced git context reliability with better error handling for timeout and command failures
Changed OAuth provider selector to display live authentication status instead of static login state
Changed logout flow to refresh OAuth provider authentication state before showing selector

Fixed

Improved error reporting in fetch tool to include HTTP status codes when URL fetching fails
Fixed fetch tool to preserve actual response metadata (finalUrl, contentType) instead of defaults when requests fail

[12.1.0] - 2026-02-13

Added

Filesystem scan cache invalidation helpers (invalidateFsScanAfterWrite, invalidateFsScanAfterDelete, invalidateFsScanAfterRename) to properly invalidate shared caches after file mutations
Named discovery profile for file mention candidates to standardize cache visibility and ignore semantics across callers
Comprehensive models.yml provider integration guide documenting custom model registration, provider overrides, API adapters, merge behavior, and practical integration patterns for Ollama, vLLM, LM Studio, and proxy endpoints
Claude Code marketplace plugin discovery: automatically loads skills, commands, hooks, tools, and agents from ~/.claude/plugins/cache/ based on installed_plugins.json registry (#48)

Changed

Moved directory path utilities from src/config.ts to @oh-my-pi/pi-utils/dirs for shared use across packages
Updated imports throughout codebase to use centralized directory path functions from @oh-my-pi/pi-utils/dirs
Updated interactive bash terminal UI label from 'InteractiveTerm' to 'Console' for clarity
Enhanced bash execution environment with comprehensive non-interactive defaults for pagers, editors, and package managers to prevent command blocking and interactive prompts
Updated custom models configuration to use ~/.omp/agent/models.yml (YAML format) while maintaining backward compatibility with legacy models.json

[12.0.0] - 2026-02-12

Added

Added getAllServerNames() method to MCPManager for enumerating all known servers

Changed

Changed default edit mode from patch to hashline for more precise code modifications
Changed readHashLines setting default from false to true to enable hash line reading by default

Fixed

Fixed omp setup crashing with uncaught exception when no component argument provided; now shows help (#35)
Fixed /mcp list showing "No MCP servers configured" when servers are loaded from discovery sources like .claude.json, .cursor/mcp.json, .vscode/mcp.json (#34)
Fixed model selector sorting to show newest models first within each provider instead of alphabetical; -latest aliases now appear before dated versions (#37)

[11.14.4] - 2026-02-12

Added

Exported renderPromptTemplate function for programmatic prompt template rendering
Exported computeLineHash function from patch utilities
Added ./cli export path for direct CLI module access

Changed

Replaced jsdom with linkedom for improved HTML parsing performance and reduced memory footprint

Removed

Removed @types/jsdom dependency

[11.14.1] - 2026-02-12

Changed

Improved Bun binary detection to check Bun.env.PI_COMPILED environment variable
Enhanced Bun package manager update to install specific version instead of latest
Added post-update verification for Bun installations to warn if expected version was not installed

Fixed

Fixed Bun update process to properly handle version pinning and report installation mismatches

[11.14.0] - 2026-02-12

Added

Added SwiftLint linter client with JSON reporter support for Swift file linting
Added --no-pty flag to disable PTY-based interactive bash execution
Added PI_NO_PTY environment variable to disable PTY-based interactive bash execution
Added bash.virtualTerminal setting to control PTY-backed interactive execution for bash commands
Added interactive PTY-based bash execution with real-time terminal rendering and input forwarding
Added sourcekit-lsp language server support for Swift files

Changed

Changed bash.virtualTerminal default from on to off for standard non-interactive bash execution
Changed SwiftLint configuration to use lint command with JSON reporter instead of analyze for improved diagnostic parsing
Changed diff line format from space-separated (+123 content) to pipe-delimited (+123|content) for improved parsing reliability
Changed bash tool to use interactive PTY execution by default when UI is available, falling back to standard execution when disabled

[11.13.1] - 2026-02-12

Added

Added /move slash command to move session to a different working directory
Added moveTo() method to SessionManager for relocating sessions with file migration and header updates
Added refreshSlashCommandState() method to reload slash commands and autocomplete when working directory changes
Added setSlashCommands() method to AgentSession for updating file-based slash commands
Added OAuth authentication support for Perplexity web search via www.perplexity.ai/rest/sse/perplexity_ask endpoint
Added automatic OAuth token refresh with 5-minute expiry buffer for Perplexity authentication
Added authMode field to search responses to indicate authentication method used (oauth or api_key)
Added display of authentication mode in search result output
Added support for streaming SSE responses from Perplexity OAuth API with proper event merging

Changed

Changed Perplexity provider to support both API key and OAuth authentication methods
Changed isAvailable() method to async to check for both API key and OAuth token availability
Changed error message to guide users to set PERPLEXITY_API_KEY or login via OAuth
Changed callPerplexity to callPerplexityApi to clarify it uses the API key endpoint

[11.13.0] - 2026-02-12

Breaking Changes

Removed support for .pi configuration directory alias; use .omp instead

Added

Added openPath utility function to centralize cross-platform URL and file path opening

Changed

Refactored browser/file opening across multiple modules to use unified openPath utility for improved maintainability

[11.12.0] - 2026-02-11

Added

Added resolveFileDisplayMode utility to centralize file display mode resolution across tools (read, grep, file mentions)
Added automatic hashline formatting to @file mentions when hashline mode is active
Added replace hashline edit operation for substr-style fuzzy text replacement without line references, with optional all flag for replace-all behavior
Added noopEdits array to applyHashlineEdits return value to report edits that produced no changes, including edit index, location, and current content for diagnostics
Added validation to detect and reject hashline edits using wrong-format fields (old_text/new_text from replace mode, diff from patch mode) with helpful error messages
Added additionalProperties: true to all hashline edit schemas (single, range, insertAfter, and root) to tolerate extra fields from models
Added whitespace normalization in line reference parsing to tolerate spaces around colons (e.g., 5 : ab now parses as 5:ab)
Added remaps property to HashlineMismatchError providing quick-fix mapping of stale line references to corrected hashes
Added warnings detection in applyHashlineEdits to alert users when edits affect significantly more lines than expected, indicating possible unintended reformatting
Added diagnostic output showing target line content when an edit produces no changes, helping users identify hash mismatches or incorrect replacement content
Added {{hashline}} Handlebars helper to compute accurate LINE:HASH references for prompt examples and documentation
Added deduplication of identical hashline edits targeting the same line(s) in a single call
Added replacement as accepted alias for content in insertAfter operations
Added graceful degradation of range edits with missing end field to single-line edits
Added additionalProperties: true to hashline edit schemas to tolerate extra fields from models

Changed

Reverted hashline display format from LINE:HASH content (two spaces) back to LINE:HASH|content (pipe separator) for consistency with legacy format
Changed hashline display format from LINE:HASH| content to LINE:HASH content (two spaces instead of pipe separator) for improved readability
Removed lines and hashes parameters from read tool—file display mode (line numbers, hashlines) now determined automatically by settings and edit mode
Simplified read tool prompt to reflect automatic display mode detection based on configuration
Updated grep tool to respect file display mode settings, showing hashline-prefixed output when hashline mode is active
Renamed hashline edit operation keys from single/range/insertAfter to set_line/replace_lines/insert_after for clearer semantics
Renamed hashline edit fields: loc → anchor, replacement → new_text, content → text for consistency across all operation types
Separated hashline anchor-based edits (set_line, replace_lines, insert_after) from content-replace edits (replace) in application pipeline
Improved no-op edit diagnostics to use noopEdits array from applyHashlineEdits, providing precise line-by-line comparison when replacements match current content
Enhanced error messages for wrong-format hashline edits to guide users toward correct operation syntax
Strengthened hashline prompt guidance to emphasize that replacement must differ from current line content and clarify no-op error recovery procedures
Improved hashline prompt to clarify atomicity: all edits in one call are validated against the original file state, with line numbers and hashes referring to the pre-edit state
Added explicit instruction in hashline prompt to preserve exact whitespace and formatting when replacing lines, changing only the targeted token/expression
Added guidance in hashline prompt for swap operations: use two single operations in one call rather than attempting to account for line number shifts
Strengthened anti-reformatting instructions in hashline prompt to reduce formatting-only failures
Improved no-op error recovery guidance in hashline prompt to prevent infinite retry loops
Renamed hashline edit operation keys from replaceLine/replaceLines to single/range for clearer semantics
Renamed hashline edit field content to replacement in single and range operations to distinguish from insertAfter.content
Improved no-op edit diagnostics to show specific line-by-line comparisons when replacements match current content, helping users identify hash mismatches or formatting issues
Enhanced no-op error messages to distinguish between literally identical replacements and content normalized back by heuristics
Reverted hash algorithm from 3-character base-36 back to 2-character hexadecimal for line references
Enhanced range validation during hashline edits to detect and reject relocations that change the scope of affected lines
Improved wrapped-line restoration logic to only attempt merging when source lines exhibit continuation patterns
Updated hashline tool documentation to emphasize direction-locking mutations and clarify recovery procedures for hash mismatches
Changed applyHashlineEdits return type to include optional warnings array for reporting suspicious edit patterns
Improved hash relocation logic to recompute touched lines after hash-based line number adjustments, preventing incorrect merge heuristics
Enhanced error messages for no-op edits to include preview of target lines with their current hashes and content
Changed hashline edit format from src/dst object structure to direct operation schemas (replaceLine, replaceLines, insertAfter)
Changed hash algorithm from 2-character hexadecimal to 3-character base-36 alphanumeric for improved readability and collision resistance
Improved hash mismatch handling to automatically relocate stale line references when the hash uniquely identifies a moved line
Changed HashlineEdit from src/dst format to direct operation schemas: replaceLine, replaceLines, insertAfter
Changed hash algorithm from hexadecimal (base-16) to base-36 alphanumeric for shorter, more readable line references
Increased maximum wrapped-line restoration from 6 to 10 lines to handle longer reflowed statements
Updated prompt examples to use {{hashline}} Handlebars helper for generating correct line references in tool instructions

Removed

Removed insertBefore hashline edit operation for inserting content before a line
Removed substr hashline edit operation for substring-based line replacement
Removed insertBefore and substr hashline edit variants

Fixed

Fixed parseLineRef to handle both legacy pipe-separator format (LINE:HASH| content) and new two-space format (LINE:HASH content) for backward compatibility
Fixed resource leak in browser query handler by properly disposing owned proxy elements for non-winning candidates
Fixed script evaluation to support async functions and await expressions in browser evaluate operations
Fixed range edits with missing end field to gracefully degrade to single-line edits instead of crashing
Fixed insertAfter operations to accept both content and replacement field names for consistency with other edit types
Fixed deduplication logic to correctly identify and remove identical hashline edits targeting the same line(s) in a single call
Fixed range-based edits to prevent invalid mutations when hash relocation changes the number of lines in the target range
Fixed multi-edit application to use original file state for all anchor references, preventing incorrect line numbers when earlier edits change file length

[11.10.4] - 2026-02-10

Added

Hashline diff computation with computeHashlineDiff function for preview rendering of hashline-mode edits
Streaming preview display for hashline edits in tool execution UI showing edit sources and destinations
Streaming hash line computation with progress updates via onUpdate callback in read tool
Optional onCollectedLine callback parameter to streamLinesFromFile for line collection tracking

Changed

Edit tool renderer now displays computed preview diffs for hashline operations before execution
Read tool now streams hash lines incrementally instead of computing them all at once, improving responsiveness for large files
Refactored hash line formatting to use async streamHashLinesFromLines for better performance

[11.10.3] - 2026-02-10

Added

Exported ./patch/* subpath for direct access to patch utilities

[11.10.2] - 2026-02-10

Added

Exported streamHashLinesFromUtf8 and streamHashLinesFromLines functions for streaming hashline-formatted output with configurable chunking
Added HashlineStreamOptions interface to control streaming behavior (startLine, maxChunkLines, maxChunkBytes)
Added streamHashLinesFromUtf8 function to incrementally format content with hash lines from a UTF-8 byte stream
Added streamHashLinesFromLines function to incrementally format content with hash lines from an iterable of lines

Changed

Updated hashline format to use 2-character hex hashes instead of 4-character hashes for more compact line references
Modified computeLineHash to normalize whitespace in line content and removed line number from hash seed for consistency
Improved CLI argument parsing to explicitly handle --help, --version, and subcommand detection instead of prefix-based routing

Removed

Removed @types/diff dev dependency
Removed AggregateError unwrapping from console.warn in CLI initialization

[11.10.1] - 2026-02-10

Changed

Migrated CLI framework from oclif to lightweight pi-utils CLI runner
Replaced oclif command registration with explicit command entries in cli.ts
Changed default root command name from 'index' to 'launch'
Updated all command imports to use @oh-my-pi/pi-utils/cli instead of @oclif/core

Removed

Removed @oclif/core and @oclif/plugin-autocomplete dependencies
Removed oclif configuration from package.json
Removed custom oclif help renderer (oclif-help.ts)

[11.10.0] - 2026-02-10

Breaking Changes

Changed HashlineEdit.src from string format (e.g., "5:ab", "5:ab..9:ef") to structured SrcSpec object with discriminated union types ({ kind: "single", ref: "..." }, { kind: "range", start: "...", end: "..." }, etc.)
Changed HashlineEdit API from old: string | string[] / new: string | string[] to src: string / dst: string; src uses range syntax "5:ab" (single), "5:ab..9:ef" (range), "5:ab.." (insert after), or "..5:ab" (insert before)
Removed support for comma and newline-separated line reference lists in hashline edits; use range syntax instead
Removed after field from HashlineEdit; insert-after is now expressed via open range syntax src: "5:ab.."
Changed HashlineEdit API from old: string | string[] / new: string | string[] to src: string / dst: string; multi-line content uses \n-separated strings, empty string "" for insert/delete operations
Replaced edit.patchMode boolean setting with edit.mode enum; existing edit.patchMode: true configurations should use edit.mode: patch
Changed getEditModelVariants() return type from Record<string, "patch" | "replace"> to Record<string, EditMode | null>
Removed after field from HashlineEdit; insert-after is now expressed via open range syntax src: "5:ab.."
Changed HashlineEdit.src from newline-separated line ref lists to range syntax: "5:ab" (single), "5:ab..9:ef" (range), "5:ab.." (insert after); comma and newline-separated lists are no longer supported

Added

Added substring-based source matching for hashline edits when line reference format is invalid, allowing fallback to unique substring search within the file
Added automatic detection and repair of single-line merges where models absorb adjacent lines, preventing content duplication
Added normalization of unicode-confusable hyphens (en-dash, em-dash, etc.) to ASCII hyphens when edits would otherwise be no-ops
Added heuristics to restore original indentation and preserve wrapped line formatting in hashline edits
Added abort signal support to MCP server connection and tool listing operations, allowing requests to be cancelled via Escape key during testing
Added MCPRequestOptions interface with signal property to support request cancellation via AbortSignal
Added abort signal support to MCP tool execution, allowing requests to be cancelled via Escape-to-interrupt or other abort mechanisms
Added HashlineMismatchError class that displays grep-style output with >>> markers showing correct LINE:HASH references when hash validation fails
Added HashMismatch type to represent individual hash mismatches with line number, expected hash, and actual hash
Added hashline edit mode for line-addressed edits using content hashes (LINE:HASH format) with integrity verification
Added readHashLines setting to include line hashes in read output for hashline edit mode
Added edit.mode setting (enum: replace, patch, hashline) to select edit tool variant, replacing edit.patchMode boolean
Added hashes parameter to read tool to output line hashes in format LINE:HASH| content
Added automatic hash line output when using hashline edit mode or readHashLines setting is enabled
Added computeLineHash, formatHashLines, parseLineRef, validateLineRef, and applyHashlineEdits functions for hashline operations
Added HashlineEdit and HashlineInput types for structured hashline edit operations
Added normalizeEditMode function to validate and normalize edit mode strings
Added subcommand definitions for /mcp command with 10 subcommands (add, list, remove, test, reauth, unauth, enable, disable, reload, help) including usage hints for argument completion
Added inline hint support for slash commands with simple arguments (/export [path], /compact [focus instructions], /handoff [focus instructions])
Added subcommand dropdown completion for /browser command (headless, visible modes)
Added SubcommandDef interface for declarative subcommand definitions with name, description, and usage hints
Added subcommands and inlineHint properties to BuiltinSlashCommand interface for enhanced command metadata
Added getArgumentCompletions and getInlineHint functions to materialized slash commands for autocomplete and ghost text hints
Added temperature setting to control sampling temperature (0 = deterministic, 1 = creative, -1 = provider default)
Added temperature option selector in settings UI with preset values (Default, 0, 0.2, 0.5, 0.7, 1)

Changed

Relaxed comma validation in src to allow trailing content after line references (e.g., 14:abexport function foo()), while still rejecting inputs that appear to contain multiple line refs
Improved parseLineRef to accept hash prefixes shorter than the full hash length, allowing partial hash matches
Enhanced error message for hash mismatches to guide users toward re-reading the file and using updated LINE:HASH references
Updated hashline tool documentation to clarify that accidental trailing text after LINE:HASH will be extracted, and to discourage merging multiple lines into single-line replacements
Treated same-line ranges (e.g., 5:ab..5:cd) as single-line replacements instead of range operations
Enhanced hashline edit robustness with heuristics to strip anchor line echoes and range boundary echoes that models may copy into replacement content
Improved whitespace preservation in hashline edits to handle mismatched line counts using loose matching strategy
Strengthened hashline edit validation to reject malformed src specs (embedded newlines, commas, invalid ranges)
Enhanced MCP connection timeout handling to properly respect abort signals and distinguish between timeout and user-initiated cancellation
Improved MCP test command UI to show cancellation hint (esc to cancel) and handle graceful cancellation without blocking cleanup
Updated HTTP transport session termination to use AbortSignal.timeout() for reliable cleanup with timeout protection
Enhanced bash executor to properly handle abort signals by registering abort event listeners and cleaning up resources in finally block
Improved bash tool error handling to distinguish between user-initiated aborts (via AbortSignal) and other cancellations, throwing ToolAbortError for aborted requests
Enhanced MCP request handling to propagate abort signals through HTTP, SSE, and stdio transports with proper cleanup
Improved stdio transport request handling to use Promise.withResolvers for cleaner async flow and better abort signal integration
Updated HTTP transport to combine operation abort signals with timeout signals using AbortSignal.any() for unified cancellation
Modified SSE response parsing to support abort signals and distinguish between timeout and user-initiated cancellation
Improved hashline edit robustness by automatically stripping LINE:HASH| display prefixes and unified-diff + markers that models may copy into replacement content
Enhanced replace edits to preserve original whitespace on lines where only whitespace differs, preventing spurious formatting diffs when models reformat code
Enhanced system prompt to display tool descriptions alongside tool names for improved clarity on available capabilities
Reduced hash length from 4 to 2 hex characters (16-bit hashes) for more concise line references
Updated HashlineEdit type to accept string | string[] for old and new fields, allowing single-line edits without array wrapping
Enhanced parseLineRef to strip display-format suffix (e.g., 5:ab| content → 5:ab), allowing models to copy full read output format
Improved validateLineRef to throw HashlineMismatchError with context lines instead of generic error, providing grep-style output with correct hashes
Modified hash computation to strip trailing carriage returns before hashing for consistent line hash values
Renamed HashlineEdit fields from src/dst to old/new for clarity in replace, delete, and insert operations
Enhanced hash validation in applyHashlineEdits to collect all mismatches before throwing, providing comprehensive error reporting with context lines
Changed edit.patchMode boolean setting to edit.mode enum (replace, patch, hashline) with default value patch
Changed edit tool to support three modes (replace, patch, hashline) instead of two, with dynamic mode selection based on model and settings
Changed read tool to prioritize hash lines over line numbers when both are requested
Changed getEditVariantForModel to return EditMode | null and removed hardcoded Kimi model detection
Renamed settingsInstance parameter to settings in CreateAgentSessionOptions for consistency
Updated all internal references from settingsInstance to settings throughout SDK and components

Fixed

Fixed substring source matching to reject ambiguous matches and provide helpful error messages showing all occurrences
Fixed substring source matching to require exactly one matching line in the file
Fixed MCP test command to properly clean up connections even when cancelled or aborted, preventing resource leaks

[11.9.0] - 2026-02-10

Added

Added /mcp slash command for runtime MCP server management (add, list, remove, enable, disable, test, reauth)
Added interactive multi-step wizard for adding MCP servers with transport auto-detection
Added OAuth auto-discovery and authentication flow for MCP servers requiring authorization
Added MCP config file writer for persisting server configurations at user and project level
Added enabled and timeout fields to MCP server configuration
Added runtime MCP manager reload and active tool registry rebind without restart
Added MCP command guide documentation

Changed

Replaced setTimeout with Bun.sleep() for improved performance in file lock retry logic
Refactored component invalidation handling to use dedicated helper function for cleaner code
Improved error handling in worktree baseline application to use isEnoent() utility instead of file existence checks
Updated bash tool to use standard Node.js fs.promises.stat() with isEnoent() error handling
Replaced tmpdir() named import with os namespace import for consistency
Migrated logging from chalk and console.error to structured logger from @oh-my-pi/pi-utils

Fixed

Improved browser script evaluation to handle both expression and statement forms, fixing evaluation failures for certain script types
Fixed unsafe OAuth endpoint extraction that could redirect token exchange to attacker-controlled URLs
Fixed PKCE code verifier stored via untyped property; now uses typed private field
Fixed refresh token fallback incorrectly using access token when no refresh token provided
Fixed MCP config files written with default permissions; now enforces 0o700/0o600 for secret protection
Fixed add wizard ignoring user-chosen environment variable name and auth header name
Fixed reauth endpoint discovery misclassifying non-OAuth servers as discovery failures
Fixed resolved OAuth tokens leaking into connection config, causing cache churn on token rotation
Fixed unvalidated type assertions for enabled/timeout config fields from user-controlled JSON
Fixed uncaught exceptions in /mcp add quick-add flow crashing the interactive loop
Fixed greedy /mcp prefix match routing /mcpfoo to MCP controller
Fixed stdio transport timeout timer leak keeping process alive after request completion

Removed

Removed GrepOperations interface from public API exports
Removed GrepToolOptions interface from public API exports
Removed unused _options parameter from GrepTool constructor

[11.8.1] - 2026-02-10

Added

Added current date to system prompt context in YYYY-MM-DD format for date-aware agent reasoning
Added file size display in UI when files are skipped due to size limits
Added support for gigabyte (GB) file size formatting in truncate utility

Changed

Changed skipped file messages to include file size information for better visibility into why files were excluded
Changed file processing to skip reading files exceeding 5MB (text) or 25MB (images) and include them as path-only references instead
Changed @mention auto-reading to skip files exceeding 5MB (text) or 25MB (images) to prevent out-of-memory issues with large files
Clarified that subagents automatically inherit full system prompt including AGENTS.md, context files, and skills — do not repeat project rules or conventions in task context
Updated task context guidance to focus on session-specific information subagents lack, eliminating redundant documentation of project constraints already available to them
Refined constraints template to emphasize task-specific rules and session decisions rather than global project conventions
Expanded anti-patterns section to explicitly flag redundant context that wastes tokens by repeating AGENTS.md rules, project constraints, and tool preferences

Fixed

Fixed bash tool hanging when commands spawn background jobs by properly detecting foreground process completion
Fixed bash tool occasionally hanging after command completion when background jobs keep stdout/stderr open
Fixed crash when auto-reading @mentions for very large files by skipping content injection with an explicit "skipped" note
Improved bash tool output draining after foreground completion to reduce tail output truncation

[11.8.0] - 2026-02-10

Added

Added ctx.reload() method to extension command context to reload extensions, skills, prompts, and themes from disk
Added ctx.ui.pasteToEditor() method to paste text into the editor with proper handling (e.g., large paste markers in interactive mode)
Added extension UI sub-protocol for RPC mode enabling dialog methods (select, confirm, input, editor) and fire-and-forget UI methods via client communication
Added support for tilde (~) expansion in custom skill directory paths
Added example extension demonstrating ctx.reload() usage with both command and LLM-callable tool patterns

Changed

Changed ctx.hasUI behavior: now true in RPC mode (previously false), with dialog methods working via extension UI sub-protocol
Changed warning output for invalid CLI arguments to use structured logging instead of console.error
Changed help text to indicate command-specific help is available via <command> --help
Changed tool result event handlers to chain like middleware, allowing each handler to see and modify results from previous handlers with partial patch support

Fixed

Fixed archive extraction security vulnerability by validating that extracted paths do not escape the extraction directory
Fixed archive format validation to reject unsupported formats before extraction attempt
Fixed archive extraction error handling to provide clear error messages on failure

[11.7.0] - 2026-02-07

Changed

Enhanced error messages for failed Python cells to include full combined output context instead of just the error message
Updated error cell output styling to use error color theme instead of standard tool output color for better visual distinction

Fixed

Improved error handling in Python cell execution to preserve and display combined output from previous cells when an error occurs
Fixed tab character rendering in Python tool output display to properly format whitespace in cell output and status events

[11.6.1] - 2026-02-07

Fixed

Fixed potential crash when rendering results with undefined details.results

[11.6.0] - 2026-02-07

Fixed

Fixed task tool renderer not sanitizing tabs, causing visual holes in TUI output
Fixed task tool expanded view showing redundant <swarm_context> block that is shared across all tasks
Fixed assistant message spacer appearing before tool executions when no visible content follows thinking block
Fixed extension runner emit() type safety with narrowed event/result types
Fixed extension runner tool_result event chaining across multiple extensions via dedicated emitToolResult()
Fixed queued messages not delivered after auto-compaction completes

Added

Added /quit slash command as alias for /exit
Added per-model overrides (modelOverrides) in models.json for customizing built-in model properties
Added mergeCustomModels to merge custom models with built-ins by provider+id instead of replacing

[11.5.2] - 2026-02-07

Fixed

Fixed TUI crash when ask tool renders long user input exceeding terminal width by using Text component for word wrapping instead of raw line output

[11.5.0] - 2026-02-06

Added

Added terminal breadcrumb tracking to remember the last session per terminal, enabling --continue to work correctly with concurrent sessions in different terminals

Changed

Changed screenshot format to always use PNG instead of supporting JPEG with quality parameter
Changed default extract_readable format from text to markdown
Changed screenshot storage to use temporary directory with Snowflake IDs instead of artifacts directory
Changed ResizedImage interface to return buffer as Uint8Array with lazy-loaded base64 data getter for improved memory efficiency

Removed

Removed JPEG quality parameter from screenshot options
Removed format selection for screenshots (now PNG only)
Removed ability to save screenshots to custom paths or artifacts directory

[11.4.1] - 2026-02-06

Fixed

Fixed tab character display in error messages and bash tool output by properly replacing tabs with spaces

[11.4.0] - 2026-02-06

Added

Visualize leading whitespace (indentation) in diff output with dim glyphs—tabs display as → and spaces as · for improved readability

Fixed

Fixed patch applicator to correctly handle context-only hunks (pure context lines between @@ markers) without altering indentation in tab-indented files
Fixed indentation conversion logic to infer tab width from space-to-tab patterns using linear regression (ax+b model) when pattern uses spaces and actual file uses tabs
Fixed tab character rendering in tool output previews and code cell displays, ensuring tabs are properly converted to spaces for consistent terminal display
Fixed newSession() to properly await session manager operations, ensuring new session is fully initialized before returning
Fixed session formatting to use XML structure for tools and tool invocations instead of YAML, improving compatibility with structured output parsing

[11.3.0] - 2026-02-06

Added

Added resumption hint printed to stderr on session exit showing command to resume the session (e.g., Resume this session with claude --resume <session-id>)
New BlobStore class for content-addressed storage of large binary data (images) externalized from session files
New getBlobsDir() function to get path to blob store directory
Support for externalizing large images to blob store during session persistence, reducing JSONL file size
New blob reference format (blob:sha256:<hash>) for tracking externalized image data in sessions
Exported ModeChangeEntry type for tracking agent mode transitions
Support for restoring plan mode state when resuming sessions
New appendModeChange() method in SessionManager to record mode transitions
New mode and modeData fields in SessionContext to track active agent mode
Support for PI_PACKAGE_DIR environment variable to override package directory (useful for Nix/Guix store paths)
New keybindings for session management: toggleSessionNamedFilter (Ctrl+N), newSession, tree, fork, and resume actions
Support for shell command execution in configuration values (API keys, headers) using ! prefix, with result caching
New clearOnShrink display setting to control whether empty rows are cleared when content shrinks
New SlashCommandInfo, SlashCommandLocation, and SlashCommandSource types for extension slash command discovery
New getCommands() method in ExtensionAPI to retrieve available slash commands
New switchSession() action in ExtensionCommandContext to switch between sessions
New SwitchSessionHandler type for extension session switching handlers
New getSystemPrompt() method in ExtensionUIContext to access current system prompt
New getToolsExpanded() and setToolsExpanded() methods in ExtensionUIContext for tool output expansion control
New WriteToolCallEvent type for write tool call events
New isToolCallEventType() type guard for tool call events
Support for image content in RPC steer and followUp commands
New GitSource type and parseGitUrl() function for parsing git URLs in plugin system
Tool input types exported: BashToolInput, FindToolInput, GrepToolInput, ReadToolInput, WriteToolInput
Support for @ prefix normalization in file paths (strips leading @ character)
New parentSessionPath field in SessionInfo to track forked session origins
Skill file relative path resolution against skill directory in system prompt
Support for Termux/Android package installation guidance for missing tools
Support for puppeteer query handlers (aria/, text/, xpath/, pierce/) in selector parameters across all browser actions
Automatic normalization of legacy p- prefixed selectors (p-aria/, p-text/, p-xpath/, p-pierce/) to modern puppeteer query handler syntax
Improved click action with intelligent element selection that prioritizes visible, actionable candidates and retries until timeout
Enhanced actionability checking for click operations, validating visibility, pointer events, opacity, viewport intersection, and element occlusion

Changed

Modified --resume flag to accept optional session ID or path (e.g., --resume abc123 or --resume /path/to/session.jsonl), with session picker shown when no value provided
Consolidated --session flag as an alias for --resume with value for improved CLI consistency
Removed read tool grouping reset logic that was breaking grouping when text or thinking blocks appeared between tool calls
Image persistence now externalizes images ≥1KB to content-addressed blob store instead of compressing inline
Session loading now automatically resolves blob references back to base64 image data
Session forking now resolves blob references in copied entries to ensure data integrity
Screenshot tool now automatically compresses images for API content using the same resizing logic as pasted images, reducing payload size while maintaining quality
Improved text truncation across tool renderers to respect terminal width constraints and prevent output overflow
Enhanced render caching to include width parameter for accurate cache invalidation when terminal width changes
HTML export filter now treats mode_change entries as settings entries alongside model changes and thinking level changes
Replaced ellipsis string (...) with Unicode ellipsis character (…) throughout UI text and truncation logic for improved typography
Improved render performance by introducing caching for tool output blocks and search results to avoid redundant text width and padding computations
Enhanced read tool grouping to reset when non-tool content (text/thinking blocks) appears between read calls, preventing unintended coalescing
Improved string preview formatting in scalar values to show line counts and truncation indicators for multi-line strings
Refactored tool execution component to use shared mutable render state for spinner frames and expansion state, reducing closure overhead
Enhanced error handling in tool renderers with logging for renderer failures instead of silent fallbacks
Made shell command execution in configuration values asynchronous to prevent blocking the TUI
Improved @ prefix normalization to only strip leading @ for well-known path syntaxes (absolute paths, home directory, internal URL shorthands) to avoid mangling literal paths
Enhanced git URL parsing to strip credentials from repository URLs and validate URL-encoded hash fragments
Improved null data handling in task submission to preserve agent output when submit_result is called with null/undefined data, enabling fallback text extraction instead of discarding output
Updated default model IDs across providers: Claude Sonnet 4.5 → Claude Opus 4.6, Gemini 2.5 Pro → Gemini 3 Pro variants, and others
Made model definition fields optional with sensible defaults for local models (Ollama, LM Studio, etc.)
Modified custom tool execute signature to reorder parameters: (toolCallId, params, signal, onUpdate, ctx) instead of (toolCallId, params, onUpdate, ctx, signal)
Changed --version and --list-models flags to exit with process.exit(0) instead of returning
Improved --export flag to exit with process.exit(0) on success
Enhanced tree selector to preserve last selected ID across filter changes
Modified tree navigation to use real leaf ID instead of skipping metadata entries
Improved footer path truncation logic to prevent invalid truncation at boundary
Enhanced model selector to display selected model name when no matches found
Improved RPC client steer() and followUp() methods to accept optional image content
Updated extension loader to check for explicit extension entries in root directory before discovering subdirectories
Removed line limiting in custom message component when collapsed
Improved API key resolution to support shell command execution via resolveConfigValue()
Enhanced session branching to preserve parent session path reference
Updated selector parameter descriptions to document support for CSS selectors and puppeteer query handlers
Modified viewport handling in headless mode to respect custom viewport parameters while disabling viewport in headed mode for better window management
Improved click action to use specialized text query handler logic with retry mechanism for better reliability with dynamic content

Fixed

Fixed background color stability in output blocks when inner content contains SGR reset sequences, preventing background color from being cleared mid-line
Fixed spurious ellipsis appended to output lines that were already padded to terminal width by trimming trailing spaces before truncation check
Fixed config file parsing to properly handle missing files instead of treating them as errors
Fixed truncation indicator in truncate tool to use ellipsis character (…) instead of verbose '[truncated]' suffix
Fixed concurrent shell command execution by de-duplicating in-flight requests for the same command
Fixed git URL parsing to properly handle URL-encoded characters in hash fragments and reject invalid encodings
Fixed task executor to properly handle agents calling submit_result with null data by treating it as missing and attempting to extract output from conversation text rather than silently failing
Fixed HTML export template to safely handle invalid argument types in tool rendering
Fixed path shortening in HTML export to handle non-string paths
Fixed custom message rendering to properly display full content without artificial line limits
Fixed tree navigation to only restore editor text when editor is empty
Fixed session creation to properly track parent session when forking
Fixed thinking level initialization to only append change entry for new sessions without existing thinking entries
Fixed tool expansion state management to properly propagate through UI context
Fixed click action to properly handle text/ query handlers with timeout and retry logic instead of failing immediately
Fixed viewport application to only apply when in headless mode or when explicitly requested, preventing conflicts in headed browser mode

Security

Added support for shell command execution in configuration values with caching to enable secure credential resolution patterns

[11.2.1] - 2026-02-05

Fixed

Fixed CLI invocation with flags only (e.g. pi --model=codex) to route to the default command instead of erroring

[11.2.0] - 2026-02-05

Added

Added omp commit command to generate commit messages and update changelogs with --push, --dry-run, --no-changelog, and model override flags
Added omp config command to manage configuration settings with actions: list, get, set, reset, path
Added omp grep command to test grep tool with pattern matching, glob filtering, context lines, and output modes
Added omp jupyter command to manage the shared Jupyter gateway with status and kill actions
Added omp plugin command to manage plugins with install, uninstall, list, link, doctor, features, config, enable, and disable actions
Added omp setup command to install dependencies for optional features like Python
Added omp shell command for interactive shell console with working directory and timeout configuration
Added omp stats command to view usage statistics with dashboard server, JSON output, and summary options
Added omp update command to check for and install updates with force and check-only modes
Added omp web-search command (alias omp q) to test web search providers with provider selection, recency filtering, and result limits
Migrated CLI from custom argument parser to oclif framework for improved command structure and help system
Added omp q CLI subcommand for testing web search providers with query, provider, recency, and limit options
Added web search provider information API with authentication requirements and provider metadata
Added support for hour recency filter option in Perplexity web search
Support for image file mentions—images are now automatically detected, resized, and attached when referenced with @filepath syntax
Image dimension information displayed in file mention UI to show image properties alongside text files

Changed

Refactored web search provider system to use individual provider classes in separate files for improved maintainability
Moved SearchProvider base class and SearchParams interface to dedicated providers/base.ts module
Updated web search execution to pass maxOutputTokens, numSearchResults, and temperature parameters to providers
Changed Perplexity search context size from 'high' to 'medium' and added search classifier, reasoning effort, and language preference settings
Increased Perplexity default max tokens from 4096 to 8192 for more comprehensive responses
Updated Anthropic and Gemini search providers to support max_tokens and temperature parameters for finer control over response generation
Simplified AuthStorage.create() to accept direct agent.db path
Renamed web search types and exports for consistency: WebSearchProvider → SearchProviderId, WebSearchResponse → SearchResponse, WebSearchTool → SearchTool, and related functions
Refactored web search provider system to use centralized provider registry with getSearchProvider() and resolveProviderChain() for improved provider management
Updated web search system prompt to emphasize comprehensive, detailed answers with concrete data and specific examples over brevity
Simplified Exa API key discovery to check environment variables only, removing .env file fallback logic
Refactored ModelRegistry instantiation to use direct constructor instead of discoverModels() helper function across codebase
Refactored CLI entry point to use oclif command framework instead of custom subcommand routing
Reorganized subcommands into individual command files under src/commands/ directory for better maintainability
Updated extension flag handling to parse raw arguments directly instead of using custom flag definitions
Refactored web search provider definitions into centralized provider-info module for better maintainability
Updated web search result rendering to support long-form answers with text wrapping in CLI mode
Removed related questions section from web search result rendering
Updated Perplexity API types to support extended message content formats including images, files, and PDFs
Updated Perplexity search to use 'pro' search type for improved search quality and relevance
File mention messages now support both text content and image attachments, with optional line count for text files
Updated file mention processing to respect image auto-resize settings

Removed

Removed legacy auth.json file—credentials are stored exclusively in agent.db

Fixed

Fixed type handling in model selector error message display to properly convert error objects to strings
Fixed web search to use search results when Perplexity API returns no citations, ensuring search results are always available to users
Fixed model switches deferred during streaming to apply correctly when the stream completes, preventing model changes from being lost
Fixed plan mode toggles during streaming to inject plan-mode context immediately, preventing file edits while in plan mode
Fixed plan mode model switches during streaming to defer model changes until the current turn completes

[11.1.0] - 2026-02-05

Added

Added sortDiagnostics() utility function to sort diagnostics by severity, location, and message for consistent output ordering
Added task.isolation.enabled setting to control whether subagents run in isolated git worktrees
Added dynamic task schema that conditionally includes isolated parameter based on isolation setting
Added openInEditor() utility function to centralize external editor handling with support for custom file extensions and stdio configuration
Added getEditorCommand() utility function to retrieve the user's preferred editor from $VISUAL or $EDITOR environment variables

Changed

Changed diagnostic output to sort results by severity (errors first), then by file location and message for improved readability
Changed task tool to validate isolation setting and reject isolated parameter when isolation is disabled
Changed task API to use assignment field instead of args for per-task instructions, with shared context prepended to every task
Changed task template rendering to use structured context/assignment separation with <swarm_context> wrapper instead of placeholder-based substitution
Changed task item schema to require assignment string (complete per-task instructions) instead of optional args object
Changed TaskItem to remove args field and add assignment field for clearer per-task instruction semantics
Changed agent frontmatter to use thinking-level field name instead of thinkingLevel for consistency
Refactored task rendering to display full task text instead of args in progress and result views
Changed SubmenuSettingDef.getOptions() method to options getter property for cleaner API access
Converted static option providers from functions to direct array definitions for improved performance
Added createSubmenuSettingDef() helper function to support both static and dynamic option providers
Modified setThinkingLevel() API to accept optional persist parameter (defaults to false) for controlling whether thinking level changes are saved to settings
Refactored hook editor and input controller to use shared external editor utilities, reducing code duplication

Removed

Removed context parameter from ExecutorOptions — context now prepended at template level before task execution
Removed args field from AgentProgress and SingleResult interfaces
Removed placeholder-based template rendering in favor of structured context/assignment model

[11.0.3] - 2026-02-05

Added

Added new subcommands to help text: commit for AI-assisted git commits, stats for AI usage statistics dashboard, and jupyter for managing the shared Jupyter gateway
Added grep subcommand to help text for testing the grep tool
Added browser tool documentation for browser automation using Puppeteer
Added todo_write tool documentation for managing todo and task lists
Added documentation for additional LLM provider API keys (Groq, Cerebras, xAI, OpenRouter, Mistral, z.ai, MiniMax, OpenCode, Cursor, Vercel AI Gateway) in environment variables reference
Added documentation for cloud provider configuration (AWS Bedrock, Google Vertex AI) in environment variables reference
Added documentation for search provider API keys (Perplexity, Anthropic Search) in environment variables reference
Added documentation for model override environment variables (PI_SMOL_MODEL, PI_SLOW_MODEL, PI_PLAN_MODEL) in CLI help text
Added comprehensive environment variables reference documentation at docs/environment-variables.md covering API keys, configuration, debugging, and testing variables
Added theme system with 44 customizable color tokens, two built-in themes (dark/light), and auto-detection based on terminal background
Added /theme command to interactively select and switch between themes
Added support for custom themes in ~/.pi/agent/themes/*.json with live editing - changes apply immediately when files are saved
Added userMessageText theme token for customizing user message text color
Added toolTitle and toolOutput theme tokens for separate coloring of tool execution box titles and output

Changed

Updated help text to reflect expanded tool availability - default now enables all tools instead of just read, bash, edit, write
Updated available tools list in help documentation to include python, notebook, task, fetch, web_search, browser, and ask
Simplified main description in help text from 'AI coding assistant with read, bash, edit, write tools' to 'AI coding assistant'
Updated --tools option documentation to clarify default behavior and list all available tools
Changed all environment variable access from process.env to Bun.env throughout the codebase for Bun runtime compatibility
Updated documentation to reference Bun.env instead of process.env in examples and comments

Fixed

Fixed Text component to properly implement invalidate() method, ensuring theme changes apply correctly to all UI elements
Fixed TruncatedText component to properly pad all lines to exactly match the specified width, preventing rendering artifacts
Fixed TruncatedText component to stop at the first newline and only display the first line
Fixed invalid or malformed themes to fall back gracefully to dark theme instead of crashing the application

[11.0.2] - 2026-02-05

Fixed

Fixed role model cycling to expand role aliases (e.g., roles pointing at pi/plan) so slow/default/smol cycles resolve correctly

[11.0.0] - 2026-02-05

Added

Added UI dropdown options for task.maxRecursion Depth setting with presets (Unlimited, None, Single, Double, Triple)
Added UI dropdown options for grep.contextBefore setting with presets (0-5 lines)
Added UI dropdown options for grep.contextAfter setting with presets (0-10 lines)
Added task.maxRecursionDepth setting to control how many levels deep subagents can spawn their own subagents (0=none, 1=one level, 2=two levels, -1=unlimited)
Added support for nested task artifact naming with parent task prefixes (e.g., "0-Auth.1-Subtask") to organize hierarchical task outputs
Added taskDepth and parentTaskPrefix options to CreateAgentSessionOptions for tracking subagent recursion depth and organizing nested artifacts
Added task.maxConcurrency setting to control concurrent limit for subagents (default: 32)
Added UI options for task concurrency configuration with presets from unlimited to 64 tasks
Added support for loading skills from ~/.agents/skills

Changed

Simplified task.maxRecursionDepth description in settings UI to remove specific value examples
Made thinking level persistence optional via persist parameter in setThinkingLevel() method, allowing temporary thinking level changes without saving to settings
Updated thinking level cycling to no longer persist changes to settings, enabling quick iteration through thinking levels without modifying user preferences
Replaced nanoid with Snowflake for ID generation throughout codebase for improved performance and collision resistance
Updated session ID format in documentation from nanoid to snowflake hex string (e.g., "a1b2c3d4e5f60001")
Renamed environment variable prefix from OMP_ to PI_ throughout codebase (e.g., OMP_DEBUG_STARTUP → PI_DEBUG_STARTUP, OMP_PYTHON_GATEWAY_URL → PI_PYTHON_GATEWAY_URL)
Removed env setting from configuration schema; environment variables are no longer automatically applied from settings
Changed venvPath property in PythonRuntime from nullable to optional (returns undefined instead of null)
Simplified notification settings from protocol-specific options (bell, osc99, osc9) to simple on/off toggle for completion.notify and ask.notify
Moved notification protocol detection and sending to TERMINAL API from local utility functions
Changed task tool spawns configuration from "explore" to "*" to allow subagents to spawn any agent type
Changed system prompt to enable parallel delegation guidance for all agents (removed coordinator-only restriction)
Changed task tool to automatically disable itself when maximum recursion depth is reached, preventing infinite nesting
Changed task concurrency from hardcoded constant to configurable setting via task.maxConcurrency
Changed concurrency limit calculation to support unlimited concurrency when set to 0

Removed

Removed nanoid dependency from package.json
Removed terminal-notify.ts utility module with detectNotificationProtocol(), sendNotification(), and isNotificationSuppressed() functions
Removed MAX_PARALLEL_TASKS constant and associated task count validation limit

Fixed

Fixed MCP tool name generation to properly sanitize server and tool names, preventing invalid characters and duplicate prefixes in tool identifiers
Fixed task ID display formatting to show hierarchical structure for nested tasks (e.g., "0.1 Auth>Subtask" instead of "0-Auth.1-Subtask")
Improved frontmatter parsing error messages to include source context for better debugging

[10.6.1] - 2026-02-04

Added

Added commit model role for dedicated commit message generation
Exported resolveModelOverride function from model resolver for external use

Changed

Updated model role resolution to accept optional roleOrder parameter for custom role priority
Made tag and color properties optional in ModelRoleInfo interface
Updated model selector to safely handle roles without tag or color definitions
Refactored role label display to use centralized MODEL_ROLES registry instead of hardcoded strings
Refactored model role system to use centralized MODEL_ROLES registry with consistent tag, name, and color definitions
Simplified model role resolution to use MODEL_ROLE_IDS array instead of hardcoded role checks
Updated model selector to dynamically generate menu actions from MODEL_ROLES registry

Removed

Removed support for omp/ model role prefix; use pi/ prefix instead

[10.6.0] - 2026-02-04

Breaking Changes

Removed output_mode parameter from grep tool—results now always use content mode with formatted match output
Renamed grep context parameters from context_pre/context_post to pre/post
Removed n (show line numbers) parameter—line numbers are now always displayed in grep results

Added

Added Jina as a web search provider option alongside Exa, Perplexity, and Anthropic
Added support for Jina Reader API integration with automatic provider detection when JINA_API_KEY is configured

Changed

Reformatted grep output to display matches grouped by file with numbered match headers and aligned context lines
Updated grep output to use >> prefix for match lines and aligned spacing for context lines for improved readability
Changed multiline matching to automatically enable when pattern contains literal newlines ( )
Split grep context parameter into separate context_pre and context_post options for independent control of lines before and after matches
Updated grep tool to use configurable default context settings from grep.contextBefore and grep.contextAfter configuration
Added configurable grep context defaults and reduced the default to 1 line before, 3 lines after
Enabled the browser tool by default

Removed

Removed filesWithMatches and count output modes from grep tool

[10.5.0] - 2026-02-04

Breaking Changes

Changed ask tool to require questions array parameter; single-question mode with question, options, multi, and recommended parameters is no longer supported
Removed support for local Python kernel gateway startup; shared gateway is now required

Added

Added browser tool powered by Ulixee Hero with support for navigation, DOM interaction, screenshots, and readable content extraction
Added /browser command to toggle browser headless vs visible mode in interactive sessions
Added browser.enabled and browser.headless settings to control browser automation behavior
Added Python prelude caching to improve startup performance by storing compiled prelude helpers and module metadata
Added OMP_DEBUG_STARTUP environment variable for conditional startup performance debugging output
Added autonomous memory system with storage, memory tools, and context injection

Changed

Updated task tool guidance to enforce small, well-defined task scope with maximum 3-5 files per task to prevent timeouts and improve parallel execution
Updated browser viewport to use 1.25x device scale factor for improved rendering on high-DPI displays
Modified device pixel ratio detection to respect actual screen capabilities instead of forcing 1x ratio
Updated system prompt guidance to state assumptions and proceed without asking for confirmation, reducing unnecessary round-trips
Tightened ask tool conditions to require multiple approaches with significantly different tradeoffs before prompting user
Strengthened ask tool guidance to default to action and only ask when genuinely blocked by decisions with materially different outcomes
Changed refactor workflow to automatically remove now-unused elements and note removals instead of asking for confirmation
Enforced exclusive concurrency mode for all file-modifying tools (edit, write, bash, python, ssh, todo-write) to prevent concurrent execution conflicts
Updated ask tool guidance to prioritize proactive problem-solving and default to action, asking only when truly blocked by decisions that materially change scope or behavior
Changed Python kernel initialization to require shared gateway mode; local gateway startup has been removed
Changed shared gateway error handling to retry on server errors (5xx status codes) before failing

Fixed

Fixed glob search returning no results when all files are ignored by gitignore by automatically retrying without gitignore filtering

[10.3.2] - 2026-02-03

Added

Added renderCall and renderResult methods to MCP tools for structured TUI display of tool calls and results
Added new mcp/render.ts module providing JSON tree rendering for MCP tool output with collapsible/expandable views

Changed

Updated renderResult signature in custom tools and extensions to accept optional args parameter for context-aware rendering
Changed environment variable from ENV_AGENT_DIR constant to hardcoded OMP_CODING_AGENT_DIR string in config and CLI help text
Fixed method binding in extension and hook tool wrappers to preserve this context for renderCall and renderResult methods

[10.3.1] - 2026-02-03

Fixed

Fixed timeout handling in LSP write-through operations to properly clear formatter and diagnostics results when operations exceed the 10-second timeout

[10.3.0] - 2026-02-03

Removed

Removed shellForceBasic setting that forced bash/sh shell selection
Removed bash.persistentShell experimental setting for shell session reuse

[10.2.3] - 2026-02-02

Added

Added find.enabled, grep.enabled, ls.enabled, notebook.enabled, fetch.enabled, web_search.enabled, lsp.enabled, and calc.enabled settings to control availability of individual tools
Added conditional tool documentation in system prompt that dynamically lists only enabled specialized tools
Added todos.enabled setting to control availability of the todo_write tool for task tracking
Added tools field to agent frontmatter for declaring agent-specific tool capabilities

Changed

Consolidated symbols action to handle both file-based document symbols and workspace symbol search (query-based)
Consolidated diagnostics action to handle both single-file and workspace-wide diagnostics (no file = workspace)
Simplified reload action to gracefully reload language server with fallback to kill
Updated LSP tool documentation to reflect simplified operation set and consolidated actions
Reorganized settings tabs from 8 to 9 tabs with clearer categorization: Display, Agent, Input, Tools, Config, Services, Bash, LSP, and TTSR
Moved behavior-related settings to new Agent tab for better organization
Moved input/interaction settings to new Input tab
Moved tool configuration settings to new Config tab
Moved provider and service settings to new Services tab
Added visual icons to settings tabs using theme symbols for improved UI clarity
Changed default settings tab from Behavior to Display on startup
Updated read tool to handle directory paths by returning formatted listings with modification times instead of redirecting to ls
Updated tool documentation to reflect that read now handles both files and directories
Updated system prompt tool precedence section to conditionally display only available specialized tools based on enabled settings
Renamed todo completion settings from todoCompletion.* to todos.reminders.* and todos.enabled for clearer organization
Updated todo reminder logic to check both todos.reminders and todos.enabled settings independently

Removed

Removed Rust-analyzer specific LSP operations: flycheck, expand_macro, ssr, runnables, related_tests, and reload_workspace
Removed workspace_diagnostics action; use diagnostics without file parameter instead
Removed workspace_symbols action; use symbols with query parameter and no file instead
Removed actions, incoming_calls, and outgoing_calls LSP operations
Removed replacement, kind, action_index, end_line, and end_character parameters from LSP tool
Removed Python prelude helper functions: pwd(), mkdir(), ls(), head(), tail(), sh(), cat(), touch(), wc(), basenames(), and batch()
Removed type guard functions (isBashToolResult, isReadToolResult, isEditToolResult, isWriteToolResult, isGrepToolResult, isFindToolResult, isLsToolResult) from public API exports
Removed ls tool—directory listing is now handled by the read tool
Removed ls.enabled setting and related configuration options
Removed bashInterceptor.simpleLs setting that redirected simple ls commands to the dedicated tool
Removed project tree snapshot generation from system prompt (unused feature)

Fixed

Fixed tool parameter schemas displaying internal TypeBox metadata fields in system prompt

[10.2.1] - 2026-02-02

Breaking Changes

Removed strippedRedirect field from NormalizedCommand interface returned by normalizeBashCommand()
Removed automatic stripping of 2>&1 stderr redirections from bash command normalization

[10.1.0] - 2026-02-01

Added

Added work scheduling profiler to debug menu for analyzing CPU scheduling patterns over the last 30 seconds
Added support for work profile data in report bundles including folded stacks, summary, and flamegraph visualization

[10.0.0] - 2026-02-01

Added

Added shell subcommand for interactive shell console testing with brush-core
Added --cwd / -C option to set working directory for shell commands
Added --timeout / -t option to configure per-command timeout in milliseconds
Added --no-snapshot option to skip sourcing snapshot from user shell

Fixed

find now returns a single match when given a file path instead of failing with "not a directory"

[9.8.0] - 2026-02-01

Breaking Changes

Removed persistent shell session support; bash execution now uses native bindings via brush-core for improved reliability

Added

Added sessionKey option to bash executor to isolate shell sessions per agent instance
Added shell snapshot support for bash execution to preserve shell state across commands
Added onChunk callback support for streaming command output in real-time

Changed

Refactored bash executor to queue output chunks asynchronously for improved reliability
Updated bash executor to pass environment variables separately as sessionEnv to native bindings
Migrated system information collection to use native bindings from brush-core instead of shell command execution
Updated CPU information to report core count alongside model name
Simplified OS version reporting to use Node.js built-in APIs
Migrated bash command execution from ptree-based persistent sessions to native shell bindings with streaming support
Simplified bash executor to use brush-core native API instead of managing long-lived shell processes
Routed clipboard copy and image paste through native arboard bindings instead of shell commands
Embedded native addon payload for compiled binaries and extract to ~/.omp/natives/<version> on first run

Removed

Removed shell configuration from environment information display
Removed shell-session.ts module providing persistent shell session management
Removed shell session test suite for persistent execution patterns

[9.6.2] - 2026-02-01

Changed

Replaced hardcoded ellipsis strings with Unicode ellipsis character (…) throughout rendering code
Removed format.ellipsis symbol from theme configuration; ellipsis now uses literal Unicode character
Updated truncate() function to truncateToWidth() with simplified API accepting default ellipsis parameter
Simplified formatMoreItems() function signature by removing theme parameter dependency

Removed

Removed format.ellipsis symbol key from theme symbol maps (Unicode, Nerd, and ASCII presets)
Removed ellipsis property from SymbolTheme type

[9.6.1] - 2026-02-01

Fixed

Fixed output handling to prioritize text/markdown over text/plain when both are available, ensuring Markdown content is displayed correctly
Fixed bash command normalization to preserve newlines in heredocs and multiline commands

[9.6.0] - 2026-02-01

Breaking Changes

Replaced SettingsManager class with new Settings singleton providing sync get/set API with background persistence
Changed settings access from method calls (e.g., getTheme()) to path-based access (e.g., settings.get("theme"))
Removed settingsManager parameter from CreateAgentSessionOptions in favor of settingsInstance
Removed loadSettings() export from public API
Removed example file examples/sdk/10-settings.ts demonstrating old SettingsManager API

Added

New Settings singleton class with sync get/set operations and background persistence
Added Settings.isolated() factory for creating isolated settings instances in tests
Added Settings.init() for initializing global settings instance
Added settings global export for convenient access to settings singleton
New settings-schema.ts providing unified, type-safe settings definitions with UI metadata
Added "none" option to doubleEscapeAction setting to disable double-escape behavior entirely (#973 by @juanibiapina)

Changed

Unified settings schema into single source of truth with settings-schema.ts replacing scattered definitions
Refactored settings CLI to use new schema-based path resolution instead of SETTINGS_DEFS
Updated config command examples to use new nested path syntax (e.g., compaction.enabled instead of autoCompact)
Changed InteractiveModeContext.settingsManager to InteractiveModeContext.settings
Updated all internal settings access throughout codebase to use new settings.get() and settings.set() API
Moved DEFAULT_BASH_INTERCEPTOR_RULES from settings-manager to bash-interceptor module

Removed

Deleted settings-manager.ts (2035 lines) - functionality replaced by new Settings singleton
Removed SettingsManager.create(), SettingsManager.acquire(), and SettingsManager.inMemory() factory methods
Removed individual getter/setter methods from settings API (e.g., getTheme(), setTheme(), getCompactionSettings())

Fixed

Respect .gitignore, .ignore, and .fdignore files when scanning package resources for skills, prompts, themes, and extensions

[9.5.1] - 2026-02-01

Changed

Changed persistent shell from opt-out to opt-in (default: off) for improved reliability; enable via Settings > Bash > Persistent shell or OMP_SHELL_PERSIST=1
Added new "Bash" settings tab grouping shell-related settings (force basic shell, persistent shell, interceptor, intercept ls)

[9.5.0] - 2026-02-01

Added

Added head and tail parameters to bash tool to limit output lines without breaking streaming
Added automatic normalization of bash commands to extract | head -n N and | tail -n N patterns into native parameters
Added maxResults parameter to find tool to limit result set at the native layer
Added context-structure template showing required sections (Goal, Constraints, Existing Code, API Contract) with examples of good vs bad context
Added explicit dependency test: 'Can agent B write correct code without seeing agent A's output?' to determine sequencing
Added detailed phased execution pattern with four phases (Foundation, Parallel Implementation, Integration, Dependent Layer) and WASM-to-N-API migration example
Added table of dependency patterns that must be sequential (API creation before bindings, interface definition before implementation, etc.)
Added phased execution guidance for migrations and refactors to prevent parallel work on dependent layers
Added example demonstrating phased execution pattern for porting WASM to N-API with sequential foundation, parallel implementation, integration, and dependent layer phases

Changed

Improved find tool performance by delegating mtime-based sorting to native layer instead of post-processing results in JavaScript
Simplified find tool result processing by removing redundant filesystem stat calls when native metadata is available
Updated bash tool documentation to recommend using head and tail parameters instead of piping through head/tail commands
Updated binary build process to exclude worker files from compilation, reducing binary size
Modified update mechanism to download and install native addon alongside CLI binary for platform-specific functionality
Updated find tool to emit streaming match updates via callback, allowing real-time progress feedback during file searches
Modified find tool to use native match metadata (mtime, fileType) from WASM layer instead of redundant filesystem stats, improving performance
Restructured Task tool documentation to emphasize context quality and explicit API contracts for subagent success
Updated task execution guidance to require structured context with Goal, Constraints, Existing Code, and API Contract sections
Reorganized parallelization rules with explicit dependency patterns and phased execution guidance for migrations
Clarified that response format requirements must go in schema parameter, never in context descriptions
Centralized Python runtime resolution into shared ipy/runtime.ts module, removing duplicate code from kernel and gateway coordinator

Removed

Removed Nushell language server configuration from LSP defaults

Fixed

Fixed race condition in shell session where command completion could occur before stream data was fully processed
Fixed Python gateway spawning console window on Windows by using windowless Python interpreter (pythonw.exe)

[9.4.0] - 2026-01-31

Changed

Migrated environment variable handling to use centralized getEnv() and getEnvApiKey() utilities from pi-ai package for consistent API key resolution across web search providers and image tools
Simplified web search error messages to remove provider-specific configuration hints
Replaced manual space padding with padding() utility function from pi-tui across UI components for consistent whitespace handling
Improved rendering performance for Python cell output by implementing caching in the table and cell results renderers
Updated task tool documentation to clarify that subagents can access parent conversation context via a searchable file, reducing need to repeat information in context parameter
Updated plan mode prompt to guide model toward using edit tool for incremental plan updates instead of defaulting to write

Removed

Removed environment variable denylist that blocked API keys from being passed to subprocesses; API keys are now controlled via allowlist only

[9.3.1] - 2026-01-31

Added

Added getCompactContext() API to retrieve parent conversation context for subagents, excluding system prompts and tool results
Added automatic submit_result tool injection for subagents with explicit tool lists
Added contextFile parameter to pass parent conversation context to subagent sessions

Changed

Updated subagent system prompt to reference parent conversation context file when available
Enhanced subagent system prompt formatting with clearer backtick notation for tool and parameter names

Removed

Removed schema override notification from task summary prompt

[9.2.5] - 2026-01-31

Changed

Clarified that user instructions about delegation override tool-use defaults
Updated coordinator guidance to emphasize Task tool preference for substantial work with improved emphasis on context window limitations
Enhanced context parameter documentation to require self-contained information for subagents, including file contents and user requirements

[9.2.4] - 2026-01-31

Fixed

Prevented interactive commands from blocking on stdin by redirecting from /dev/null in POSIX and Fish shell sessions

[9.2.3] - 2026-01-31

Added

Persistent shell session support for bash tool with environment variable preservation across commands
New shellForceBasic setting to force bash/sh even if user's default shell is different (default: true)
New OMP_SHELL_PERSIST environment variable to control persistent shell behavior (set to 0 to disable)

Changed

Bash tool now reuses a persistent shell session by default on Unix systems for improved performance and state preservation
Replaced Bun file APIs with Node.js fs module for better cross-runtime compatibility
LSP configuration loading is now synchronous instead of async
Shell snapshot generation now sanitizes BASH_ENV and ENV variables to prevent shell exit issues
Shell snapshot caching now per-shell-binary instead of global to avoid cross-shell contamination
System prompt restructured with coordinator-specific guidance for parallel task delegation
Bash tool now reuses a persistent shell session by default on Unix. Set OMP_SHELL_PERSIST=0 to disable or fall back to per-command execution on Windows/unsupported shells.
Added a shellForceBasic setting to force bash/sh and keep environment changes across bash commands (default: true).

Fixed

Shell snapshots now filter unsafe bash options (onecmd, monitor, restricted) to prevent session exits
Git branch detection in status line now works synchronously without race conditions
Shell session initialization properly restores trap handlers and shell functions after command execution
Sanitized BASH_ENV/ENV during persistent shell startup and snapshot creation to prevent basic shells from exiting immediately.
Cached shell snapshots per shell binary to avoid sourcing zsh snapshots in bash sessions.
Filtered unsafe bash options (onecmd/monitor/restricted) out of shell snapshots to prevent session exits.

[9.2.2] - 2026-01-31

Added

Added grep CLI subcommand (omp grep) for testing pattern matching
Added fuzzy matching for model resolution with scoring and ranking fallback
Added 'Open: artifact folder' menu option to debug selector for quick access to session artifacts
Added Kimi API format setting for selecting between OpenAI and Anthropic formats
Added Codex and Gemini web search providers with OAuth and grounding support
Added /debug command with interactive menu for profiling, heap snapshots, session dumps, and diagnostics
Added configurable ask timeout and notification settings
Added gitignore-aware project tree scanning with ripgrep integration
Added project tree visualization to system prompts with configurable depth and entry limits
Added reset() method to CountdownTimer with integration into HookSelectorComponent
Added custom message support to AgentSession via promptCustomMessage() method
Added skill message component for rendering /skill command messages as compact entries
Added model preference matching system for intelligent model selection based on usage history
Added designer agent with UI/UX review and accessibility audit capabilities
Added model-specific edit variant configuration for patch/replace modes
Added automatic browser opening when stats dashboard starts
Added model statistics table and TTFT/throughput metrics to stats dashboard
Added artifact allocation for truncated fetch responses to preserve full content
Added 30-second timeout to ask tool with auto-selection of recommended option
Added recommended parameter (0-indexed) to ask tool for specifying default option
Added JTD to TypeScript converter for rendering schemas in system prompts
Added tools list to system prompt for better agent awareness
Added synthetic message flag for system-injected prompts
Added session compaction enhancements with auto-continue, tool pruning, and remote endpoint support
Added detection and rendering of missing complete tool warning in subagent output
Added outline UI components for bordered list containers
Added macOS NFD normalization and curly quote variant resolution for file paths
Enhanced session compaction with dynamic token ratio adjustment and improved summary preservation

Changed

Simplified find tool API by consolidating path and pattern parameters
Replaced bulk file loading with streaming for read tool to reduce memory overhead
Migrated grep and find tools to WASM-based implementation
Replaced ripgrep-based file listing with glob-based file discovery for project scans
Updated minimum Bun runtime requirement to >=1.3.7
Renamed task parameter from output to schema
Renamed complete tool to submit_result for clarity and consistency
Improved output preview logic: shows full output for ≤30 lines, truncates to 10 lines for larger output

Fixed

Enhanced error reporting with debug stack trace when DEBUG env is set
Improved OAuth token refresh error handling to distinguish transient vs definitive failures
Added windowsHide option to child process spawn calls to prevent console windows on Windows
External edits to config.yml are now preserved when omp reloads or saves settings
Exposed LSP server startup errors in session display and logs
Improved error handling and security in agent storage initialization with restrictive file permissions
Fixed LSP server display showing unknown when server warmup fails
Preserved null timeout when user disables ask timeout setting
Removed incorrect timeout unit conversion logic in cursor, fetch, gemini-image, and ssh tools
Blocked /fork command while streaming to prevent split session logs

[9.0.0] - 2026-01-29

Fixed

External edits to config.yml are now preserved when omp reloads or saves unrelated settings. Previously, editing config.yml directly (e.g., removing a package from packages array) would be silently reverted on next omp startup when automatic setters like setLastChangelogVersion() triggered a save. (#1046 by @nicobailonMD)

[8.13.0] - 2026-01-29

Added

Added /debug command with interactive menu for bug report generation:
- Report: performance issue - CPU profiling with reproduction flow
- Report: dump session - Immediate session bundle creation
- Report: memory issue - Heap snapshot with bundle
- View: recent logs - Display last 50 log entries
- View: system info - Show environment details
- Clear: artifact cache - Remove old session artifacts

Fixed

Fixed LSP server errors not being visible in /session output or logs when startup fails

[8.12.7] - 2026-01-29

Fixed

Fixed LSP servers showing as "unknown" in status display when server warmup fails
Fixed Read tool loading entire file into memory when offset/limit was specified

[8.12.2] - 2026-01-28

Changed

Replaced ripgrep-based file listing with fs.glob for project scans and find/read tooling

[8.11.14] - 2026-01-28

Changed

Rendered /skill command messages as compact skill entries instead of full prompt text

[8.8.8] - 2026-01-28

Added

Added /fork command to create a new session with the exact same state (entries and artifacts) as the current session

Changed

Renamed the complete tool to submit_result for subagent result submission

[8.6.0] - 2026-01-27

Added

Added plan model role for specifying the model used by the plan agent
Added --plan CLI flag and OMP_PLAN_MODEL environment variable for ephemeral plan model override
Added plan model selection in model selector UI with PLAN badge

Changed

Task tool subagents now execute in-process instead of using worker threads

Fixed

Queued skill commands as follow-ups when the agent is already streaming to avoid load failures
Deduplicated repeated review findings in subagent progress rendering
Restored MCP proxy tool timeout handling to prevent subagent hangs

[8.5.0] - 2026-01-27

Added

Added subagent support for preloading skill contents into the system prompt instead of listing available skills
Added session init entries to capture system prompt, task, tools, and output schema for subagent session logs

Fixed

Reduced Task tool progress update overhead to keep the UI responsive during high-volume streaming output
Fixed subagent session logs dropping pre-assistant entries (user/task metadata) before the first assistant response

Removed

Removed enter-plan-mode tool

[8.4.5] - 2026-01-26

Added

Model usage tracking to record and retrieve most recently used models
Model sorting in selector based on usage history

Changed

Renamed head_limit parameter to limit in grep and find tools for consistency
Added context as an alias for the c context parameter in grep tool
Made hidden files inclusion configurable in find tool via hidden parameter (defaults to true)
Added support for reading ignore patterns from .gitignore and .ignore files in find tool

Fixed

Respected .gitignore rules when filtering find tool results by glob pattern

[8.4.2] - 2026-01-25

Changed

Clarified and condensed plan mode prompts for improved clarity and consistency

[8.4.1] - 2026-01-25

Added

Added core plan mode with plan file approval workflow and tool gating
Added plan:// internal URLs for plan file access and subagent plan-mode system prompt
Added plan mode toggle shortcut with paused status indicator

Fixed

Fixed plan reference injection and workflow prompt parameters for plan mode
Fixed tool downloads hanging on slow/blocked GitHub by adding timeouts and zip extraction fallback
Fixed missing UI notification when tools are downloaded or installed on demand

[8.4.0] - 2026-01-25

Added

Added extension API to set working/loading messages during streaming
Added task worker propagation of context files, skills, and prompt templates
Added subagent option to skip Python preflight checks when Python tooling is unused
Model field now accepts string arrays for fallback model prioritization

Changed

Merged patch application warnings into edit tool diagnostics output
Cached Python prelude docs for subagent workers to avoid repeated warmups
Simplified image placeholders inserted on paste to match Claude-style markers

Fixed

Rewrote empty or corrupted session files to restore valid headers
Improved patch applicator ambiguity errors with match previews and overlap detection
Fixed Task tool agent model resolution to honor comma-separated model lists

[8.3.0] - 2026-01-25

Changed

Added request parameter tracking to LSP tool rendering for better diagnostics visibility
Added async diff computation and Kitty protocol support to tool execution rendering
Refactored patch applicator with improved fuzzy matching (7-pass sequence matching with Levenshtein distance) and indentation adjustment
Added inline rendering flag to bash and fetch tool renderers
Extracted constants for preview formatting to improve code maintainability
Exposed mergeCallAndResult and inline rendering options from tools to their wrappers
Added timeout validation and normalization for tool timeout parameters

Fixed

Fixed output block border rendering (bottom-right corner was missing)
Added background control parameter to output block rendering

[8.2.2] - 2026-01-24

Removed

Removed git utility functions (_git, git_status, git_diff, git_log, git_show, git_file_at, git_branch, git_has_changes) from IPython prelude

[8.2.0] - 2026-01-24

Added

Added omp commit command to generate conventional commits with changelog updates
Added agentic commit mode with commit-specific tools and --legacy fallback
Added configurable settings for map-reduce analysis including concurrency, timeout, file thresholds, and token limits
Added support for excluding YAML lock files (.lock.yml, .lock.yaml, -lock.yml, -lock.yaml) from commit analysis
Added new TUI component library with reusable rendering utilities including code cells, file lists, tree lists, status lines, and output blocks
Added renderCodeCell component for displaying code with optional output sections, supporting syntax highlighting and status indicators
Added renderFileList component for rendering file/directory listings with language icons and metadata
Added renderTreeList component for hierarchical tree-based item rendering with expand/collapse support
Added renderStatusLine component for standardized tool status headers with icons, descriptions, and metadata
Added renderOutputBlock component for bordered output containers with structured sections
Added renderOutputBlock to Bash tool for improved output formatting with status indicators
Added --legacy flag to omp commit for using the deterministic pipeline instead of agentic mode
Added split commit support to automatically create multiple atomic commits for unrelated changes
Added git hunk inspection tools for fine-grained diff analysis in commit generation
Added commit message validation with filler word and meta phrase detection
Added automatic unicode normalization in commit summaries
Added real-time progress output to agentic commit mode showing thinking status, tool calls, and completion summary
Added hunk-level staging support in split commits allowing partial file changes per commit
Added dependency ordering for split commits ensuring commits are applied in correct sequence
Added circular dependency detection with validation errors for split commit plans
Added parallel file analysis with cross-file context awareness via analyze_files tool
Added AGENTS.md context file discovery for commit generation
Added progress indicators during changelog generation and model resolution
Added propose_changelog tool for agent-provided changelog entries in agentic commit workflow
Added fallback commit generation when agentic mode fails, using file pattern analysis and heuristic-based type inference
Added trivial change detection to automatically classify whitespace-only and import-reorganization commits
Added support for pre-computed file observations in commit agent to skip redundant analyze_files calls
Added diff content caching with smart file prioritization to optimize token usage in large changesets
Added lock file filtering (17 patterns including Cargo.lock, package-lock.json, bun.lock) from commit analysis
Added changelog deletion support to remove outdated entries via the changelog proposal interface
Added support for pre-computed changelog entries in commit agent to display existing unreleased sections for potential deletion
Added ExistingChangelogEntries interface to track changelog sections by path for changelog proposal context
Added conditional analyze_files skipping in commit agent when pre-analyzed observations are provided
Added guidance to commit agent prompts instructing subagents to write files directly instead of returning changes for manual application
Added mermaid diagram rendering with terminal graphics support (Kitty/iTerm2) for markdown output
Added renderMermaidToPng utility for converting mermaid code blocks to terminal-displayable PNG images via mmdc CLI
Added mermaid block extraction with content-addressed hashing for deduplication and cache lookup
Added background mermaid pre-rendering in assistant messages for responsive diagram display
Added two-level mermaid caching with pending deduplication to prevent redundant renders
Added Python kernel session pooling with MAX_KERNEL_SESSIONS limit and automatic eviction of oldest sessions
Added automatic idle kernel session cleanup timer (5-minute timeout, 30-second interval)
Added types/assets/index.d.ts for global TypeScript module declarations supporting .md, .py, and .wasm?raw imports
Added bunfig.toml loader configuration for importing markdown, Python, and WASM files as text modules
Added color manipulation utilities (hexToHsv, hsvToHex, shiftHue) to pi-utils for accessible theme adjustments
Added color-blind mode setting for improved accessibility
Added filesystem error type guards (isEnoent, isEacces, isPerm, isEnotempty, isFsError, hasFsCode) to pi-utils for safe error handling
Added tarball installation test Dockerfile to validate npm publish/install flow

Changed

Changed changelog diff truncation limit to be configurable via settings
Changed tool result rendering to use new TUI component library across multiple tools (bash, calculator, fetch, find, grep, ls, notebook, python, read, ssh, write, lsp, web search) for consistent output formatting
Changed Bash tool output rendering to use renderOutputBlock with proper section handling and width-aware truncation
Changed Python tool output rendering to use renderCodeCell component for code cell display with status indicators
Changed Read tool output rendering to use renderCodeCell with syntax highlighting and warnings display
Changed Write tool output rendering to use renderCodeCell for code display with streaming preview support
Changed Fetch tool output rendering to use renderOutputBlock with metadata and content preview sections
Changed LSP tool output rendering to use renderStatusLine and renderOutputBlock for structured output display
Changed Web Search result rendering to use renderOutputBlock with answer, sources, related questions, and metadata sections
Changed Find, Grep, and Ls tools to use renderFileList and renderTreeList for consistent file/item listing
Changed Calculator tool result rendering to use renderTreeList for result item display
Changed Notebook and TodoWrite tools to use new TUI rendering components for consistent output format
Refactored render-utils to move tree-related utilities to TUI module (getTreeBranch, getTreeContinuePrefix)
Changed import organization in sdk.ts for consistency
Changed tool result rendering to merge call and result displays, showing tool arguments (command, pattern, query, path) in result headers for Bash, Calculator, Fetch, Find, Grep, Ls, LSP, Notebook, Read, SSH, TodoWrite, Web Search, and Write tools
Changed Read tool title to display line range when offset or limit arguments are provided
Changed worker instantiation to use direct URL import instead of pre-bundled worker files
Changed omp commit to use agentic mode by default with tool-based git inspection
Changed agentic commit progress output to show real-time thinking previews and structured tool argument details
Changed agentic commit progress output to display full multi-line assistant messages and render tool arguments with tree-style formatting for improved readability
Changed agentic commit progress output to render assistant messages as formatted Markdown with proper word wrapping
Changed output block border color to reflect state (error, success, warning) for improved visual feedback
Changed LSP hover rendering to display documentation text before code blocks in both collapsed and expanded views
Changed Write tool to show streaming preview of content being written with syntax highlighting
Changed Read tool to display resolved path information when reading from URLs or symlinks
Changed Calculator tool result display to show both expression and output (e.g., 2+2 = 4) instead of just the result
Changed Python tool output to group status information under a labeled section for clearer organization
Changed SSH tool output to apply consistent styling to non-ANSI output lines
Changed Todo Write tool to respect expanded/collapsed state and use standard preview limits
Changed Web Search related questions to respect expanded/collapsed state instead of always showing all items
Changed empty and error state rendering across multiple tools (Find, Grep, Ls, Notebook, Calculator, Ask) to include consistent status headers
Changed split commit to support hunk selectors (all, indices, or line ranges) instead of whole-file staging
Changed analyze_file tool to analyze_files for batch parallel analysis of multiple files
Switched agentic commit from auto-generated changelogs to agent-proposed entries with validation and retry logic
Commit agent now resolves a separate smaller model for commit generation instead of reusing the primary model
Normalized code formatting and indentation across tool renderers and UI components
Changed git-file-diff tool to prioritize files by type and respect token budget limits with intelligent truncation
Changed git-overview tool to filter and report excluded lock files separately from staged files
Changed analyze-file tool to include file type inference and enriched related files with line counts
Changed propose-changelog tool to support optional deletion entries for removing existing changelog items
Changed commit agent to accept pre-computed file observations and format them into session prompts
Changed changelog skip condition in applyChangelogProposals to also check for empty deletions object
Changed createCommitTools() to build tools array incrementally with conditional analyze_files inclusion based on enableAnalyzeFiles flag
Changed system prompt guidance to clarify that pre-computed observations prevent redundant analyze_files calls
Removed map-reduce preprocessing phase from commit agent for faster iteration
Changed commit agent to process full diff text directly instead of pre-computed file observations
Changed commit agent initialization to load settingsManager, authStorage, modelRegistry, and stagedFiles in parallel
Changed commit agent prompt to remove pre-computed observations guidance and encourage direct analyze_files usage
Changed AuthStorage from constructor-based instantiation to async factory method (AuthStorage.create())
Changed Python kernel resource management with gateway shutdown on session disposal
Updated TypeScript configuration for better publish-time configuration handling with tsconfig.publish.json
Updated TypeScript and Bun configuration for monorepo-wide build consistency and reduced boilerplate
Removed WASM base64 encoding build script; imports now use Bun loader with wasm?raw query parameter
Unified TypeScript checking pipeline with tsgo-based configuration instead of per-package tsconfig.publish.json boilerplate
Refactored scanDirectoryForSkills to use async/await with concurrent directory scanning via Promise.all
Improved error logging in settings manager for config file access failures
Migrated node module imports from named to namespace imports across all packages for consistency with project guidelines
Improved filesystem error handling in extension loader with additional type guards (isEacces, hasFsCode) for permission and EPERM errors
Changed model discovery to synchronous file operations for more immediate initialization

Fixed

Fixed database busy errors during concurrent access by adding retry logic with exponential backoff when opening storage
Find tool now rejects searches from root directory and enforces a 5-second timeout on fd operations
Commit command now exits cleanly with exit code 0 on success
Handle undefined code parameter in code cell renderer
Fixed indentation formatting in split-commit tool function signature
Fixed changelog application to process proposals containing only deletion entries without additions
Fixed indentation formatting in Python tool output renderer
Fixed Python kernel resource management with proper timing instrumentation for performance monitoring
Fixed model discovery to re-check file existence after JSON to YAML migration
Fixed branch change callbacks in footer component to properly update state after git resolution
Added guard clause in plugin-settings to prevent null reference when settings list is undefined
Fixed agent task discovery to support symlinks and improved error handling for file access failures

[8.0.0] - 2026-01-23

Added

Added antigravity provider support for image generation with Google Cloud authentication
Added support for google-antigravity API credentials in model registry
Added antigravity-specific request handling with SSE streaming
Added projectId parameter to antigravity credentials parsing
Added antigravity provider to preferred image provider selection
Added list-limit utility for consistent result limiting across tools
Added output-utils module with tail buffer and artifact allocation helpers
Added tool-result module for standardized tool result construction
Added truncation summary options to OutputMetaBuilder for better output tracking
Added artifact storage system for truncated tool outputs with artifact:// URL protocol
Added structured output metadata system with fluent OutputMetaBuilder for consistent notices
Added standardized tool error types (ToolError, MultiError, ToolAbortError) for better error handling
Added internal URL routing system with protocol handlers:
agent://<id> - access agent output artifacts
agent://<id>/<path> and agent://<id>?q=<query> - JSON extraction from agent outputs
skill://<name> and skill://<name>/<path> - read skill files and relative paths
rule://<name> - read rule content
URL resolution includes filesystem path in output for bash/python interop
Added fetch tool for URL content retrieval with enhanced processing capabilities
Added isolated option to task tool for git worktree execution with automatic patch generation and application
Added format-prompts script to standardize prompt file formatting

Changed

Updated default line limit from 4000 to 3000 lines for output truncation
Reordered truncation notice to show offset continuation before artifact reference
Applied meta notice wrapper to all tools in createTools function
Updated test expectations to reflect new 3000 line limit
Removed output tool from schema validation test list
Replaced inline output truncation notices with structured metadata system across all tools
Updated bash, python, and ssh executors to track detailed output statistics (total lines/bytes vs output lines/bytes)
Modified artifact storage to use pre-allocated paths instead of inline file writing
Changed message format to use meta field instead of fullOutputPath for truncation information
Updated interactive components to display truncation metadata from structured format
Standardized tool result building with new ToolResultBuilder for consistent metadata handling
Simplified Python gateway coordination by removing reference counting and client tracking
Updated Python gateway to use global shared instance instead of per-process coordination
Modified Python kernel initialization to set working directory and environment per kernel
Updated interactive status display to show Python and venv paths instead of client count
Changed system prompt to clarify CHECKPOINT step 0 timing
Updated Python environment warming to use await instead of void for proper error handling
Updated interactive mode shutdown to use postmortem.quit instead of process.exit
Updated bash tool documentation to clarify specialized tool usage
Updated task tool documentation to escape placeholder syntax in examples
Updated Python environment warming to use await instead of void for proper error handling
Updated interactive mode shutdown to use postmortem.quit instead of process.exit
Updated bash tool documentation to clarify specialized tool usage
Updated task tool documentation to escape placeholder syntax in examples
Updated all tools to use structured metadata instead of inline notices for truncation, limits, and diagnostics
Replaced manual error formatting with ToolError.render() and standardized error handling
Enhanced bash and python executors to save full output as artifacts when truncated
Improved abort signal handling across all tools with consistent ToolAbortError
Renamed task parameter from vars to args throughout task tool interface and updated template rendering to support built-in {{id}} and {{description}} placeholders
Simplified todo-write tool by removing active_form parameter, using single content field for task descriptions
Updated system prompt structure with <important> and <avoid> tags, clearer critical sections, and standardized whitespace handling
Renamed web_fetch tool to fetch and removed internal URL handling (use read tool instead)
Standardized tool parameter names from camelCase to snake_case across edit, grep, python, and todo-write tools
Unified timeout parameters across all tools with auto-conversion from milliseconds and reasonable clamping (1s-3600s for bash/ssh, 1s-600s for python/gemini-image)
Simplified web-search tool by removing advanced parameters (max_tokens, model, search_domain_filter, search_context_size, return_related_questions) and using recency instead of search_recency_filter
Restructured tool documentation with standardized <instruction>, <output>, <critical>, and <avoid> sections across all 18 tools
Updated find tool to always sort results by modification time
Updated bash prompt to use cwd parameter instead of workdir
Improved output truncation limits: bash to 50KB/2000 lines, python to 100KB
Removed model parameter from task and gemini-image tools to use session/provider defaults
Improved MCP tool name handling with explicit server and tool name properties
Marked read tool as non-abortable to improve performance
Converted dynamic imports to static imports in installer and exa tools

Removed

Removed output tool (replaced by agent:// URLs via read tool)
Removed web_fetch tool (replaced by fetch tool)

Fixed

Fixed Python kernel environment initialization for external and shared gateways
Fixed gateway status reporting to include Python and virtual environment paths
Fixed inconsistent error formatting across tools by standardizing on ToolError types
Fixed timeout parameter handling to auto-convert milliseconds to seconds and clamp to reasonable ranges
Fixed whitespace formatting in json-query.ts comment
Fixed interactive shutdown to await postmortem cleanup so Python kernel gateways are terminated
Fixed shared Python gateway reuse across working directories by initializing kernel cwd and env per kernel
Fixed Python gateway coordination to use a single global gateway without ref counting

[7.0.0] - 2026-01-21

Added

Added usage report deduplication to prevent duplicate account entries
Added debug logging for usage fetch operations to aid diagnostics
Added provider sorting in usage display by total usage amount
Added isolated parameter to task tool for running each task in separate git worktrees
Added git worktree management for isolated task execution with patch generation
Added patch application system that applies changes only when all patches are valid
Added working directory information to environment info display
Added /usage command to display provider usage and limits
Added support for multiple usage providers beyond Codex
Added usage report caching with configurable TTL
Added visual usage bars and account aggregation in usage display
Added fetchUsageReports() method to agent session
Added output() function to read task/agent outputs by ID with support for multiple formats and queries
Added session file support to Python executor for accessing task outputs
Added support for jq-like queries when reading JSON outputs
Added offset and limit parameters for reading specific line ranges from outputs
Added "." and "c" shortcuts to continue agent without sending visible message
Added debug logging for usage fetch results to aid /usage diagnostics

Changed

Updated discoverSkills function to return object with skills property
Enhanced usage report merging to combine limits and metadata from duplicate accounts
Improved OAuth credential handling to preserve existing fields when updating
Removed cd function from Python prelude to encourage using cwd parameter
Updated task tool to generate and apply patches when running in isolated mode
Enhanced task tool rendering to display isolated execution status and patch paths
Updated system prompt structure and formatting for better readability
Reorganized tool hierarchy and discipline sections
Added parallel work guidance for task-based workflows
Enhanced verification and integration methodology sections
Updated skills and rules formatting for cleaner presentation
Added stronger emphasis on completeness and quality standards
Refactored usage tracking from Codex-specific to generic provider system
Updated usage limit detection to work with multiple provider APIs
Changed usage cache to use persistent storage instead of in-memory only
Limited diagnostic messages to 50 items to prevent overwhelming output when processing files with many issues
Changed /dump command to include complete agent context: system prompt, model config, available tools with schemas, and all message types (bash/python executions, custom messages, branch summaries, compaction summaries, file mentions)
Changed /dump format to use YAML instead of JSON for tool schemas and arguments (more readable)

Fixed

Fixed TypeScript error in bash executor by properly typing caught exception
Fixed usage display ordering to show providers with lowest usage first
Fixed task tool result rendering to show fallback text when no results are available
Fixed external editor to work properly on Unix systems by correctly handling terminal I/O
Fixed external editor to show warning message when it fails to open instead of silently failing
Fixed find tool to properly handle no matches case without treating as error
Fixed find tool to wait for fd exit so error messages no longer report exit null
Fixed read tool to properly handle no matches case without treating as error
Fixed orphaned Python kernel gateway processes not being killed on process exit
Fixed /usage provider ordering to sort by aggregate usage (most used last)
Fixed /usage account dedupe to collapse identical accounts using usage metadata

[6.9.69] - 2026-01-21

Added

Added cell-by-cell status tracking with duration and exit code for Python execution
Added syntax highlighting for Python code in execution display
Added template system with {{placeholders}} for task tool context
Added task variables support for filling context placeholders
Added enhanced task progress display with variable values
Added concurrent work handling guidance in system prompt
Added extension system support for user Python execution events
Added Python mode border color theming across all themes
Added Python execution indicator to welcome screen help text
Added omp stats command for viewing AI usage statistics dashboard
Added support for JSON output and console summary of usage statistics
Added configurable port option for stats dashboard server
Added multi-cell Python execution with sequential processing in persistent kernel
Added cell titles for better Python code organization and debugging
Added $ command prefix for user-initiated Python execution in shared kernel
Added $$ prefix variant for Python execution excluded from LLM context

Changed

Updated Python execution to display cells in bordered blocks with status indicators
Changed task tool to use template-based context instead of simple concatenation
Enhanced Python execution component with proper syntax highlighting
Improved patch applicator to preserve exact indentation when intended
Updated task tool schema to require vars instead of task field
Updated Python execution component to use pythonMode theming instead of bashMode
Enhanced UI helpers to handle pending Python components properly
Changed Python tool to use cells array instead of single code parameter
Renamed workdir parameter to cwd in Bash and Python tools for consistency
Updated Python tool to display cell-by-cell output when multiple cells are provided

Fixed

Fixed indentation preservation for exact matches and indentation-only patches
Fixed Python execution status updates to show real-time cell progress
Fixed indentation adjustment logic to handle edge cases with mixed indentation levels
Fixed patch indentation normalization for fuzzy matches, tab/space diffs, and ambiguous context alignment

[6.9.0] - 2026-01-21

Removed

Removed Git tool and all related functionality
Removed voice control and TTS features
Removed worktree management system
Removed bundled wt custom command
Removed voice-related settings and configuration options
Removed @oh-my-pi/pi-git-tool dependency

[6.8.5] - 2026-01-21

Breaking Changes

Changed timeout parameter from seconds to milliseconds in Python tool
Updated PythonExecutorOptions interface to use timeoutMs instead of timeout

Changed

Updated default timeout to 30000ms (30 seconds) for Python tool
Improved streaming output handling and buffer management

[6.8.4] - 2026-01-21

Changed

Updated output sink to properly handle large outputs
Improved error message formatting in SSH executor
Updated web fetch timeout bounds and conversion

Fixed

Fixed output truncation handling in streaming output
Fixed timeout handling in web fetch tool
Fixed async stream dumping in executors

[6.8.3] - 2026-01-21

Changed

Updated keybinding system to normalize key IDs to lowercase
Changed label edit shortcut from 'l' to 'Shift+L' in tree selector
Changed output file extension from .out.md to .md for artifacts

Removed

Removed bundled worktree command from custom commands loader

Fixed

Fixed keybinding case sensitivity issues by normalizing all key IDs
Fixed task artifact path handling and simplified file structure

[6.8.2] - 2026-01-21

Fixed

Improved error messages when multiple text occurrences are found by showing line previews and context
Enhanced patch application to better handle duplicate content in context lines
Added occurrence previews to help users disambiguate between multiple matches
Fixed cache invalidation for streaming edits to prevent stale data
Fixed file existence check for prompt templates directory
Fixed bash output streaming to prevent premature stream closure
Fixed LSP client request handling when signal is already aborted
Fixed git apply operations with stdin input handling

Security

Updated Anthropic authentication to handle manual code input securely

[6.8.1] - 2026-01-20

Fixed

Fixed unhandled promise rejection when tool execution fails by adding missing .catch() to floating .finally() chain in createAbortablePromise

[6.8.0] - 2026-01-20

Added

Added streaming abort setting to control edit tool behavior when patch preview fails

Changed

Replaced internal logger with @oh-my-pi/pi-utils logger across all modules
Updated process spawning to use cspawn and ptree utilities from pi-utils
Migrated file operations to use async fs/promises and Bun file APIs
Refactored promise handling to use Promise.withResolvers and utility functions
Updated timeout and abort handling to use standardized utility functions
Refactored authentication login method to use OAuthController interface instead of individual callbacks

Fixed

Fixed Python package installation to handle async operations properly
Fixed streaming output truncation to use consistent column limits
Fixed shell command execution to properly handle process cleanup and timeouts
Fixed SSH connection management to properly await async operations
Fixed voice supervisor process cleanup to use proper async handling
Added automatic regex pattern validation in grep tool to handle invalid patterns by switching to literal mode

Security

Updated temporary file cleanup to use secure async removal methods

[6.7.67] - 2026-01-19

Added

Added normative rewrite setting to control tool call argument normalization in session history
Added read line numbers setting to prepend line numbers to read tool output by default
Added streaming preview for edit and write tools with spinner animation
Added automatic anchor derivation for normative patches when anchors not specified

Changed

Enhanced edit and write tool renderers to show streaming content preview
Updated read tool to respect default line numbers setting
Improved normative patch anchor handling to support undefined anchors

[6.7.0] - 2026-01-19

Added

Normative patch generation to canonicalize edit tool output with tool call argument rewriting for session history
Patch matching fallback variants: trimmed context, collapsed duplicates, single-line reduction, comment-prefix normalization
Extended anchor syntax: ellipsis placeholders, top of file/start of file, @@ line N, nested @@ anchors, space-separated hierarchical contexts
Relaxed fuzzy threshold fallback and unique substring acceptance for context matching
Added --no-title flag to disable automatic session title generation
Environment variables for edit tool configuration (OMP_EDIT_VARIANT, OMP_EDIT_FUZZY, OMP_EDIT_FUZZY_THRESHOLD)
Configurable fuzzy matching threshold setting (0.85 lenient to 0.98 strict)
Apply-patch mode for edit tool (edit.patchMode setting) with create, update, delete, and rename operations
Added MCP tool caching for faster startup with cached tool definitions

Changed

Patch applicator now supports normalized input, implicit context lines, and improved indentation adjustment
Patch operation schema uses 'op' instead of 'operation' and 'rename' instead of 'moveTo'
Fuzzy matching tries comment-prefix normalized matches before unicode normalization
Updated patch prompts with clearer anchor selection rules and verbatim context requirements
Changed default behavior of read tool to omit line numbers by default
Changed default edit tool mode to use apply-patch format instead of oldText/newText
Converted tool implementations from factory functions to class-based architecture
Refactored edit tool with modular patch architecture (moved from edit/ to patch/ module)
Enhanced patch parsing: unified diff format, Codex-style patches, nested anchors, multi-file markers
Improved fuzzy matching with multiple match tracking, ambiguity detection, and out-of-order hunk processing
Better diff rendering: smarter truncation, optional line numbers, trailing newline preservation
Improved error messages with hierarchical context display using > separator
Centralized output sanitization in streaming-output module
Enhanced MCP startup with deferred tool loading and cached fallback

Fixed

Patch application handles repeated context blocks, preserves original indentation on fuzzy match
Ambiguous context matching resolves duplicates using adjacent @@ anchor positioning
Patch parser handles bare *** terminators, model hallucination markers, line hint ranges
Function context matching handles signatures with and without empty parentheses
Fixed session title generation to respect OMP_NO_TITLE environment variable
Fixed Python module discovery to use import.meta.dir for ES module compatibility
Fixed LSP writethrough batching to flush when delete operations complete a batch
Fixed line number validation, BOM detection, and trailing newline preservation in patches
Fixed hierarchical context matching and space-separated anchor parsing
Fixed fuzzy matching to avoid infinite loops when allowFuzzy is disabled
Fixed tool completion logic to only mark tools as complete when streaming is not aborted or in error state
Fixed MCP tool path formatting to correctly display provider information

[6.2.0] - 2026-01-19

Changed

Improved LSP batching to coalesce formatting and diagnostics for parallel edits
Updated edit and write tools to support batched LSP operations

Fixed

Coalesced LSP formatting/diagnostics for parallel edits so only the final write triggers LSP across touched files

[6.1.0] - 2026-01-19

Added

Added lspmux integration for LSP server multiplexing to reduce startup time and memory usage
Added LSP tool proxy support for subagent workers
Updated LSP status command to show lspmux connection state
Added maxdepth and mindepth parameters to find function for depth-controlled file search
Added counter function to count occurrences and sort by frequency
Added basenames function to extract base names from paths

Changed

Simplified rust-analyzer default configuration by removing custom initOptions and settings

[6.0.0] - 2026-01-19

Added

Added Cursor and OpenAI Codex OAuth providers
Added Windows installer bash shell auto-configuration
Added dedicated TTSR settings tab (separated from Voice/TTS)

Fixed

Fixed TTSR abbreviation expansion from TTSR to Time Traveling Stream Rules

[5.8.0] - 2026-01-19

Changed

Updated WASM loading to use streaming for development environments with base64 fallback
Added scripts directory to published package files

[5.7.68] - 2026-01-18

Changed

Updated WASM loading to use base64-encoded WASM for better compatibility with compiled binaries

Fixed

Fixed WASM loading issues in compiled binary builds

[5.7.67] - 2026-01-18

Changed

Replaced external photon-node dependency with vendored WebAssembly implementation
Updated image processing to use local photon library for better performance

[5.6.70] - 2026-01-18

Added

Added support for loading Python prelude extension modules from user and project directories
Added automatic discovery of Python modules from .omp/modules and .pi/modules directories
Added prioritized module loading with project-level modules overriding user-level modules

[5.6.7] - 2026-01-18

Added

Added Python shared gateway setting to enable resource-efficient kernel reuse across sessions
Added Python tool cancellation support with proper timeout and cleanup handling
Added enhanced Python prelude helpers including file operations, text processing, and Git utilities
Added Python tool documentation rendering with categorized helper functions
Added session-scoped Python kernel isolation with workdir-aware session IDs
Added structured status events for Python prelude functions with TUI rendering
Added status event display system with operation icons and formatted descriptions
Added support for rich output using IPython.display.display() in Python tool
Added setup subcommand to install dependencies for optional features
Added Python setup component to install Jupyter kernel dependencies
Added setup command help with component and option documentation
Added Python tool dependency check in help output
Added file locking mechanism for shared Python gateway to prevent race conditions
Added Python gateway status monitoring with URL, PID, client count, and uptime information
Added comprehensive Git helpers to Python prelude including status, diff, log, show, branch, and file operations
Added line-based operations to Python prelude including line extraction, deletion, insertion, and pattern matching
Added automatic categorization system for Python prelude functions with discoverable documentation
Added enhanced /status command display showing Python gateway, LSP servers, and MCP server connections
Added shared Python gateway coordinator for resource-efficient kernel management across sessions
Added Python shared gateway setting with session-scoped kernel reuse and fallback behavior
Added automatic idle shutdown for shared Python gateway after 30 seconds of inactivity
Added environment filtering for shared Python gateway to exclude sensitive API keys
Added virtual environment detection and automatic PATH configuration for Python gateway
Added IPython-backed Python tool with streaming output, image/JSON rendering, and Jupyter kernel gateway integration
Added Python prelude with 30+ shell-like utility functions for file operations
Added Python tool exposure settings with session-scoped kernel reuse and fallback behavior
Added streaming output system with automatic spill-to-disk for large outputs
Added extension input interception with source metadata and command argument completion
Added extension command context compact() helper plus context usage accessors
Added ExtensionAPI setLabel() for extension and entry labels
Added startup quiet setting to suppress welcome screen and startup messages
Added support for auto-discovering APPEND_SYSTEM.md files
Added support for piped input in non-interactive mode (auto-print mode)
Added global session listing across all project directories with enhanced search metadata
Added session fork prompt when resolving sessions from other projects
Added key hint formatting utilities plus public exports for getShellConfig/getAgentDir/VERSION
Added bash tool timeout display in tool output
Added fuzzy text normalization for improved edit diff matching
Added $@ argument slicing syntax in prompt templates
Added configurable keybindings for expand tools and dequeue actions
Added process title update on CLI startup

Changed

Updated Python tool description to display categorized helper functions with improved formatting
Enhanced Python kernel startup to use shared gateway by default for better resource utilization
Improved Python prelude functions to emit structured status events instead of text output
Updated agent prompts to use bash tool instead of exec for git operations
Changed default Python tool mode from ipy-only to both to enable shell execution
Enhanced Python gateway coordination with Windows environment support and stale process cleanup
Updated Python prelude functions to emit structured status events instead of text output
Enhanced Python tool renderer to display status events alongside output
Improved Python tool output formatting with status event integration
Improved shared Python gateway coordination with environment validation and stale process cleanup
Updated Python prelude to rename bash() function to sh() for consistency
Changed default Python tool mode from "ipy-only" to "both" to enable both IPython and shell execution
Enhanced Python gateway metadata tracking to include Python path and virtual environment information
Improved Python kernel startup to use shared gateway by default for better resource utilization
Updated Python tool to support proxy execution mode for worker processes
Enhanced Python kernel availability checking with faster validation
Optimized Python environment warming to avoid blocking during tool initialization
Reorganized settings interface into behavior, tools, display, voice, status, lsp, and exa tabs
Migrated environment variables from PI_ to OMP_ prefix with automatic migration
Updated model selector to use TabBar component for provider navigation
Changed role badges to inverted style with colored backgrounds
Added support for /models command alias in addition to /model
Improved error retry detection to include fetch failures
Enhanced session selector search and overflow handling
Updated skill command execution to include skill path metadata
Surfaced loaded prompt templates during initialization
Updated compaction summarization to use serialized prompt text
Cleaned up Python prelude sh() and run() output to only show stdout/stderr without noisy metadata

Fixed

Fixed Python kernel cancellation handling and WebSocket cleanup for in-flight executions
Fixed Python tool session scoping to include workdir and honor sharedGateway settings
Fixed gist sharing output draining to avoid truncated URLs
Fixed streaming output byte accounting and UTF-8 decoder flushing
Fixed Python prelude integration tests to detect virtual environments and cover helper exports
Fixed Python kernel cancellation/timeout handling and WebSocket close cleanup for in-flight executions
Fixed Python output byte accounting and UTF-8 decoder flushing in streaming output
Fixed shared Python gateway coordination (Windows env allowlist, lock staleness, refcount recovery)
Fixed Python tool session scoping to include workdir and honor sharedGateway settings
Fixed subagent Python proxy session isolation and cancellation/timeout propagation
Fixed print-mode cleanup to dispose Python sessions before exit
Fixed gist share output draining to avoid truncated URLs
Fixed explore agent tool list to use bash for git operations
Fixed Python prelude integration tests to detect venv-only Python and cover helper exports

Security

Enhanced Python gateway environment filtering to exclude sensitive API keys and Windows system paths

[5.5.0] - 2026-01-18

Changed

Updated task execution guidelines to improve prompt framing and parallelization instructions

[5.4.2] - 2026-01-16

Changed

Updated model resolution to accept pre-serialized settings for better performance
Improved system prompt guidance for position-addressed vs content-addressed file edits
Enhanced edit tool documentation with clear use cases for bash alternatives

[5.3.0] - 2026-01-15

Changed

Expanded bash tool guidance to explicitly list appropriate use cases including file operations, build commands, and process management

[5.2.1] - 2026-01-14

Fixed

Fixed stale diagnostic results by tracking diagnostic versions before file sync operations
Fixed race condition where LSP diagnostics could return outdated results after file modifications

[5.2.0] - 2026-01-14

Added

Added withLines parameter to read tool for optional line number output (default: true, cat -n format)

Changed

Changed find/grep/ls tool output to render inline without background box for cleaner visual flow

Fixed

Fixed task tool abort to return partial results instead of failing (completed tasks preserved, cancelled tasks shown as skipped)
Fixed TUI crash when bash output metadata lines exceed terminal width on narrow terminals
Fixed find tool not matching **/filename patterns (was incorrectly using --full-path for glob depth wildcards)

[5.1.1] - 2026-01-14

Fixed

Fixed clipboard image paste getting stuck on Wayland when no image is present (was falling back to X11 and timing out)

[5.1.0] - 2026-01-14

Changed

Updated light theme colors for WCAG AA compliance (4.5:1 contrast against white background)
Changed dequeue hint text from "restore" to "edit all queued messages"

Fixed

Fixed session selector staying open when current folder has no sessions (shows hint to press Tab)
Fixed print mode JSON output to emit session header at start
Fixed "database is locked" SQLite errors when running subagents by serializing settings to workers instead of opening the database
Fixed /new command to create a new session file (previously reused the same file when --session was specified)
Fixed session selector page up/down navigation

[5.0.1] - 2026-01-12

Changed

Replaced wasm-vips with Photon for more stable WASM image processing
Added graceful fallback to original images when image resizing fails
Added error handling for image conversion failures in interactive mode to prevent crashes
Replace wasm-vips with Photon for more stable WASM image processing (fixes worker thread crashes)

[5.0.0] - 2026-01-12

Added

Implemented xhigh thinking level for Anthropic models with increased reasoning limits

[4.8.3] - 2026-01-12

Changed

Replace sharp with wasm-vips for cross-platform image processing without native dependencies

[4.8.1] - 2026-01-12

Fixed

Prevent bun from statically resolving sharp import to fix runtime errors on arm64

[4.8.0] - 2026-01-12

Fixed

Move sharp to optional dependencies with all platform binaries to fix arm64 runtime errors

[4.7.0] - 2026-01-12

Added

Add omp config subcommand for managing settings (list, get, set, reset, path)
Add todoCompletion setting to warn agent when it stops with incomplete todos (up to 3 reminders)
Add multi-part questions support to ask tool via questions array parameter

Changed

Updated multi-select cursor behavior in ask tool to stay on the toggled option instead of jumping to top
Single-file reads now render inline (e.g., Read AGENTS.md:23) instead of tree structure

Fixed

Subagent model resolution now respects explicit provider prefix (e.g., zai/glm-4.7 no longer matches cerebras/zai-glm-4.7)
Auto-compaction now skips to next model candidate when retry delay exceeds 30 seconds

[4.6.0] - 2026-01-12

Added

Add /skill:name slash commands for quick skill access (toggle via skills.enableSkillCommands setting)
Add cwd to SessionInfo for session list display
Add custom summarization instructions option in tree selector
Add Alt+Up (dequeue) to restore all queued messages at once
Add shutdownRequested and checkShutdownRequested() for extension-initiated shutdown

Fixed

Component invalidate() now properly rebuilds content on theme changes
Force full re-render after returning from external editor

[4.4.8] - 2026-01-12

Changed

Changed review finding priority format from numeric (0-3) to string labels (P0-P3) for clearer severity indication
Replaced Type.Union with Type.Literal patterns with StringEnum helper across tool schemas for cleaner enum definitions

[4.4.5] - 2026-01-11

Changed

Removed format: "date-time" from timestamp type conversion in JTD to JSON Schema transformation
Reorganized system prompt to display context, environment, and tools sections before discipline guidelines
Updated system prompt to show file paths more clearly in output
Improved YAML frontmatter parsing with better error messages including source file information

Fixed

Fixed frontmatter parsing to properly report source location when YAML parsing fails

[4.4.4] - 2026-01-11

Added

Added todo_write tool for creating and managing structured task lists during coding sessions
Added persistent todo panel above the editor that displays task progress
Added Ctrl+T keybinding to toggle todo list expansion
Added grouped display for consecutive Read tool calls, showing multiple file reads in a compact tree view
Added todo_write tool and persistent todo panel above the editor

Changed

Changed Ctrl+Enter to insert a newline when not streaming (previously Alt+Enter)
Changed Ctrl+T from toggling thinking block visibility to toggling todo list expansion
Changed system prompt to use more direct, field-oriented language with emphasis on verification and assumptions
Changed temporary model selector keybinding from Ctrl+Y to Alt+P
Changed expand hint text from "Ctrl+O to expand" to "Ctrl+O for more"
Changed Read tool result display to hide content by default, showing only file path and status
Changed Ctrl+T to toggle todo panel expansion

Removed

Removed yaml package dependency in favor of Bun's built-in YAML parser

Fixed

Fixed Alt+Enter to insert a newline when not streaming, instead of submitting the message
Fixed Alt+Enter inserting a new line when not streaming instead of submitting a message
Fixed Cursor provider to avoid advertising the Edit tool, relying on full-file Write operations instead
Fixed prompt template loading to strip leading HTML comment metadata blocks

[4.3.2] - 2026-01-11

Changed

Increased default bash output preview from 5 to 10 lines when collapsed
Updated expanded bash output view to show full untruncated output when available

[4.3.1] - 2026-01-11

Changed

Expanded system prompt with defensive reasoning guidance and assumption checks
Allowed agent frontmatter to override subagent thinking level, clamped to model capabilities

Fixed

Ensured reviewer agents use structured output schemas and include reported findings in task outputs

[4.3.0] - 2026-01-11

Added

Added Cursor provider support with browser-based OAuth authentication
Added default model configuration for Cursor provider (claude-sonnet-4-5)
Added execution bridge for Cursor tool calls including read, ls, grep, write, delete, shell, diagnostics, and MCP operations

Fixed

Improved fuzzy matching accuracy for edit operations when file and target have inconsistent indentation patterns

[4.2.3] - 2026-01-11

Changed

Changed default for hidden option in find tool from false to true, now including hidden files by default

Fixed

Fixed serialized auth storage initialization so OAuth refreshes in subagents don't crash

[4.2.2] - 2026-01-11

Added

Added persistent cache storage for Codex usage data that survives application restarts
Added --no-lsp to disable LSP tools, formatting, diagnostics, and warmup for a session

Changed

Changed SettingsManager.create() to be async, requiring await when creating settings managers
Changed loadSettings() to be async, requiring await when loading settings
Changed discoverSkills() to be async, requiring await when discovering skills
Changed loadSlashCommands() to be async, requiring await when loading slash commands
Changed buildSystemPrompt() to be async, requiring await when building system prompts
Changed loadSkills() to be async, requiring await when loading skills
Changed loadProjectContextFiles() to be async, requiring await when loading context files
Changed getShellConfig() to be async, requiring await when getting shell configuration
Changed capability provider load() methods to be async-only, removing synchronous loadSync API
Updated plan agent with enhanced structured planning process, parallel exploration via explore agent spawning, and improved output format with examples
Removed planner agent command template, consolidating planning functionality into the plan agent

[4.2.1] - 2026-01-11

Added

Added automatic discovery and listing of AGENTS.md files in the system prompt, providing agents with an authoritative list of project-specific instruction files without runtime searching
Added planner built-in agent for comprehensive implementation planning with slow model

Changed

Refactored skill discovery to use unified loadSkillsFromDir helper across all providers, reducing code duplication
Updated skill discovery to scan only skills/*/SKILL.md entries instead of recursive walks in Codex provider
Added guidance to Task tool documentation to isolate file scopes when assigning tasks to prevent agent conflicts
Updated Task tool documentation to emphasize that subagents have no access to conversation history and require all relevant context to be explicitly passed
Revised task agent prompt to clarify that subagents have full tool access and can make file edits, run commands, and create files
OpenAI Codex: updated to use bundled system prompt from upstream
Changed complete tool to make data parameter optional when aborting, while still requiring it for successful completions
Skills discovery now scans only skills/*/SKILL.md entries instead of recursive walks

Removed

Removed architect-plan, implement, and implement-with-critic built-in agent commands

Fixed

Fixed editor border rendering glitch after canceling slash command autocomplete
Fixed login/logout credential path message to reference agent.db

[4.2.0] - 2026-01-10

Added

Added /dump slash command to copy the full session transcript to the clipboard
Added automatic Nerd Fonts detection for terminals like iTerm, WezTerm, Kitty, Ghostty, and Alacritty to set appropriate symbol preset
Added NERD_FONTS environment variable override (1 or 0) to manually control Nerd Fonts symbol preset
Added Handlebars templating engine for prompt template rendering with {{arg}} helper for positional arguments
Added support for custom share scripts at ~/.omp/agent/share.ts to replace default GitHub Gist sharing

Changed

Changed rules system to use read tool for loading rule content instead of dedicated rulebook tool
Separated /export and /dump commands—/export now only exports to HTML file, while /dump copies session transcript to clipboard
Updated /export command to no longer accept --copy flag (use /dump instead)
Changed prompt template rendering to use Handlebars instead of simple string replacement
Updated prompt layout optimization to normalize indentation and collapse excessive blank lines
Changed auth migration to merge credentials per-provider instead of skipping when any credentials exist in database
Migrated settings and auth credential storage from JSON files to SQLite database (agent.db)
Updated credential migration message to reference agent.db instead of auth.json
Renamed Glob tool references to Find tool throughout prompts and documentation
Updated project context formatting to use XML-style tags for clearer structure
Refined bash tool guidance to prefer dedicated tools (read/grep/find/ls) over bash for file operations
Updated system prompt with clearer tone guidelines emphasizing directness and conciseness
Revised workflow instructions to require explicit planning for non-trivial tasks
Enhanced verification guidance to prefer external feedback loops like tests and linters
Added explicit alignment and prohibited behavior sections to improve response quality

Removed

Removed rulebook tool - rules are now loaded via the read tool instead of a dedicated tool

Fixed

Fixed message submission lag caused by synchronous history database writes by deferring DB operations with setImmediate

Security

Hardened file permissions on agent database directory (700) and database file (600) to restrict access

[4.1.0] - 2026-01-10

Added

Added persistent prompt history with SQLite-backed storage and Ctrl+R search

Fixed

Fixed credential blocking logic to correctly check for remaining available credentials instead of always returning true

[4.0.1] - 2026-01-10

Added

Added usage limit error detection to enable automatic credential switching when Codex accounts hit rate limits
Added Codex usage API integration to proactively check account limits before credential selection
Added credential backoff tracking to temporarily skip rate-limited accounts during selection
Multi-credential usage-aware selection for OpenAI Codex OAuth accounts with automatic fallback when rate limits are reached
Consistent session-to-credential hashing (FNV-1a) for stable credential assignment across sessions
Codex usage API integration to detect and cache rate limit status per account
Automatic mid-session credential switching when usage limits are hit

Changed

Changed credential selection to use deterministic FNV-1a hashing for consistent session-to-credential mapping
Changed OAuth credential resolution to try credentials in priority order, skipping blocked ones

[4.0.0] - 2026-01-10

Added

Exported InteractiveModeOptions type for programmatic SDK usage
Exported additional UI components for extensions: ArminComponent, AssistantMessageComponent, BashExecutionComponent, BranchSummaryMessageComponent, CompactionSummaryMessageComponent, CustomEditor, CustomMessageComponent, FooterComponent, ExtensionEditorComponent, ExtensionInputComponent, ExtensionSelectorComponent, LoginDialogComponent, ModelSelectorComponent, OAuthSelectorComponent, SessionSelectorComponent, SettingsSelectorComponent, ShowImagesSelectorComponent, ThemeSelectorComponent, ThinkingSelectorComponent, ToolExecutionComponent, TreeSelectorComponent, UserMessageComponent, UserMessageSelectorComponent
Exported renderDiff, truncateToVisualLines, and related types for extension use
setFooter() and setHeader() methods on ExtensionUIContext for custom footer/header components
setEditorComponent() method on ExtensionUIContext for custom editor components
supportsUsageInStreaming model config option to control stream_options: { include_usage: true } behavior
Terminal setup documentation for Kitty keyboard protocol configuration (Ghostty, wezterm, Windows Terminal)
Documentation for paid Cloud Code Assist subscriptions via GOOGLE_CLOUD_PROJECT env var
Environment variables reference section in README
--no-tools flag to disable all built-in tools, enabling extension-only setups
--no-extensions flag to disable extension discovery while still allowing explicit -e paths
blockImages setting to prevent images from being sent to LLM providers
thinkingBudgets setting to customize token budgets per thinking level
PI_SKIP_VERSION_CHECK environment variable to disable new version notifications at startup
Anthropic OAuth support via /login to authenticate with Claude Pro/Max subscription
OpenCode Zen provider support via OPENCODE_API_KEY env var and opencode/<model-id> syntax
Session picker (pi -r) and --session flag support searching/resuming by session ID (UUID prefix)
Session ID forwarding to LLM providers for session-based caching (used by OpenAI Codex for prompt caching)
dequeue keybinding (Alt+Up) to restore queued steering/follow-up messages back into the editor
Pluggable operations for built-in tools enabling remote execution via SSH or other transports (ReadOperations, WriteOperations, EditOperations, BashOperations, LsOperations, GrepOperations, FindOperations)
/model <search> pre-filters the model selector or auto-selects on exact match; use provider/model syntax to disambiguate
Managed binaries directory (~/.omp/bin/) for fd and rg tools
FooterDataProvider for custom footers with getGitBranch(), getExtensionStatuses(), and onBranchChange()
ctx.ui.custom() accepts { overlay: true } option for floating modal components
ctx.ui.getAllThemes(), ctx.ui.getTheme(name), ctx.ui.setTheme(name | Theme) for theme management
setActiveTools() for dynamic tool management
setModel(), getThinkingLevel(), setThinkingLevel() methods for runtime model and thinking level changes
ctx.shutdown() for requesting graceful shutdown
pi.sendUserMessage() for sending user messages from extensions
Extension UI dialogs (select, confirm, input) support timeout option with live countdown display
Extension UI dialogs accept optional AbortSignal to programmatically dismiss dialogs
Async extension factories for dynamic imports and lazy-loaded dependencies
user_bash event for intercepting user !/!! commands
Built-in renderers used automatically for tool overrides without custom renderCall/renderResult
InteractiveMode, runPrintMode(), runRpcMode() exported for building custom run modes
Copy link button on messages for deep linking to specific entries
Codex injection info display showing system prompt modifications
URL parameter support for leafId and targetId deep linking
Wayland clipboard support for /copy command using wl-copy with xclip/xsel fallback

Changed

Bash tool output truncation now recalculates on terminal resize instead of using cached width
Web search tool headers updated to match Claude Code client format for better compatibility
discoverSkills() return type documented as { skills: Skill[], warnings: SkillWarning[] } in SDK docs
Default model for OpenCode provider changed from claude-sonnet-4-5 to claude-opus-4-5
Terminal color mode detection defaults to truecolor for modern terminals instead of 256color
System prompt restructured with XML tags and clearer instructions format
before_agent_start event receives systemPrompt in the event object and returns systemPrompt (full replacement) instead of systemPromptAppend
discoverSkills() returns { skills: Skill[], warnings: SkillWarning[] } instead of Skill[]
ctx.ui.custom() factory signature changed from (tui, theme, done) to (tui, theme, keybindings, done)
ExtensionRunner.initialize() signature changed from options object to positional params (actions, contextActions, commandContextActions?, uiContext?)

Fixed

Wayland clipboard copy (wl-copy) no longer blocks when the process doesn't exit promptly
Empty --tools flag now correctly enables all built-in tools instead of disabling them
Bash tool handles spawn errors gracefully instead of crashing the agent
Components properly rebuild their content on theme change via invalidate() override
setTheme() triggers a full rerender so previously rendered components update with new theme colors
Session ID updates correctly when branching sessions
External edits to settings.json while pi is running are preserved when pi saves settings
Default thinking level from settings applies correctly when enabledModels is configured
LM Studio compatibility for OpenAI Responses tool strict mapping
Symlinked directories in prompts/ folders are followed when loading prompt templates
String systemPrompt in createAgentSession() works as a full replacement instead of having context files and skills appended
Update notification for bun binary installs shows release download URL instead of npm command
ESC key works during "Working..." state after auto-retry
Abort messages show correct retry attempt count
Antigravity provider returning 429 errors despite available quota
Malformed thinking text in Gemini/Antigravity responses where thinking content appeared as regular text
--no-skills flag correctly prevents skills from loading in interactive mode
Overflow-based compaction skips if error came from a different model or was already handled
OpenAI Codex context window reduced from 400k to 272k tokens to match Codex CLI defaults
Context overflow detection recognizes context_length_exceeded errors
Key presses no longer dropped when input is batched over SSH
Clipboard image support works on Alpine Linux and other musl-based distros
Queued steering/follow-up messages no longer wipe unsent editor input
OAuth token refresh failure no longer crashes app at startup
Status bar shows correct git branch when running in a git worktree
Ctrl+V clipboard image paste works on Wayland sessions
Extension directories in settings.json respect package.json manifests

[3.37.0] - 2026-01-10

Changed

Improved bash command display to show relative paths for working directories within the current directory, and hide redundant cd prefix when working directory matches current directory

[3.36.0] - 2026-01-10

Added

Added calc tool for basic mathematical calculations with support for arithmetic operators, parentheses, and hex/binary/octal literals
Added support for multiple API credentials per provider with round-robin distribution across sessions
Added file locking for auth.json to prevent concurrent write corruption
Added clickable OAuth login URL display in terminal
Added workdir parameter to bash tool to execute commands in a specific directory without requiring cd commands

Changed

Updated bash tool rendering to display working directory context when workdir parameter is used

Fixed

Fixed completion notification to only send when interactive mode is in foreground
Improved completion notification message to include session title when available

[3.35.0] - 2026-01-09

Added

Added retry logic with exponential backoff for auto-compaction failures
Added fallback to alternative models when auto-compaction fails with the primary model
Added support for pi/<role> model aliases in task tool (e.g., pi/slow, pi/default)
Added visual cycle indicator when switching between role models showing available roles
Added automatic model inheritance for subtasks when parent uses default model
Added -- separator in grep tool to prevent pattern interpretation as flags

Changed

Changed role model cycling to remember last selected role instead of matching current model
Changed edit tool to merge call and result displays into single block
Changed model override behavior to persist in settings when explicitly set via CLI

Fixed

Fixed retry-after parsing from error messages supporting multiple header formats (retry-after, retry-after-ms, x-ratelimit-reset)
Fixed image attachments being dropped when steering/follow-up messages are queued during streaming
Fixed image auto-resize not applying to clipboard images before sending
Fixed clipboard image attachments being dropped when steering/follow-up messages are queued while streaming
Fixed clipboard image attachments ignoring the auto-resize setting before sending

[3.34.0] - 2026-01-09

Added

Added caching for system environment detection to improve startup performance
Added disk usage information to automatic environment detection in system prompt
Added compat option for SSH hosts to wrap commands in a POSIX shell on Windows systems
Added automatic working directory handling for PowerShell and cmd.exe on Windows SSH hosts
Added automatic environment detection to system prompt including OS, distro, kernel, CPU, GPU, shell, terminal, desktop environment, and window manager information
Added SSH tool with project ssh.json/.ssh.json discovery, persistent connections, and optional sshfs mounts
Added SSH host OS/shell detection with compat mode and persistent host info cache

Changed

Changed GPU detection on Linux to prioritize discrete GPUs (NVIDIA, AMD) over integrated graphics and skip server management adapters
Changed SSH host info cache to use versioned format for automatic refresh on schema changes
Changed SSH compat shell detection to actively probe for bash/sh availability on Windows hosts
Changed SSH tool description to show detected shell type and available commands per host

[3.33.0] - 2026-01-08

Added

Added env support in settings.json for automatically setting environment variables on startup
Added environment variable management methods to SettingsManager (get/set/clear)

Fixed

Fixed bash output previews to recompute on resize, preventing TUI line width overflow crashes
Fixed session title generation to retry alternate smol models when the primary model errors or is rate-limited
Fixed file mentions to resolve extensionless paths and directories, using read tool truncation limits for injected content
Fixed interactive UI to show auto-read file mention indicators
Fixed task tool tree rendering to use consistent tree connectors for progress, findings, and results
Fixed last-branch tree connector symbol in the TUI
Fixed output tool previews to use compact JSON when outputs are formatted with leading braces

[3.32.0] - 2026-01-08

Added

Added progress indicator when starting LSP servers at session startup
Added bundled /init slash command available by default

Changed

Changed LSP server warmup to use a 5-second timeout, falling back to lazy initialization for slow servers

Fixed

Fixed Task tool subagent model selection to inherit explicit CLI --model overrides

[3.31.0] - 2026-01-08

Added

Added temporary model selection: Ctrl+Y opens model selector for session-only model switching (not persisted to settings)
Added setModelTemporary() method to AgentSession for ephemeral model changes
Added empty Enter to flush queued messages: pressing Enter with empty editor while streaming aborts current stream
Added auto-chdir to temp directories when starting in home unless --allow-home is set
Added upfront diff parsing and filtering for code review command to exclude lock files, generated code, and binary assets

Fixed

Fixed auto-chdir to only use existing directories and fall back to tmpdir()
Added automatic reviewer agent count recommendation based on diff weight and file count
Added file grouping guidance for parallel review distribution across multiple agents
Added diff preview mode for large changesets that exceed size thresholds
Added in-memory session storage implementation for testing and ephemeral sessions
Added createToolUIKit helper to consolidate common UI formatting utilities across tool renderers
Added configurable bash interceptor rules via bashInterceptor.patterns setting for custom command blocking
Added bashInterceptor.simpleLs setting to control interception of bare ls commands
Added LSP server configuration via external JSON defaults file for easier customization
Added abort signal propagation to web scrapers for improved cancellation handling
Added diagnosticsVersion tracking to LSP client for more reliable diagnostic polling
Added 80+ specialized web scrapers for structured content extraction from popular sites including GitHub, GitLab, npm, PyPI, crates.io, Wikipedia, YouTube, Stack Overflow, Hacker News, Reddit, arXiv, PubMed, and many more
Added site-specific API integrations for package registries (npm, PyPI, crates.io, Hex, Hackage, NuGet, Maven, RubyGems, Packagist, pub.dev, Go packages)
Added scrapers for social platforms (Mastodon, Bluesky, Lemmy, Lobsters, Dev.to, Discourse)
Added scrapers for academic sources (arXiv, bioRxiv, PubMed, Semantic Scholar, ORCID, CrossRef, IACR)
Added scrapers for security databases (NVD, OSV, CISA KEV)
Added scrapers for documentation sites (MDN, Read the Docs, RFC Editor, W3C, SPDX, tldr, cheat.sh)
Added scrapers for media platforms (YouTube, Vimeo, Spotify, Discogs, MusicBrainz)
Added scrapers for AI/ML platforms (Hugging Face, Ollama)
Added scrapers for app stores and marketplaces (VS Code Marketplace, JetBrains Marketplace, Firefox Add-ons, Open VSX, Flathub, F-Droid, Snapcraft)
Added scrapers for business data (SEC EDGAR, OpenCorporates, CoinGecko)
Added scrapers for reference sources (Wikipedia, Wikidata, OpenLibrary, Choose a License)
Fixed session persistence to call fsync before renaming temp file for durability
Fixed duplicate persistence error logging by tracking whether error was already reported
Fixed byte counting in task output truncation to correctly handle multi-byte Unicode characters
Fixed parallel task execution to propagate abort signals and fail fast on first error
Fixed task worker abort handling to properly clean up on cancellation
Fixed parallel task execution to fail fast on first error instead of waiting for all workers
Fixed byte counting in task output truncation to handle multi-byte Unicode characters correctly

Changed

Changed Ctrl+P to cycle through role models (slow → default → smol) instead of all available models
Changed Shift+Ctrl+P to cycle role models temporarily (not persisted)
Changed Extension Control Center to scale with terminal height instead of fixed 25-line limit
Changed review command to parse git diff upfront and provide structured context to reviewer agents
Changed session persistence to use structured logging instead of console.error for persistence failures
Changed find tool to use fd command for .gitignore discovery instead of Bun.Glob for better abort handling
Changed LSP config loading to only mark overrides when servers are actually defined
Changed task tool to require explicit task id field instead of auto-generating names from agent type
Changed grep and find tools to use native Bun file APIs instead of Node.js fs module for improved performance
Changed YouTube scraper to use async command execution with proper stream handling
Improved rust-analyzer diagnostic polling to use version-based stability detection instead of time-based delays
Changed theme icons for extension types to use Unicode symbols (✧, ⚒) instead of text abbreviations (SK, TL, MCP)
Changed task tool to use short CamelCase task IDs instead of agent-based naming (e.g., 'SessionStore' instead of 'explore_0')
Changed task tool to accept single agent parameter at top level instead of per-task agent specification
Changed reviewer agent to use complete tool instead of submit_review for finishing reviews
Changed theme icons for extensions to use Unicode symbols instead of text abbreviations
Changed LSP file type matching to support exact filename matches in addition to extensions
Improved rust-analyzer diagnostic polling to use version-based stability detection
Refactored web-fetch tool to use modular scraper architecture for improved maintainability

Removed

Removed submit_review tool - reviewers now finish via complete tool with structured output

[3.30.0] - 2026-01-07

Added

Added environment variable configuration for task limits: OMP_TASK_MAX_PARALLEL, OMP_TASK_MAX_CONCURRENCY, OMP_TASK_MAX_OUTPUT_BYTES, OMP_TASK_MAX_OUTPUT_LINES, and OMP_TASK_MAX_AGENTS_IN_DESCRIPTION
Added specialized web-fetch handlers for 50+ platforms including GitHub, GitLab, npm, PyPI, crates.io, Stack Overflow, Wikipedia, arXiv, PubMed, Hacker News, Reddit, Mastodon, Bluesky, and many more
Added automatic yt-dlp installation for YouTube transcript extraction
Added YouTube video support with automatic transcript extraction via yt-dlp

Changed

Changed task executor to gracefully handle worker termination with proper cleanup and timeout handling

Fixed

Fixed Lobsters front page handler to use correct API endpoint (/hottest.json instead of invalid .json)
Fixed task worker error handling to prevent hanging on worker crashes, uncaught errors, and unhandled rejections
Fixed double-stringified JSON output from subagents being returned as escaped strings instead of parsed objects
Fixed markitdown tool installation to use automatic tool installer instead of requiring manual installation

[3.25.0] - 2026-01-07

Added

Added complete tool for structured subagent output with JSON schema validation
Added query parameter to output tool for jq-like JSON querying
Added output_schema parameter to task tool for structured subagent completion
Added JTD (JSON Type Definition) to JSON Schema converter for schema flexibility
Added memorable two-word task identifiers (e.g., SwiftFalcon) for better task tracking

Changed

Changed task output IDs from agent_index format to memorable names for easier reference
Changed subagent completion flow to require explicit complete tool call with retry reminders
Simplified worker agent system prompt to be more concise and focused

[3.24.0] - 2026-01-07

Added

Added ToolSession interface to unify tool creation with session context including cwd, UI availability, and rulebook rules
Added Bun Worker-based execution for subagent tasks, replacing subprocess spawning for improved performance and event streaming
Added toolNames option to filter which built-in tools are included in agent sessions
Added BUILTIN_TOOLS registry constant for programmatic access to available tool factories
Added unit tests for createTools function covering tool filtering and conditional tool creation

Changed

Changed subagent execution from spawning separate omp processes to running in Bun Workers with direct event streaming
Changed tool factories to accept ToolSession parameter instead of separate cwd and options arguments
Changed createTools to return tools as a Map and support conditional tool creation based on session context
Changed system prompt builder to dynamically generate tool descriptions from the tool registry
Changed task tool description to be generated from a template with dynamic agent list injection
Changed tool creation to use a unified ToolSession interface instead of separate parameters for cwd, options, and callbacks
Changed createTools to return tools as a Map instead of an array for consistent tool registry access
Changed system prompt builder to receive tool registry Map for dynamic tool description generation
Changed subprocess usage tracking to accumulate incrementally from message_end events rather than parsing stored events after completion

Removed

Removed browser embedded agent from task tool agent discovery
Removed recursive property from agent definitions
Removed environment variables OMP_NO_SUBAGENTS, OMP_BLOCKED_AGENT, and OMP_SPAWNS for subagent control
Removed pre-instantiated tool exports (readTool, bashTool, editTool, writeTool, grepTool, findTool, lsTool) in favor of factory functions
Removed createCodingTools and createReadOnlyTools helper functions
Removed codingTools and readOnlyTools convenience exports
Removed wrapToolsWithExtensions function from extensions API
Removed hidden property support from custom tools
Removed subagent and question custom tool examples

Fixed

Fixed memory accumulation in task subprocess by streaming events directly to disk instead of storing in memory
Fixed session persistence to exclude transient streaming data (partialJson, jsonlEvents) that was causing unnecessary storage bloat
Fixed createTools respecting explicit tool lists instead of returning all non-hidden tools

[3.21.0] - 2026-01-06

Changed

Switched from local @oh-my-pi/pi-ai to upstream @mariozechner/pi-ai package
Refactored tool renderers to be co-located with their respective tool implementations for improved code organization
Changed web search to try all configured providers in sequence with fallback before reporting errors
Changed default Anthropic web search model from claude-sonnet-4-5-20250514 to claude-haiku-4-5
Changed read tool to show first 50KB of oversized lines instead of directing users to bash sed
Changed web_fetch to use Bun.which() instead of spawning which/where for command detection
Changed web_fetch to check Content-Length header before downloading to reject oversized files early
Changed generate_image tool to save images to temp files and report paths instead of inline base64
Changed system prompt with tool usage guidance (ground answers with tools, minimize context, iterate on results)
Changed Task tool prompt with plan-then-execute guidance and output tool hints
Changed edit tool success message to report count when replacing multiple occurrences with all: true
Changed default image generation model to gemini-3-pro-image-preview
Changed error message for multiple occurrences to suggest using all: true option
Changed web_fetch tool label from web_fetch to Web Fetch for improved display
Changed argument substitution order in slash commands to process positional args ($1, $2) before wildcards ($@, $ARGUMENTS) to prevent re-substitution issues
Changed image tool name from gemini_image to generate_image with label GenerateImage

Added

Added webSearchProvider setting to override auto-detection priority (Exa > Perplexity > Anthropic)
Added imageProvider setting to override auto-detection priority (OpenRouter > Gemini)
Added git.enabled setting to enable/disable the structured git tool
Added offset and limit parameters to Output tool for paginated reading of large outputs
Added provider fallback chain for web search that tries all configured providers before failing
Added bash interceptor rule to block git commands when structured git tool is enabled
Added validation requiring message parameter for git commit operations (prevents interactive editor)
Added output ID hints in multi-agent Task results pointing to Output tool for full logs
Added fuzzy matching support for all: true mode in edit tool, enabling replacement of similar text blocks with whitespace differences
Added all parameter to edit tool for replacing all occurrences instead of requiring unique matches
Added OpenRouter support for image generation when OPENROUTER_API_KEY is set
Added ImageMagick fallback for image processing when sharp module is unavailable
Added slash commands to the extensions inspector panel for visibility and management
Added support for file-based slash commands from commands/ directories
Added $ARGUMENTS placeholder for slash command argument substitution, aligning with Claude and Codex conventions

Fixed

Fixed read tool markitdown truncation message using broken template string (missing ${ around format call)
Fixed web_fetch URL normalization order to run before special handlers
Fixed TUI image display for generate_image tool by sourcing images from details.images in addition to content blocks
Fixed context file preview in inspector panel to display content correctly instead of attempting async file reads
Fixed Linux ARM64 installs failing on fresh Debian when the sharp module is unavailable during session image compression

[3.20.1] - 2026-01-06

Fixed

Fixed find tool failing to match patterns with path separators (e.g., reports/**) by enabling full-path matching in fd

Changed

Changed multi-task display to show task descriptions instead of agent names when available
Changed ls tool to show relative modification times (e.g., "2d ago", "just now") for each entry

[3.20.0] - 2026-01-06

Added

Added extensions API with auto-discovery (.omp/extensions) and --extension/-e loading for custom tools, commands, and lifecycle hooks
Added prompt templates loaded from global and project .omp/prompts directories with /template expansion in the input box
Built-in provider overrides in models.json: override just baseUrl to route a built-in provider through a proxy while keeping all its models, or define models to fully replace the provider
Shell commands without context contribution: use !!command to execute a bash command that is shown in the TUI and saved to session history but excluded from LLM context. Useful for running commands you don't want the AI to see
Added VoiceSupervisor class for realtime voice mode using OpenAI Realtime API with continuous mic streaming and semantic VAD turn detection
Added VoiceController class for steering user input and deciding presentation of assistant responses
Added echo suppression and noise floor filtering for microphone input during voice playback
Added fallback transcript handling when realtime assistant produces no tool call or audio output
Added voice progress notifications that speak partial results after 15 seconds of streaming
Added platform-specific audio tool detection with helpful installation instructions for missing tools
Added realtime voice mode using OpenAI gpt-5-realtime with continuous mic streaming, interruptible input, and supervisor-controlled spoken updates
Added gemini_image tool for Gemini Nano Banana image generation when GEMINI_API_KEY (or GOOGLE_API_KEY) is set
Added description field to task tool for displaying short user-facing summaries in progress output
Added getApiKeyForProvider() method to ModelRegistry for retrieving API keys by provider name
Added voice settings configuration for transcription model, TTS model, voice, and audio format
Added shared render utilities module with standardized formatting functions for truncation, byte/token/duration display, and tree rendering
Added resolveOmpCommand() helper to resolve subprocess command from environment or entry point
Added /background (or /bg) command to detach UI and continue agent execution in the background
Added completion notification system with configurable methods (bell, osc99, osc9, auto, off) when agent finishes
Added completionNotification setting to configure how the agent notifies on completion
Added OMP_NOTIFICATIONS environment variable to suppress notifications globally
Added /wt slash command for git worktree management with create, list, merge, remove, status, spawn, and parallel operations
Added worktree library with collapse strategies (simple, merge-base, rebase) for merging changes between worktrees
Added worktree session tracking for managing agent tasks across isolated worktrees
Added structured git tool with safety guards, caching, and GitHub operations
Added cycleRoleModels() method to cycle through configured role-based models in a fixed order with deduplication
Added language-specific file icons to LSP diagnostics output showing file locations
Added language-specific file icon to edit tool header display

Changed

Changed voice mode toggle from Caps Lock to Ctrl+Y with auto-send on silence behavior
Changed default TTS model from gpt-4o-mini-tts to tts-1
Changed voice mode description to reflect realtime input/output with auto-send on silence
Updated hotkeys help to show Ctrl+Y for voice mode toggle instead of Caps Lock
Voice mode now uses OpenAI Realtime (gpt-5-realtime) with Ctrl+Y toggle and auto-send on silence
Updated web search tool to support auto as explicit provider option for auto-detection
Standardized tool result rendering across grep, find, ls, notebook, ask, output, and web search tools with consistent tree formatting and expand hints
Updated grep and find tool output to display language-specific icons for files and folder icons for directories
Updated file listing to display language-specific icons based on file extension instead of generic file icons

Fixed

Fixed task tool race condition where subprocess stdout events were skipped due to resolved flag being set before stream readers finished, causing completed tasks to display "0 tools · 0 tokens"
/model selector now opens instantly instead of waiting for OAuth token refresh. Token refresh is deferred until a model is actually used
Fixed cross-platform browser opening to work on Windows (via cmd /c start) and fail gracefully when unavailable

[3.15.1] - 2026-01-05

Added

Added 65 new built-in color themes including dark variants (abyss, aurora, cavern, copper, cosmos, eclipse, ember, equinox, lavender, lunar, midnight, nebula, rainforest, reef, sakura, slate, solstice, starfall, swamp, taiga, terminal, tundra, twilight, volcanic), light variants (aurora-day, canyon, cirrus, coral, dawn, dunes, eucalyptus, frost, glacier, haze, honeycomb, lagoon, lavender, meadow, mint, opal, orchard, paper, prism, sand, savanna, soleil, wetland, zenith), and material themes (alabaster, amethyst, anthracite, basalt, birch, graphite, limestone, mahogany, marble, obsidian, onyx, pearl, porcelain, quartz, sandstone, titanium)

Fixed

Fixed status line end cap rendering to properly apply background colors and use correct powerline separator characters

[3.15.0] - 2026-01-05

Added

Added spinner type variants (status and activity) with distinct animation frames per symbol preset
Added animated spinner for task tool progress display during subagent execution
Added language/file type icons for read tool output with support for 35+ file types
Added async cleanup registry for graceful session flush on SIGINT, SIGTERM, and SIGHUP signals
Added subagent token usage aggregation to session statistics and task tool results
Added streaming NDJSON writer for session persistence with proper backpressure handling
Added flush() method to SessionManager for explicit control over pending write completion
Added /exit slash command to exit the application from interactive mode
Added fuzzy path matching suggestions when read tool encounters file-not-found errors, showing closest matches using Levenshtein distance
Added status.shadowed symbol for theme customization to properly indicate shadowed extension state
Added Biome CLI-based linter client as alternative to LSP for more reliable diagnostics
Added LinterClient interface for pluggable formatter/linter implementations
Added status line segment editor for arranging and toggling status line components
Added status line presets (default, minimal, compact, developer, balanced) for quick configuration
Added status line separator styles (powerline, powerline-thin, arrow, slash, pipe, space)
Added configurable status line segments including time, hostname, and subagent count
Added symbol customization via theme overrides for icons, separators, and glyphs
Added 30+ built-in color themes including Catppuccin, Dracula, Nord, Gruvbox, Tokyo Night, and more
Added configurable status line with customizable segments, presets, and separators
Added status line segment editor for arranging and toggling status line components
Added symbol preset setting to switch between Unicode, Nerd Font, and ASCII glyphs
Added file size limit (20MB) for image files to prevent memory issues during serialization

Changed

Changed isError property in tool result events to be optional instead of required
Changed SessionManager.open() and SessionManager.continueRecent() to async methods for proper initialization
Changed session file writes to use atomic rename pattern with fsync for crash-safe persistence
Changed read tool display to show file type icons and metadata inline with path
Changed AgentSession.dispose() to async method that flushes pending writes before cleanup
Changed read tool result display to hide content by default with expand hint, showing only metadata until expanded
Changed diagnostics display to group messages by file with tree structure and severity icons
Changed diff stats formatting to use colored +/- indicators with slash separators
Changed session persistence to use streaming writes instead of synchronous file appends for better performance
Changed read tool to automatically redirect to ls when given a directory path instead of a file
Changed tool description prompts to be more concise with clearer usage guidelines and structured formatting
Moved tool description prompts from inline strings to external markdown files in src/prompts/tools/ directory for better maintainability
Changed Exa web search provider from MCP protocol to direct REST API for simpler integration
Changed web search result rendering to handle malformed response data with fallback text display
Changed compaction prompts to preserve tool outputs, command results, and repository state in context summaries
Changed init prompt to include runtime/tooling preferences section and improved formatting guidelines
Changed reviewer prompt to require evidence-backed findings anchored to diff hunks with stricter suggestion block formatting
Changed system prompt to include explicit core behavior guidelines for task completion and progress updates
Changed task prompt to emphasize end-to-end task completion and tool verification
Moved all prompt templates from inline strings to external markdown files in src/prompts/ directory for better maintainability
Changed tool result renderers to use structured tree layouts with consistent expand hints and truncation indicators
Changed grep, find, and ls tools to show scope path and detailed truncation reasons in output
Changed web search and web fetch result rendering to display structured metadata sections with bounded content previews
Changed task/subagent progress rendering to use badge-style status labels and structured output sections
Changed notebook tool to display cell content preview with line counts
Changed ask tool result to show checkbox-style selection indicators
Changed output tool to include provenance metadata and content previews for retrieved outputs
Changed collapsed tool views to show consistent "Ctrl+O to expand" hints with remaining item counts
Changed Biome integration to use CLI instead of LSP to avoid stale diagnostics issues
Changed hardcoded UI symbols throughout codebase to use theme-configurable glyphs
Changed tree drawing characters to use theme-defined box-drawing symbols
Changed status line rendering to support left/right segment positioning with separators
Changed hardcoded UI symbols to use theme-configurable glyphs throughout the interface
Changed tree drawing characters to use theme-defined box-drawing symbols
Changed CLI image attachments to resize if larger than 2048px (fit within 1920x1080) and convert >2MB images to JPEG

Removed

Removed custom renderers for ls, find, and grep tools in favor of generic tool display

Fixed

Fixed spinner animation crash when spinner frames array is empty by adding length check
Fixed session persistence to properly await all queued writes before closing or switching sessions
Fixed session persistence to truncate oversized content blocks before writing to prevent memory exhaustion
Fixed extension list and inspector panel to use correct symbols for disabled and shadowed states instead of reusing unrelated status icons
Fixed token counting for subagent progress to handle different usage object formats (camelCase and snake_case)
Fixed image file handling by adding 20MB size limit to prevent memory issues during serialization
Fixed session persistence to truncate oversized entries before writing JSONL to prevent out-of-memory errors

[3.14.0] - 2026-01-04

Added

Added getUsageStatistics() method to SessionManager for tracking cumulative token usage and costs across session messages

Changed

Changed status line to display usage statistics more efficiently by using centralized session statistics instead of recalculating from entries

[3.9.1337] - 2026-01-04

Changed

Changed default for lsp.formatOnWrite setting from true to false
Updated status line thinking level display to use emoji icons instead of abbreviated text
Changed auto-compact indicator from "(auto)" text to icon

Fixed

Fixed status line not updating token counts and cost after starting a new session
Fixed stale diagnostics persisting after file content changes in LSP client

[3.8.1337] - 2026-01-04

Added

Added automatic browser opening after exporting session to HTML
Added automatic browser opening after sharing session as a Gist

Fixed

Fixed session titles not persisting to file when set before first flush

[3.7.1337] - 2026-01-04

Added

Added EditMatchError class for structured error handling in edit operations
Added utils module export with once and untilAborted helper functions
Added in-memory LSP content sync via syncContent and notifySaved client methods

Changed

Refactored LSP integration to use writethrough callbacks for edit and write tools, improving performance by syncing content in-memory before disk writes
Simplified FileDiagnosticsResult interface with renamed fields: diagnostics → messages, hasErrors → errored, serverName → server
Session title generation now triggers before sending the first message rather than after agent work begins

Fixed

Fixed potential text decoding issues in bash executor by using streaming TextDecoder instead of Buffer.toString()

[3.5.1337] - 2026-01-03

Added

Added session header and footer output in text mode showing version, model, provider, thinking level, and session ID
Added Extension Control Center dashboard accessible via /extensions command for unified management of all providers and extensions
Added ability to enable/disable individual extensions with persistent settings
Added three-column dashboard layout with sidebar tree, extension list, and inspector panel
Added fuzzy search filtering for extensions in the dashboard
Added keyboard navigation with Tab to cycle panes, j/k for navigation, Space to toggle, Enter to expand/collapse

Changed

Redesigned Extension Control Center from 3-column layout to tabbed interface with horizontal provider tabs and 2-column grid
Replaced sidebar tree navigation with provider tabs using TAB/Shift+TAB cycling

Fixed

Fixed title generation flag not resetting when starting a new session

[3.4.1337] - 2026-01-03

Added

Added Time Traveling Stream Rules (TTSR) feature that monitors agent output for pattern matches and injects rule reminders mid-stream
Added ttsr_trigger frontmatter field for rules to define regex patterns that trigger mid-stream injection
Added TTSR settings for enabled state, context mode (keep/discard partial output), and repeat mode (once/after-gap)

Fixed

Fixed excessive subprocess spawns by caching git status for 1 second in the footer component

[3.3.1337] - 2026-01-03

Changed

Improved /status command output formatting to use consistent column alignment across all sections
Updated version update notification to suggest omp update instead of manual npm install command

[3.1.1337] - 2026-01-03

Added

Added spawns frontmatter field for agent definitions to control which sub-agents can be spawned
Added spawn restriction enforcement preventing agents from spawning unauthorized sub-agents

Fixed

Fixed duplicate skill loading when the same SKILL.md file was discovered through multiple paths

[3.0.1337] - 2026-01-03

Added

Added unified capability-based discovery system for loading configuration from multiple AI coding tools (Claude Code, Cursor, Windsurf, Gemini, Codex, Cline, GitHub Copilot, VS Code)
Added support for discovering MCP servers, rules, skills, hooks, tools, slash commands, prompts, and context files from tool-specific config directories
Added Discovery settings tab in interactive mode to enable/disable individual configuration providers
Added provider source attribution showing which tool contributed each configuration item
Added support for Cursor MDC rule format with frontmatter (description, globs, alwaysApply)
Added support for Windsurf rules from .windsurf/rules/*.md and global_rules.md
Added support for Cline rules from .clinerules file or directory
Added support for GitHub Copilot instructions with applyTo glob patterns
Added support for Gemini extensions and system.md customization files
Added support for Codex AGENTS.md and config.toml settings
Added automatic migration of PI_* environment variables to OMP_* equivalents for backwards compatibility
Added multi-path config discovery supporting .omp, .pi, and .claude directories with priority ordering
Added getConfigDirPaths(), findConfigFile(), and readConfigFile() functions for unified config resolution
Added documentation for config module usage patterns

Changed

Changed MCP tool name parsing to use last underscore separator for better server name handling
Changed /config output to show provider attribution for discovered items
Renamed CLI binary from pi to omp and updated all command references
Changed config directory from .pi to .omp with fallback support for legacy paths
Renamed environment variables from PI_* to OMP_* prefix (e.g., OMP_SMOL_MODEL, OMP_SLOW_MODEL)
Changed model role alias prefix from pi/ to omp/ (e.g., omp/slow instead of pi/slow)

[2.1.1337] - 2026-01-03

Added

Added pi update command to check for and install updates from GitHub releases or via bun

Changed

Changed HTML export to use compile-time bundled templates via Bun macros for improved performance
Changed exportToHtml and exportFromFile functions to be async
Simplified build process by embedding assets (themes, templates, agents, commands) directly into the binary at compile time
Removed separate asset copying steps from build scripts

[2.0.1337] - 2026-01-03

Added

Added shell environment snapshot to preserve user aliases, functions, and shell options when executing bash commands
Added support for PI_BASH_NO_CI, PI_BASH_NO_LOGIN, and PI_SHELL_PREFIX environment variables for shell customization
Added zsh support alongside bash for shell detection and configuration

Changed

Changed shell detection to prefer user's $SHELL when it's bash or zsh, with improved fallback path resolution
Changed Edit tool to reject .ipynb files with guidance to use NotebookEdit tool instead

[1.500.0] - 2026-01-03

Added

Added provider tabs to model selector with Tab/Arrow navigation for filtering models by provider
Added context menu to model selector for choosing model role (Default, Smol, Slow) instead of keyboard shortcuts
Added LSP diagnostics display in tool execution output showing errors and warnings after file edits
Added centralized file logger with daily rotation to ~/.pi/logs/ for debugging production issues
Added logger property to hook and custom tool APIs for error/warning/debug logging
Added output tool to read full agent/task outputs by ID when truncated previews are insufficient
Added task tool to reviewer agent, enabling parallel exploration of large codebases during reviews
Added subprocess tool registry for extracting and rendering tool data from subprocess agents in real-time
Added combined review result rendering showing verdict and findings in a tree structure
Auto-read file mentions: Reference files with @path/to/file.ext syntax in prompts to automatically inject their contents, eliminating manual Read tool calls
Added hidden property for custom tools to exclude them from default tool list unless explicitly requested
Added explicitTools option to createAgentSession for enabling hidden tools by name
Added example review tools (report_finding, submit_review) with structured findings accumulation and verdict rendering
Added /review example command for interactive code review with branch comparison, uncommitted changes, and commit review modes
Custom TypeScript slash commands: Create programmable commands at ~/.pi/agent/commands/[name]/index.ts or .pi/commands/[name]/index.ts. Commands export a factory returning { name, description, execute(args, ctx) }. Return a string to send as LLM prompt, or void for fire-and-forget actions. Full access to HookCommandContext for UI dialogs, session control, and shell execution.
Claude command directories: Markdown slash commands now also load from ~/.claude/commands/ and .claude/commands/ (parallel to existing .pi/commands/ support)
commands.enableClaudeUser and commands.enableClaudeProject settings to disable Claude command directory loading
/export --copy option to copy entire session as formatted text to clipboard

Changed

Changed model selector keyboard shortcuts from S/L keys to a context menu opened with Enter
Changed model role indicators from symbols (✓ ⚡ 🧠) to labeled badges ([ DEFAULT ] [ SMOL ] [ SLOW ])
Changed model list sorting to include secondary sort by model ID within each provider
Changed silent error suppression to log warnings and debug info for tool errors, theme loading, and command loading failures
Changed Task tool progress display to show agent index (e.g., reviewer(0)) for easier Output tool ID derivation
Changed Task tool output to only include file paths when Output tool is unavailable, providing Read tool fallback
Changed Task tool output references to use simpler ID format (e.g., reviewer_0) with line/char counts for Output tool integration
Changed subagent recursion prevention from blanket blocking to same-agent blocking. Non-recursive agents can now spawn other agent types (e.g., reviewer can spawn explore agents) but cannot spawn themselves.
Changed /review command from markdown to interactive TypeScript with mode selection menu (branch comparison, uncommitted changes, commit review, custom)
Changed bundled commands to be overridable by user/project commands with same name
Changed subprocess termination to wait for message_end event to capture accurate token counts
Changed token counting in subprocess to accumulate across messages instead of overwriting
Updated bundled reviewer agent to use structured review tools with priority-based findings (P0-P3) and formal verdict submission
Task tool now streams artifacts in real-time: input written before spawn, session jsonl written by subprocess, output written at completion

Removed

Removed separate Exa error logger in favor of centralized logging system
Removed findings_count parameter from submit_review tool - findings are now counted automatically
Removed artifacts location display from task tool output

Fixed

Fixed race condition in event listener iteration by copying array before iteration to prevent mutation during callbacks
Fixed potential memory leak from orphaned abort controllers by properly aborting existing controllers before replacement
Fixed stream reader resource leak by adding proper releaseLock() calls in finally blocks
Fixed hook API methods throwing clear errors when handlers are not initialized instead of silently failing
Fixed LSP client race conditions with concurrent client creation and file operations using proper locking
Fixed Task tool progress display showing stale data by cloning progress objects before passing to callbacks
Fixed Task tool missing final progress events by waiting for readline to close before resolving
Fixed RPC mode race condition with concurrent prompt commands by serializing execution
Fixed pre-commit hook race condition causing index.lock errors when GitKraken/IDE git integrations detect file changes during formatting
Fixed Task tool output artifacts (out.md) containing duplicated text from streaming updates
Fixed Task tool progress display showing repeated nearly-identical lines during streaming
Fixed Task tool subprocess model selection ignoring agent's configured model and falling back to settings default. The --model flag now accepts provider/model format directly.
Fixed Task tool showing "done + succeeded" when aborted; now correctly displays "⊘ aborted" status

[1.341.0] - 2026-01-03

Added

Added interruptMode setting to control when queued messages are processed during tool execution.
Implemented getter and setter methods in SettingsManager for interrupt mode persistence.
Exposed interruptMode configuration in interactive settings UI with immediate/wait options.
Wired interrupt mode through AgentSession and SDK to enable runtime configuration.
Model roles: Configure different models for different purposes (default, smol, slow) via /model selector
Model selector key bindings: Enter sets default, S sets smol, L sets slow, Escape closes
Model selector shows role markers: ✓ for default, ⚡ for smol, 🧠 for slow
pi/<role> model aliases in Task tool agent definitions (e.g., model: pi/smol, haiku, flash, mini)
Smol model auto-discovery using priority chain: haiku > flash > mini
Slow model auto-discovery using priority chain: gpt-5.2-codex > codex > gpt > opus > pro
CLI args for model roles: --smol <model> and --slow <model> (ephemeral, not persisted)
Env var overrides: PI_SMOL_MODEL and PI_SLOW_MODEL
Title generation now uses configured smol model from settings
LSP diagnostics on edit: Edit tool can now return LSP diagnostics after editing code files. Disabled by default to avoid noise during multi-edit sequences. Enable via lsp.diagnosticsOnEdit setting.
LSP workspace diagnostics: New lsp action=workspace_diagnostics command checks the entire project for errors. Auto-detects project type and uses appropriate checker (rust-analyzer/cargo for Rust, tsc for TypeScript, go build for Go, pyright for Python).
LSP local binary resolution: LSP servers installed in project-local directories are now discovered automatically. Checks node_modules/.bin/ for Node.js projects, .venv/bin//venv/bin/ for Python projects, and vendor/bundle/bin/ for Ruby projects before falling back to $PATH.
LSP format on write: Write tool now automatically formats code files using LSP after writing. Uses the language server's built-in formatter (e.g., rustfmt for Rust, gofmt for Go). Controlled via lsp.formatOnWrite setting (enabled by default).
LSP diagnostics on write: Write tool now returns LSP diagnostics (errors/warnings) after writing code files. This gives immediate feedback on syntax errors and type issues. Controlled via lsp.diagnosticsOnWrite setting (enabled by default).
LSP server warmup at startup: LSP servers are now started at launch to avoid cold-start delays when first writing files.
LSP server status in welcome banner: Shows which language servers are active and ready.
Edit fuzzy match setting: Added edit.fuzzyMatch setting (enabled by default) to control whether the edit tool accepts high-confidence fuzzy matches for whitespace/indentation differences. Toggle via /settings.
Multi-server LSP diagnostics: Diagnostics now query all applicable language servers for a file type. For TypeScript/JavaScript projects with Biome, this means both type errors (from tsserver) and lint errors (from Biome) are reported together.
Comprehensive LSP server configurations for 40+ languages including Rust, Go, Python, Java, Kotlin, Scala, Haskell, OCaml, Elixir, Ruby, PHP, C#, Lua, Nix, and many more. Each server includes sensible defaults for args, settings, and init options.
Extended LSP config file search paths: Now searches for lsp.json, .lsp.json in project root and .pi/ subdirectory, plus user-level configs in ~/.pi/ and home directory.

Changed

LSP settings moved to dedicated "LSP" tab in /settings for better organization
Improved grep tool description to document pagination options (headLimit, offset) and clarify recursive search behavior
LSP idle timeout now disabled by default. Configure via idleTimeoutMs in lsp.json to auto-shutdown inactive servers.
Model settings now use role-based storage (modelRoles map) instead of single defaultProvider/defaultModel fields. Supports multiple model roles (default, small, etc.)
Session model persistence now uses "provider/modelId" string format with optional role field

Fixed

Recent sessions now show in welcome banner (was never wired up).
Auto-generated session titles: Sessions are now automatically titled based on the first message using a small model (Haiku/GPT-4o-mini/Flash). Titles are shown in the terminal window title, recent sessions list, and --resume picker. The resume picker shows title with dimmed first message preview below.

[1.340.0] - 2026-01-03

Changed

Replaced vendored highlight.js and marked.js with CDN-hosted versions for smaller exports
Added runtime minification for HTML, CSS, and JS in session exports
Session share URL now uses gistpreview.github.io instead of shittycodingagent.ai

[1.339.0] - 2026-01-03

Added

MCP project config setting to disable loading .mcp.json/mcp.json from project root
Support for both mcp.json and .mcp.json filenames (prefers mcp.json if both exist)
Automatic Exa MCP server filtering with API key extraction for native integration

[1.338.0] - 2026-01-03

Added

Bash interceptor setting to block shell commands that have dedicated tools (disabled by default, enable via /settings)

Changed

Refactored settings UI to declarative definitions for easier maintenance
Shell detection now respects $SHELL environment variable before falling back to bash/sh
Tool binary detection now uses Bun.which() instead of spawning processes

Fixed

CLI help text now accurately lists all default tools

[1.337.1] - 2026-01-02

Added

MCP support and plugin system for external tool integration
Git context to system prompt for repo awareness
Bash interception to guide tool selection
Fuzzy matching to handle indentation variance in edit tool
Specialized Exa tools with granular toggles
/share command for exporting conversations to HTML
Edit diff preview before tool execution

Changed

Renamed package scope to @oh-my-pi for consistent branding
Simplified toolset and enhanced navigation
Improved process cleanup with tree kill
Updated CI/CD workflows for GitHub Actions with provenance-signed npm publishing

Fixed

Template string interpolation in image read output
Prevented full re-renders during write tool streaming
Edit tool failing on files with UTF-8 BOM

[1.337.0] - 2026-01-02

Initial release under @oh-my-pi scope. See previous releases at badlogic/pi-mono.

[1.5.1] - 2026-01-03

Added

Added shell environment snapshot to preserve user aliases, functions, and shell options when executing bash commands
Added support for PI_BASH_NO_CI, PI_BASH_NO_LOGIN, and PI_SHELL_PREFIX environment variables for shell customization
Added zsh support alongside bash for shell detection and configuration

Changed

Changed shell detection to prefer user's $SHELL when it's bash or zsh, with improved fallback path resolution
Changed Edit tool to reject .ipynb files with guidance to use NotebookEdit tool instead

[1.5.0] - 2026-01-03

Added

Added provider tabs to model selector with Tab/Arrow navigation for filtering models by provider
Added context menu to model selector for choosing model role (Default, Smol, Slow) instead of keyboard shortcuts
Added LSP diagnostics display in tool execution output showing errors and warnings after file edits
Added centralized file logger with daily rotation to ~/.pi/logs/ for debugging production issues
Added logger property to hook and custom tool APIs for error/warning/debug logging
Added output tool to read full agent/task outputs by ID when truncated previews are insufficient
Added task tool to reviewer agent, enabling parallel exploration of large codebases during reviews
Added subprocess tool registry for extracting and rendering tool data from subprocess agents in real-time
Added combined review result rendering showing verdict and findings in a tree structure
Auto-read file mentions: Reference files with @path/to/file.ext syntax in prompts to automatically inject their contents, eliminating manual Read tool calls
Added hidden property for custom tools to exclude them from default tool list unless explicitly requested
Added explicitTools option to createAgentSession for enabling hidden tools by name
Added example review tools (report_finding, submit_review) with structured findings accumulation and verdict rendering
Added /review example command for interactive code review with branch comparison, uncommitted changes, and commit review modes
Custom TypeScript slash commands: Create programmable commands at ~/.pi/agent/commands/[name]/index.ts or .pi/commands/[name]/index.ts. Commands export a factory returning { name, description, execute(args, ctx) }. Return a string to send as LLM prompt, or void for fire-and-forget actions. Full access to HookCommandContext for UI dialogs, session control, and shell execution.
Claude command directories: Markdown slash commands now also load from ~/.claude/commands/ and .claude/commands/ (parallel to existing .pi/commands/ support)
commands.enableClaudeUser and commands.enableClaudeProject settings to disable Claude command directory loading
/export --copy option to copy entire session as formatted text to clipboard

Changed

Changed model selector keyboard shortcuts from S/L keys to a context menu opened with Enter
Changed model role indicators from symbols (✓ ⚡ 🧠) to labeled badges ([ DEFAULT ] [ SMOL ] [ SLOW ])
Changed model list sorting to include secondary sort by model ID within each provider
Changed silent error suppression to log warnings and debug info for tool errors, theme loading, and command loading failures
Changed Task tool progress display to show agent index (e.g., reviewer(0)) for easier Output tool ID derivation
Changed Task tool output to only include file paths when Output tool is unavailable, providing Read tool fallback
Changed Task tool output references to use simpler ID format (e.g., reviewer_0) with line/char counts for Output tool integration
Changed subagent recursion prevention from blanket blocking to same-agent blocking. Non-recursive agents can now spawn other agent types (e.g., reviewer can spawn explore agents) but cannot spawn themselves.
Changed /review command from markdown to interactive TypeScript with mode selection menu (branch comparison, uncommitted changes, commit review, custom)
Changed bundled commands to be overridable by user/project commands with same name
Changed subprocess termination to wait for message_end event to capture accurate token counts
Changed token counting in subprocess to accumulate across messages instead of overwriting
Updated bundled reviewer agent to use structured review tools with priority-based findings (P0-P3) and formal verdict submission
Task tool now streams artifacts in real-time: input written before spawn, session jsonl written by subprocess, output written at completion

Removed

Removed separate Exa error logger in favor of centralized logging system
Removed findings_count parameter from submit_review tool - findings are now counted automatically
Removed artifacts location display from task tool output

Fixed

Fixed race condition in event listener iteration by copying array before iteration to prevent mutation during callbacks
Fixed potential memory leak from orphaned abort controllers by properly aborting existing controllers before replacement
Fixed stream reader resource leak by adding proper releaseLock() calls in finally blocks
Fixed hook API methods throwing clear errors when handlers are not initialized instead of silently failing
Fixed LSP client race conditions with concurrent client creation and file operations using proper locking
Fixed Task tool progress display showing stale data by cloning progress objects before passing to callbacks
Fixed Task tool missing final progress events by waiting for readline to close before resolving
Fixed RPC mode race condition with concurrent prompt commands by serializing execution
Fixed pre-commit hook race condition causing index.lock errors when GitKraken/IDE git integrations detect file changes during formatting
Fixed Task tool output artifacts (out.md) containing duplicated text from streaming updates
Fixed Task tool progress display showing repeated nearly-identical lines during streaming
Fixed Task tool subprocess model selection ignoring agent's configured model and falling back to settings default. The --model flag now accepts provider/model format directly.
Fixed Task tool showing "done + succeeded" when aborted; now correctly displays "⊘ aborted" status

[0.50.1] - 2026-01-26

Fixed

Git extension updates now handle force-pushed remotes gracefully instead of failing (#961 by @aliou)
Extension ctx.newSession({ setup }) now properly syncs agent state and renders messages after setup callback runs (#968)
Fixed extension UI bindings not initializing when starting with no extensions, which broke UI methods after /reload
Fixed /hotkeys output to title-case extension hotkeys (#969 by @Perlence)
Fixed model catalog generation to exclude deprecated OpenCode Zen models (#970 by @DanielTatarkin)
Fixed git extension removal to prune empty directories

[0.50.0] - 2026-01-26

New Features

Pi packages for bundling and installing extensions, skills, prompts, and themes. See docs/packages.md.
Hot reload (/reload) of resources including AGENTS.md, SYSTEM.md, APPEND_SYSTEM.md, prompt templates, skills, themes, and extensions. See README.md#commands and README.md#context-files.
Custom providers via pi.registerProvider() for proxies, custom endpoints, OAuth or SSO flows, and non-standard streaming APIs. See docs/custom-provider.md.
Azure OpenAI Responses provider support with deployment-aware model mapping. See docs/providers.md#azure-openai.
OpenRouter routing support for custom models via openRouterRouting. See docs/providers.md#api-keys and docs/models.md.
Skill invocation messages are now collapsible and skills can opt out of model invocation via disable-model-invocation. See docs/skills.md#frontmatter.
Session selector renaming and configurable keybindings. See README.md#commands and docs/keybindings.md.
models.json headers can resolve environment variables and shell commands. See docs/models.md#value-resolution.
--verbose CLI flag to override quiet startup. See README.md#cli-reference.

Read the fully revamped docs in README.md, or have your clanker read them for you.

SDK Migration Guide

There are multiple SDK breaking changes since v0.49.3. For the quickest migration, point your agent at packages/coding-agent/docs/sdk.md, the SDK examples in packages/coding-agent/examples/sdk, and the SDK source in packages/coding-agent/src/core/sdk.ts and related modules.

Breaking Changes

Header values in models.json now resolve environment variables (if a header value matches an env var name, the env var value is used). This may change behavior if a literal header value accidentally matches an env var name. (#909)
External packages (npm/git) are now configured via packages array in settings.json instead of extensions. Existing npm:/git: entries in extensions are auto-migrated. (#645)
Resource loading now uses ResourceLoader only and settings.json uses arrays for extensions, skills, prompts, and themes (#645)
Removed discoverAuthStorage and discoverModels from the SDK. AuthStorage and ModelRegistry now default to ~/.pi/agent paths unless you pass an agentDir (#645)

Added

Session renaming in /resume picker via Ctrl+R without opening the session (#863 by @svkozak)
Session selector keybindings are now configurable (#948 by @aos)
disable-model-invocation frontmatter field for skills to prevent agentic invocation while still allowing explicit /skill:name commands (#927)
Exposed copyToClipboard utility for extensions (#926 by @mitsuhiko)
Skill invocation messages are now collapsible in chat output, showing collapsed by default with skill name and expand hint (#894)
Header values in models.json now support environment variables and shell commands, matching apiKey resolution (#909)
Added HTTP proxy environment variable support for API requests (#942 by @haoqixu)
Added OpenRouter provider routing support for custom models via openRouterRouting compat field (#859 by @v01dpr1mr0s3)
Added azure-openai-responses provider support for Azure OpenAI Responses API. (#890 by @markusylisiurunen)
Added changelog link to update notifications (#925 by @dannote)
Added --verbose CLI flag to override quietStartup setting (#906 by @Perlence)
markdown.codeBlockIndent setting to customize code block indentation in rendered output
Extension package management with pi install, pi remove, pi update, and pi list commands (#645)
Package filtering: selectively load resources from packages using object form in packages array (#645)
Glob pattern support with minimatch in package filters, top-level settings arrays, and pi manifest (e.g., "!funky.json", "*.ts") (#645)
/reload command to reload extensions, skills, prompts, and themes (#645)
pi config command with TUI to enable/disable package and top-level resources via patterns (#938)
CLI flags for --skill, --prompt-template, --theme, --no-prompt-templates, and --no-themes (#645)
Package deduplication: if same package appears in global and project settings, project wins (#645)
Unified collision reporting with ResourceDiagnostic type for all resource types (#645)
Show provider alongside the model in the footer if multiple providers are available
Custom provider support via pi.registerProvider() with streamSimple for custom API implementations
Added custom-provider.ts example extension demonstrating custom Anthropic provider with OAuth

Changed

/resume picker sort toggle moved to Ctrl+S to free Ctrl+R for rename (#863 by @svkozak)
HTML export: clicking a sidebar message now navigates to its newest leaf and scrolls to it, instead of truncating the branch (#853 by @mitsuhiko)
HTML export: active path is now visually highlighted with dimmed off-path nodes (#929 by @hewliyang)
Azure OpenAI Responses provider now uses base URL configuration with deployment-aware model mapping and no longer includes service tier handling
/reload now re-renders the entire scrollback so updated extension components are visible immediately (#928 by @ferologics)
Skill, prompt template, and theme discovery now use settings and CLI path arrays instead of legacy filters (#645)

Fixed

Extension setWorkingMessage() calls in agent_start handlers now work correctly; previously the message was silently ignored because the loading animation didn't exist yet (#935)
Fixed package auto-discovery to respect loader rules, config overrides, and force-exclude patterns
Fixed /reload restoring the correct editor after reload (#949 by @Perlence)
Fixed distributed themes breaking /export (#946 by @mitsuhiko)
Fixed startup hints to clarify thinking level selection and expanded thinking guidance
Fixed SDK initial model resolution to use findInitialModel and default to Claude Opus 4.5 for Anthropic models
Fixed no-models warning to include the /model instruction
Fixed authentication error messages to point to the authentication documentation
Fixed bash output hint lines to truncate to terminal width
Fixed custom editors to honor the paddingX setting (#936 by @Perlence)
Fixed system prompt tool list to show only built-in tools
Fixed package manager to check npm package versions before using cached copies
Fixed package manager to run npm install after cloning git repositories with a package.json
Fixed extension provider registrations to apply before model resolution
Fixed editor multi-line insertion handling and lastAction tracking (#945 by @Perlence)
Fixed editor word wrapping to reserve a cursor column (#934 by @Perlence)
Fixed editor word wrapping to use single-pass backtracking for whitespace handling (#924 by @Perlence)
Fixed Kitty image ID allocation and cleanup to prevent image ID collisions
Fixed overlays staying centered after terminal resizes (#950 by @nicobailon)
Fixed streaming dispatch to use the model api type instead of hardcoded API defaults
Fixed Google providers to default tool call arguments to an empty object when omitted
Fixed OpenAI Responses streaming to handle arguments.done events on OpenAI-compatible endpoints (#917 by @williballenthin)
Fixed OpenAI Codex Responses tool strictness handling after the shared responses refactor
Fixed Azure OpenAI Responses streaming to guard deltas before content parts and correct metadata and handoff gating
Fixed OpenAI completions tool-result image batching after consecutive tool results (#902 by @terrorobe)
Off-by-one error in bash output "earlier lines" count caused by counting spacing newline as hidden content (#921)
User package filters now layer on top of manifest filters instead of replacing them (#645)
Auto-retry now handles "terminated" errors from Codex API mid-stream failures
Follow-up queue (Alt+Enter) now sends full paste content instead of [paste #N ...] markers (#912)
Fixed Alt-Up not restoring messages queued during compaction (#923 by @aliou)
Fixed session corruption when loading empty or invalid session files via --session flag (#932 by @armanddp)
Fixed extension shortcuts not firing when extension also uses setEditorComponent() (#947 by @Perlence)
Session "modified" time now uses last message timestamp instead of file mtime, so renaming doesn't reorder the recent list (#863 by @svkozak)

[0.49.3] - 2026-01-22

Added

markdown.codeBlockIndent setting to customize code block indentation in rendered output (#855 by @terrorobe)
Added inline-bash.ts example extension for expanding !{command} patterns in prompts (#881 by @scutifer)
Added antigravity-image-gen.ts example extension for AI image generation via Google Antigravity (#893 by @benvargas)
Added PI_SHARE_VIEWER_URL environment variable for custom share viewer URLs (#889 by @andresaraujo)
Added Alt+Delete as hotkey for delete word forwards (#878 by @Perlence)

Changed

Tree selector: changed label filter shortcut from l to Shift+L so users can search for entries containing "l" (#861 by @mitsuhiko)
Fuzzy matching now scores consecutive matches higher for better search relevance (#860 by @mitsuhiko)

Fixed

Fixed error messages showing hardcoded ~/.pi/agent/ paths instead of respecting PI_CODING_AGENT_DIR (#887 by @aliou)
Fixed write tool not displaying errors in the UI when execution fails (#856)
Fixed HTML export using default theme instead of user's active theme (#870 by @scutifer)
Show session name in the footer and terminal / tab title (#876 by @scutifer)
Fixed 256color fallback in Terminal.app to prevent color rendering issues (#869 by @Perlence)
Fixed viewport tracking and cursor positioning for overlays and content shrink scenarios
Fixed autocomplete to allow searches with / characters (e.g., folder1/folder2) (#882 by @richardgill)
Fixed autolinked emails displaying redundant (mailto:...) suffix (#888 by @terrorobe)
Fixed @ file autocomplete adding space after directories, breaking continued autocomplete into subdirectories

[0.49.2] - 2026-01-19

Added

Added widget placement option for extension widgets via widgetPlacement in pi.addWidget() (#850 by @marckrenn)
Added AWS credential detection for ECS/Kubernetes environments: AWS_CONTAINER_CREDENTIALS_RELATIVE_URI, AWS_CONTAINER_CREDENTIALS_FULL_URI, AWS_WEB_IDENTITY_TOKEN_FILE (#848)
Add "quiet startup" setting to /settings (#847 by @unexge)

Changed

HTML export now includes JSONL download button, jump-to-last-message on click, and fixed missing labels (#853 by @mitsuhiko)
Improved error message for OAuth authentication failures (expired credentials, offline) instead of generic 'No API key found' (#849 by @zedrdave)

Fixed

Fixed /model selector scope toggle so you can switch between all and scoped models when scoped models are saved (#844)
Fixed OpenAI Responses 400 error "reasoning without following item" when replaying aborted turns (#838)
Fixed pi exiting with code 0 when cancelling resume session selection

Removed

Removed strictResponsesPairing compat option from models.json schema (no longer needed)

[0.49.1] - 2026-01-18

Added

Added strictResponsesPairing compat option for custom OpenAI Responses models on Azure (#768 by @nicobako)
Session selector (/resume) now supports path display toggle (Ctrl+P) and session deletion (Ctrl+D) with inline confirmation (#816 by @w-winter)
Added undo support in interactive mode with Ctrl+- hotkey. (#831 by @Perlence)

Changed

Share URLs now use hash fragments (#) instead of query strings (?) to prevent session IDs from being sent to buildwithpi.ai (#829 by @terrorobe)
API keys in models.json can now be retrieved via shell command using ! prefix (e.g., "apiKey": "!security find-generic-password -ws 'anthropic'" for macOS Keychain) (#762 by @cv)

Fixed

Fixed IME candidate window appearing in wrong position when filtering menus with Input Method Editor (e.g., Chinese IME). Components with search inputs now properly propagate focus state for cursor positioning. (#827)
Fixed extension shortcut conflicts to respect user keybindings when built-in actions are remapped. (#826 by @richardgill)
Fixed photon WASM loading in standalone compiled binaries.
Fixed tool call ID normalization for cross-provider handoffs (e.g., Codex to Antigravity Claude) (#821)

[0.49.0] - 2026-01-17

Added

pi.setLabel(entryId, label) in ExtensionAPI for setting per-entry labels from extensions (#806)
Export keyHint, appKeyHint, editorKey, appKey, rawKeyHint for extensions to format keybinding hints consistently (#802 by @dannote)
Exported VERSION from the package index and updated the custom-header example. (#798 by @tallshort)
Added showHardwareCursor setting to control cursor visibility while still positioning it for IME support. (#800 by @ghoulr)
Added Emacs-style kill ring editing with yank and yank-pop keybindings, plus legacy Alt+letter handling and Alt+D delete word forward support in the interactive editor. (#810 by @Perlence)
Added ctx.compact() and ctx.getContextUsage() to extension contexts for programmatic compaction and context usage checks.
Added documentation for delete word forward and kill ring keybindings in interactive mode. (#810 by @Perlence)

Changed

Updated the default system prompt wording to clarify the pi harness and documentation scope.
Simplified Codex system prompt handling to use the default system prompt directly for Codex instructions.

Fixed

Fixed photon module failing to load in ESM context with "require is not defined" error (#795 by @dannote)
Fixed compaction UI not showing when extensions trigger compaction.
Fixed orphaned tool results after errored assistant messages causing Codex API errors. When an assistant message has stopReason: "error", its tool calls are now excluded from pending tool tracking, preventing synthetic tool results from being generated for calls that will be dropped by provider-specific converters. (#812)
Fixed Bedrock Claude max_tokens handling to always exceed thinking budget tokens, preventing compaction failures. (#797 by @pjtf93)
Fixed Claude Code tool name normalization to match the Claude Code tool list case-insensitively and remove invalid mappings.

Removed

Removed pi-internal:// path resolution from the read tool.

[0.48.0] - 2026-01-16

Added

Added quietStartup setting to silence startup output (version header, loaded context info, model scope line). Changelog notifications are still shown. (#777 by @ribelo)
Added editorPaddingX setting for horizontal padding in input editor (0-3, default: 0)
Added shellCommandPrefix setting to prepend commands to every bash execution, enabling alias expansion in non-interactive shells (e.g., "shellCommandPrefix": "shopt -s expand_aliases") (#790 by @richardgill)
Added bash-style argument slicing for prompt templates (#770 by @airtonix)
Extension commands can provide argument auto-completions via getArgumentCompletions in pi.registerCommand() (#775 by @ribelo)
Bash tool now displays the timeout value in the UI when a timeout is set (#780 by @dannote)
Export getShellConfig for extensions to detect user's shell environment (#766 by @dannote)
Added thinkingText and selectedBg to theme schema (#763 by @scutifer)
navigateTree() now supports replaceInstructions option to replace the default summarization prompt entirely, and label option to attach a label to the branch summary entry (#787 by @mitsuhiko)

Fixed

Fixed crash during auto-compaction when summarization fails (e.g., quota exceeded). Now displays error message instead of crashing (#792)
Fixed --session <UUID> to search globally across projects if not found locally, with option to fork sessions from other projects (#785 by @ribelo)
Fixed standalone binary WASM loading on Linux (#784)
Fixed string numbers in tool arguments not being coerced to numbers during validation (#786 by @dannote)
Fixed --no-extensions flag not preventing extension discovery (#776)
Fixed extension messages rendering twice on startup when pi.sendMessage({ display: true }) is called during session_start (#765 by @dannote)
Fixed PI_CODING_AGENT_DIR env var not expanding tilde (~) to home directory (#778 by @aliou)
Fixed session picker hint text overflow (#764)
Fixed Kitty keyboard protocol shifted symbol keys (e.g., @, ?) not working in editor (#779 by @iamd3vil)
Fixed Bedrock tool call IDs causing API errors from invalid characters (#781 by @pjtf93)

Changed

Hardware cursor is now disabled by default for better terminal compatibility. Set PI_HARDWARE_CURSOR=1 to enable (replaces PI_NO_HARDWARE_CURSOR=1 which disabled it).

[0.47.0] - 2026-01-16

Breaking Changes

Extensions using Editor directly must now pass TUI as the first constructor argument: new Editor(tui, theme). The tui parameter is available in extension factory functions. (#732)

Added

OpenAI Codex official support: Full compatibility with OpenAI's Codex CLI models (gpt-5.1, gpt-5.2, gpt-5.1-codex-mini, gpt-5.2-codex). Features include static system prompt for OpenAI allowlisting, prompt caching via session ID, and reasoning signature retention across turns. Set OPENAI_API_KEY and use --provider openai-codex or select a Codex model. (#737)
pi-internal:// URL scheme in read tool for accessing internal documentation. The model can read files from the coding-agent package (README, docs, examples) to learn about extending pi.
New input event in extension system for intercepting, transforming, or handling user input before the agent processes it. Supports three result types: continue (pass through), transform (modify text/images), handled (respond without LLM). Handlers chain transforms and short-circuit on handled. (#761 by @nicobailon)
Extension example: input-transform.ts demonstrating input interception patterns (quick mode, instant commands, source routing) (#761 by @nicobailon)
Custom tool HTML export: extensions with renderCall/renderResult now render in /share and /export output with ANSI-to-HTML color conversion (#702 by @aliou)
Direct filter shortcuts in Tree mode: Ctrl+D (default), Ctrl+T (no-tools), Ctrl+U (user-only), Ctrl+L (labeled-only), Ctrl+A (all) (#747 by @kaofelix)

Changed

Skill commands (/skill:name) are now expanded in AgentSession instead of interactive mode. This enables skill commands in RPC and print modes, and allows the input event to intercept /skill:name before expansion.

Fixed

Editor no longer corrupts terminal display when loading large prompts via setEditorText. Content now scrolls vertically with indicators showing lines above/below the viewport. (#732)
Piped stdin now works correctly: echo foo | pi is equivalent to pi -p foo. When stdin is piped, print mode is automatically enabled since interactive mode requires a TTY (#708)
Session tree now preserves branch connectors and indentation when filters hide intermediate entries so descendants attach to the nearest visible ancestor and sibling branches align. Fixed in both TUI and HTML export (#739 by @w-winter)
Added upstream connect, connection refused, and reset before headers patterns to auto-retry error detection (#733)
Multi-line YAML frontmatter in skills and prompt templates now parses correctly. Centralized frontmatter parsing using the yaml library. (#728 by @richardgill)
ctx.shutdown() now waits for pending UI renders to complete before exiting, ensuring notifications and final output are visible (#756)
OpenAI Codex provider now retries on transient errors (429, 5xx, connection failures) with exponential backoff (#733)

[0.46.0] - 2026-01-15

Fixed

Scoped models (--models or enabledModels) now remember the last selected model across sessions instead of always starting with the first model in the scope (#736 by @ogulcancelik)
Show bun install instead of npm install in update notification when running under Bun (#714 by @dannote)
/skill prompts now include the skill path (#711 by @jblwilliams)
Use configurable expandTools keybinding instead of hardcoded Ctrl+O (#717 by @dannote)
Compaction turn prefix summaries now merge correctly (#738 by @vsabavat)
Avoid unsigned Gemini 3 tool calls (#741 by @roshanasingh4)
Fixed signature support for non-Anthropic models in Amazon Bedrock provider (#727 by @unexge)
Keyboard shortcuts (Ctrl+C, Ctrl+D, etc.) now work on non-Latin keyboard layouts (Russian, Ukrainian, Bulgarian, etc.) in terminals supporting Kitty keyboard protocol with alternate key reporting (#718 by @dannote)

Added

Edit tool now uses fuzzy matching as fallback when exact match fails, tolerating trailing whitespace, smart quotes, Unicode dashes, and special spaces (#713 by @dannote)
Support APPEND_SYSTEM.md to append instructions to the system prompt (#716 by @tallshort)
Session picker search: Ctrl+R toggles sorting between fuzzy match (default) and most recent; supports quoted phrase matching and re: regex mode (#731 by @ogulcancelik)
Export getAgentDir for extensions (#749 by @dannote)
Show loaded prompt templates on startup (#743 by @tallshort)
MiniMax China (minimax-cn) provider support (#725 by @tallshort)
gpt-5.2-codex models for GitHub Copilot and OpenCode Zen providers (#734 by @aadishv)

Changed

Replaced wasm-vips with @silvia-odwyer/photon-node for image processing (#710 by @can1357)
Extension example: plan-mode/ shortcut changed from Shift+P to Ctrl+Alt+P to avoid conflict with typing capital P (#746 by @ferologics)
UI keybinding hints now respect configured keybindings across components (#724 by @dannote)
CLI process title is now set to pi for easier process identification (#742 by @richardgill)

[0.45.7] - 2026-01-13

Added

Exported highlightCode and getLanguageFromPath for extensions (#703 by @dannote)

[0.45.6] - 2026-01-13

Added

ctx.ui.custom() now accepts overlayOptions for overlay positioning and sizing (anchor, margins, offsets, percentages, absolute positioning) (#667 by @nicobailon)
ctx.ui.custom() now accepts onHandle callback to receive the OverlayHandle for controlling overlay visibility (#667 by @nicobailon)
Extension example: overlay-qa-tests.ts with 10 commands for testing overlay positioning, animation, and toggle scenarios (#667 by @nicobailon)
Extension example: doom-overlay/ - DOOM game running as an overlay at 35 FPS (auto-downloads WAD on first run) (#667 by @nicobailon)

[0.45.5] - 2026-01-13

Fixed

Skip changelog display on fresh install (only show on upgrades)

[0.45.4] - 2026-01-13

Changed

Light theme colors adjusted for WCAG AA compliance (4.5:1 contrast ratio against white backgrounds)
Replaced sharp with wasm-vips for image processing (resize, PNG conversion). Eliminates native build requirements that caused installation failures on some systems. (#696)

Added

Extension example: summarize.ts for summarizing conversations using custom UI and an external model (#684 by @scutifer)
Extension example: question.ts enhanced with custom UI for asking user questions (#693 by @ferologics)
Extension example: plan-mode/ enhanced with explicit step tracking and progress widget (#694 by @ferologics)
Extension example: questionnaire.ts for multi-question input with tab bar navigation (#695 by @ferologics)
Experimental Vercel AI Gateway provider support: set AI_GATEWAY_API_KEY and use --provider vercel-ai-gateway. Token usage is currently reported incorrectly by Anthropic Messages compatible endpoint. (#689 by @timolins)

Fixed

Fix API key resolution after model switches by using provider argument (#691 by @joshp123)
Fixed z.ai thinking/reasoning: thinking toggle now correctly enables/disables thinking for z.ai models (#688)
Fixed extension loading in compiled Bun binary: extensions with local file imports now work correctly. Updated @mariozechner/jiti to v2.6.5 which bundles babel for Bun binary compatibility. (#681)
Fixed theme loading when installed via mise: use wrapper directory in release tarballs for compatibility with mise's strip_components=1 extraction. (#681)

[0.45.2] - 2026-01-13

Fixed

Extensions now load correctly in compiled Bun binary by using jiti for module resolution with proper alias handling

[0.45.1] - 2026-01-13

Changed

/share now outputs buildwithpi.ai session preview URLs instead of shittycodingagent.ai

[0.45.0] - 2026-01-13

Added

MiniMax provider support: set MINIMAX_API_KEY and use minimax/MiniMax-M2.1 (#656 by @dannote)
/scoped-models: Alt+Up/Down to reorder enabled models. Order is preserved when saving with Ctrl+S and determines Ctrl+P cycling order. (#676 by @thomasmhr)
Amazon Bedrock provider support (experimental, tested with Anthropic Claude models only) (#494 by @unexge)
Extension example: sandbox/ for OS-level bash sandboxing using @anthropic-ai/sandbox-runtime with per-project config (#673 by @dannote)

[0.44.0] - 2026-01-12

Breaking Changes

pi.getAllTools() now returns ToolInfo[] (with name and description) instead of string[]. Extensions that only need names can use .map(t => t.name). (#648 by @carsonfarmer)

Added

Session naming: /name <name> command sets a display name shown in the session selector instead of the first message. Useful for distinguishing forked sessions. Extensions can use pi.setSessionName() and pi.getSessionName(). (#650 by @scutifer)
Extension example: notify.ts for desktop notifications via OSC 777 escape sequence (#658 by @ferologics)
Inline hint for queued messages showing the Alt+Up restore shortcut (#657 by @tmustier)
Page-up/down navigation in /resume session selector to jump by 5 items (#662 by @aliou)
Fuzzy search in /settings menu: type to filter settings by label (#643 by @ninlds)

Fixed

Session selector now stays open when current folder has no sessions, allowing Tab to switch to "all" scope (#661 by @aliou)
Extensions using theme utilities like getSettingsListTheme() now work in dev mode with tsx

[0.43.0] - 2026-01-11

Breaking Changes

Extension editor (ctx.ui.editor()) now uses Enter to submit and Shift+Enter for newlines, matching the main editor. Previously used Ctrl+Enter to submit. Extensions with hardcoded "ctrl+enter" hints need updating. (#642 by @mitsuhiko)
Renamed /branch command to /fork (#641)
- RPC: branch → fork, get_branch_messages → get_fork_messages
- SDK: branch() → fork(), getBranchMessages() → getForkMessages()
- AgentSession: branch() → fork(), getUserMessagesForBranching() → getUserMessagesForForking()
- Extension events: session_before_branch → session_before_fork, session_branch → session_fork
- Settings: doubleEscapeAction: "branch" | "tree" → "fork" | "tree"
SessionManager.list() and SessionManager.listAll() are now async, returning Promise<SessionInfo[]>. Callers must await them. (#620 by @tmustier)

Added

/resume selector now toggles between current-folder and all sessions with Tab, showing the session cwd in the All view and loading progress. (#620 by @tmustier)
SessionManager.list() and SessionManager.listAll() accept optional onProgress callback for progress updates
SessionInfo.cwd field containing the session's working directory (empty string for old sessions)
SessionListProgress type export for progress callbacks
/scoped-models command to enable/disable models for Ctrl+P cycling. Changes are session-only by default; press Ctrl+S to persist to settings.json. (#626 by @CarlosGtrz)
model_select extension hook fires when model changes via /model, model cycling, or session restore with source field and previousModel (#628 by @marckrenn)
ctx.ui.setWorkingMessage() extension API to customize the "Working..." message during streaming (#625 by @nicobailon)
Skill slash commands: loaded skills are registered as /skill:name commands for quick access. Toggle via /settings or skills.enableSkillCommands in settings.json. (#630 by @Dwsy)
Slash command autocomplete now uses fuzzy matching (type /skbra to match /skill:brave-search)
/tree branch summarization now offers three options: "No summary", "Summarize", and "Summarize with custom prompt". Custom prompts are appended as additional focus to the default summarization instructions. (#642 by @mitsuhiko)

Fixed

Session picker respects custom keybindings when using --resume (#633 by @aos)
Custom footer extensions now see model changes: ctx.model is now a getter that returns the current model instead of a snapshot from when the context was created (#634 by @ogulcancelik)
Footer git branch not updating after external branch switches. Git uses atomic writes (temp file + rename), which changes the inode and breaks fs.watch on the file. Now watches the directory instead.
Extension loading errors are now displayed to the user instead of being silently ignored (#639 by @aliou)

[0.42.5] - 2026-01-11

Fixed

Reduced flicker by only re-rendering changed lines (#617 by @ogulcancelik). No worries tho, there's still a little flicker in the VS Code Terminal. Praise the flicker.
Cursor position tracking when content shrinks with unchanged remaining lines
TUI renders with wrong dimensions after suspend/resume if terminal was resized while suspended (#599)
Pasted content containing Kitty key release patterns (e.g., :3F in MAC addresses) was incorrectly filtered out (#623 by @ogulcancelik)

[0.42.4] - 2026-01-10

Fixed

Bash output expanded hint now says "(ctrl+o to collapse)" (#610 by @tallshort)
Fixed UTF-8 text corruption in remote bash execution (SSH, containers) by using streaming TextDecoder (#608)

[0.42.3] - 2026-01-10

Changed

OpenAI Codex: updated to use bundled system prompt from upstream

[0.42.2] - 2026-01-10

Added

/model <search> now pre-filters the model selector or auto-selects on exact match. Use provider/model syntax to disambiguate (e.g., /model openai/gpt-4). (#587 by @zedrdave)
FooterDataProvider for custom footers: ctx.ui.setFooter() now receives a third footerData parameter providing getGitBranch(), getExtensionStatuses(), and onBranchChange() for reactive updates (#600 by @nicobailon)
Alt+Up hotkey to restore queued steering/follow-up messages back into the editor without aborting the current run (#604 by @tmustier)

Fixed

Fixed LM Studio compatibility for OpenAI Responses tool strict mapping in the ai provider (#598 by @gnattu)

[0.42.1] - 2026-01-09

Fixed

Symlinked directories in prompts/ folders are now followed when loading prompt templates (#601 by @aliou)

[0.42.0] - 2026-01-09

Added

Added OpenCode Zen provider support. Set OPENCODE_API_KEY env var and use opencode/<model-id> (e.g., opencode/claude-opus-4-5).

[0.41.0] - 2026-01-09

Added

Anthropic OAuth support is back! Use /login to authenticate with your Claude Pro/Max subscription.

[0.40.1] - 2026-01-09

Removed

Anthropic OAuth support (/login). Use API keys instead.

[0.40.0] - 2026-01-08

Added

Documentation on component invalidation and theme changes in docs/tui.md

Fixed

Components now properly rebuild their content on theme change (tool executions, assistant messages, bash executions, custom messages, branch/compaction summaries)

[0.39.1] - 2026-01-08

Fixed

setTheme() now triggers a full rerender so previously rendered components update with the new theme colors
mac-system-theme.ts example now polls every 2 seconds and uses osascript for real-time macOS appearance detection

[0.39.0] - 2026-01-08

Breaking Changes

before_agent_start event now receives systemPrompt in the event object and returns systemPrompt (full replacement) instead of systemPromptAppend. Extensions that were appending must now use event.systemPrompt + extra pattern. (#575)
discoverSkills() now returns { skills: Skill[], warnings: SkillWarning[] } instead of Skill[]. This allows callers to handle skill loading warnings. (#577 by @cv)

Added

ctx.ui.getAllThemes(), ctx.ui.getTheme(name), and ctx.ui.setTheme(name | Theme) methods for extensions to list, load, and switch themes at runtime (#576)
--no-tools flag to disable all built-in tools, allowing extension-only tool setups (#557 by @cv)
Pluggable operations for built-in tools enabling remote execution via SSH or other transports (#564). Interfaces: ReadOperations, WriteOperations, EditOperations, BashOperations, LsOperations, GrepOperations, FindOperations
user_bash event for intercepting user !/!! commands, allowing extensions to redirect to remote systems (#528)
setActiveTools() in ExtensionAPI for dynamic tool management
Built-in renderers used automatically for tool overrides without custom renderCall/renderResult
ssh.ts example: remote tool execution via --ssh user@host:/path
interactive-shell.ts example: run interactive commands (vim, git rebase, htop) with full terminal access via !i prefix or auto-detection
Wayland clipboard support for /copy command using wl-copy with xclip/xsel fallback (#570 by @OgulcanCelik)
Experimental: ctx.ui.custom() now accepts { overlay: true } option for floating modal components that composite over existing content without clearing the screen (#558 by @nicobailon)
AgentSession.skills and AgentSession.skillWarnings properties to access loaded skills without rediscovery (#577 by @cv)

Fixed

String systemPrompt in createAgentSession() now works as a full replacement instead of having context files and skills appended, matching documented behavior (#543)
Update notification for bun binary installs now shows release download URL instead of npm command (#567 by @ferologics)
ESC key now works during "Working..." state after auto-retry (#568 by @tmustier)
Abort messages now show correct retry attempt count (e.g., "Aborted after 2 retry attempts") (#568 by @tmustier)
Fixed Antigravity provider returning 429 errors despite available quota (#571 by @ben-vargas)
Fixed malformed thinking text in Gemini/Antigravity responses where thinking content appeared as regular text or vice versa. Cross-model conversations now properly convert thinking blocks to plain text. (#561)
--no-skills flag now correctly prevents skills from loading in interactive mode (#577 by @cv)

[0.38.0] - 2026-01-08

Breaking Changes

ctx.ui.custom() factory signature changed from (tui, theme, done) to (tui, theme, keybindings, done) for keybinding access in custom components
LoadedExtension type renamed to Extension
LoadExtensionsResult.setUIContext() removed, replaced with runtime: ExtensionRuntime
ExtensionRunner constructor now requires runtime: ExtensionRuntime as second parameter
ExtensionRunner.initialize() signature changed from options object to positional params (actions, contextActions, commandContextActions?, uiContext?)
ExtensionRunner.getHasUI() renamed to hasUI()
OpenAI Codex model aliases removed (gpt-5, gpt-5-mini, gpt-5-nano, codex-mini-latest). Use canonical IDs: gpt-5.1, gpt-5.1-codex-mini, gpt-5.2, gpt-5.2-codex. (#536 by @ghoulr)

Added

--no-extensions flag to disable extension discovery while still allowing explicit -e paths (#524 by @cv)
SDK: InteractiveMode, runPrintMode(), runRpcMode() exported for building custom run modes. See docs/sdk.md.
PI_SKIP_VERSION_CHECK environment variable to disable new version notifications at startup (#549 by @aos)
thinkingBudgets setting to customize token budgets per thinking level for token-based providers (#529 by @melihmucuk)
Extension UI dialogs (ctx.ui.select(), ctx.ui.confirm(), ctx.ui.input()) now support a timeout option with live countdown display (#522 by @nicobailon)
Extensions can now provide custom editor components via ctx.ui.setEditorComponent(). See examples/extensions/modal-editor.ts and docs/tui.md Pattern 7.
Extension factories can now be async, enabling dynamic imports and lazy-loaded dependencies (#513 by @austinm911)
ctx.shutdown() is now available in extension contexts for requesting a graceful shutdown. In interactive mode, shutdown is deferred until the agent becomes idle (after processing all queued steering and follow-up messages). In RPC mode, shutdown is deferred until after completing the current command response. In print mode, shutdown is a no-op as the process exits automatically when prompts complete. (#542 by @kaofelix)

Fixed

Default thinking level from settings now applies correctly when enabledModels is configured (#540 by @ferologics)
External edits to settings.json while pi is running are now preserved when pi saves settings (#527 by @ferologics)
Overflow-based compaction now skips if error came from a different model or was already handled by a previous compaction (#535 by @mitsuhiko)
OpenAI Codex context window reduced from 400k to 272k tokens to match Codex CLI defaults and prevent 400 errors (#536 by @ghoulr)
Context overflow detection now recognizes context_length_exceeded errors.
Key presses no longer dropped when input is batched over SSH (#538)
Clipboard image support now works on Alpine Linux and other musl-based distros (#533)

[0.37.6] - 2026-01-06

Added

Extension UI dialogs (ctx.ui.select(), ctx.ui.confirm(), ctx.ui.input()) now accept an optional AbortSignal to programmatically dismiss dialogs. Useful for implementing timeouts. See examples/extensions/timed-confirm.ts. (#474)
HTML export now shows bridge prompts in model change messages for Codex sessions (#510 by @mitsuhiko)

[0.37.5] - 2026-01-06

Added

ExtensionAPI: setModel(), getThinkingLevel(), setThinkingLevel() methods for extensions to change model and thinking level at runtime (#509)
Exported truncation utilities for custom tools: truncateHead, truncateTail, truncateLine, formatSize, DEFAULT_MAX_BYTES, DEFAULT_MAX_LINES, TruncationOptions, TruncationResult
New example truncated-tool.ts demonstrating proper output truncation with custom rendering for extensions
New example preset.ts demonstrating preset configurations with model/thinking/tools switching (#347)
Documentation for output truncation best practices in docs/extensions.md
Exported all UI components for extensions: ArminComponent, AssistantMessageComponent, BashExecutionComponent, BorderedLoader, BranchSummaryMessageComponent, CompactionSummaryMessageComponent, CustomEditor, CustomMessageComponent, DynamicBorder, ExtensionEditorComponent, ExtensionInputComponent, ExtensionSelectorComponent, FooterComponent, LoginDialogComponent, ModelSelectorComponent, OAuthSelectorComponent, SessionSelectorComponent, SettingsSelectorComponent, ShowImagesSelectorComponent, ThemeSelectorComponent, ThinkingSelectorComponent, ToolExecutionComponent, TreeSelectorComponent, UserMessageComponent, UserMessageSelectorComponent, plus utilities renderDiff, truncateToVisualLines
docs/tui.md: Common Patterns section with copy-paste code for SelectList, BorderedLoader, SettingsList, setStatus, setWidget, setFooter
docs/tui.md: Key Rules section documenting critical patterns for extension UI development
docs/extensions.md: Exhaustive example links for all ExtensionAPI methods and events
System prompt now references docs/tui.md for TUI component development

[0.37.4] - 2026-01-06

Added

Session picker (pi -r) and --session flag now support searching/resuming by session ID (UUID prefix) (#495 by @arunsathiya)
Extensions can now replace the startup header with ctx.ui.setHeader(), see examples/extensions/custom-header.ts (#500 by @tudoroancea)

Changed

Startup help text: fixed misleading "ctrl+k to delete line" to "ctrl+k to delete to end"
Startup help text and /hotkeys: added !! shortcut for running bash without adding output to context

Fixed

Queued steering/follow-up messages no longer wipe unsent editor input (#503 by @tmustier)
OAuth token refresh failure no longer crashes app at startup, allowing user to /login to re-authenticate (#498)

[0.37.3] - 2026-01-06

Added

Extensions can now replace the footer with ctx.ui.setFooter(), see examples/extensions/custom-footer.ts (#481)
Session ID is now forwarded to LLM providers for session-based caching (used by OpenAI Codex for prompt caching).
Added blockImages setting to prevent images from being sent to LLM providers (#492 by @jsinge97)
Extensions can now send user messages via pi.sendUserMessage() (#483)

Fixed

Add minimatch as a direct dependency for explicit imports.
Status bar now shows correct git branch when running in a git worktree (#490 by @kcosr)
Interactive mode: Ctrl+V clipboard image paste now works on Wayland sessions by using wl-paste with xclip fallback (#488 by @ghoulr)

[0.37.2] - 2026-01-05

Fixed

Extension directories in settings.json now respect package.json manifests, matching global extension behavior (#480 by @prateekmedia)
Share viewer: deep links now scroll to the target message when opened via /share
Bash tool now handles spawn errors gracefully instead of crashing the agent (missing cwd, invalid shell path) (#479 by @robinwander)

[0.37.1] - 2026-01-05

Fixed

Share viewer: copy-link buttons now generate correct URLs when session is viewed via /share (iframe context)

[0.37.0] - 2026-01-05

Added

Share viewer: copy-link button on messages to share URLs that navigate directly to a specific message (#477 by @lockmeister)
Extension example: add claude-rules to load .claude/rules/ entries into the system prompt (#461 by @vaayne)
Headless OAuth login: all providers now show paste input for manual URL/code entry, works over SSH without DISPLAY (#428 by @ben-vargas, #468 by @crcatala)

Changed

OAuth login UI now uses dedicated dialog component with consistent borders
Assume truecolor support for all terminals except dumb, empty, or linux (fixes colors over SSH)
OpenAI Codex clean-up: removed per-thinking-level model variants, thinking level is now set separately and the provider clamps to what each model supports internally (initial implementation in #472 by @ben-vargas)

Fixed

Messages submitted during compaction are queued and delivered after compaction completes, preserving steering and follow-up behavior. Extension commands execute immediately during compaction. (#476 by @tmustier)
Managed binaries (fd, rg) now stored in ~/.pi/agent/bin/ instead of tools/, eliminating false deprecation warnings (#470 by @mcinteerj)
Extensions defined in settings.json were not loaded (#463 by @melihmucuk)
OAuth refresh no longer logs users out when multiple pi instances are running (#466 by @Cursivez)
Migration warnings now ignore fd.exe and rg.exe in tools/ on Windows (#458 by @carlosgtrz)
CI: add examples/extensions/with-deps to workspaces to fix typecheck (#467 by @aliou)
SDK: passing extensions: [] now disables extension discovery as documented (#465 by @aliou)

[0.36.0] - 2026-01-05

Added

Experimental: OpenAI Codex OAuth provider support: access Codex models via ChatGPT Plus/Pro subscription using /login openai-codex (#451 by @kim0)

[0.35.0] - 2026-01-05

This release unifies hooks and custom tools into a single "extensions" system and renames "slash commands" to "prompt templates". (#454)

Before migrating, read:

docs/extensions.md - Full API reference
README.md - Extensions section with examples
examples/extensions/ - Working examples

Extensions Migration

Hooks and custom tools are now unified as extensions. Both were TypeScript modules exporting a factory function that receives an API object. Now there's one concept, one discovery location, one CLI flag, one settings.json entry.

Automatic migration:

commands/ directories are automatically renamed to prompts/ on startup (both ~/.pi/agent/commands/ and .pi/commands/)

Manual migration required:

Move files from hooks/ and tools/ directories to extensions/ (deprecation warnings shown on startup)
Update imports and type names in your extension code
Update settings.json if you have explicit hook and custom tool paths configured

Directory changes:

# Before
~/.pi/agent/hooks/*.ts       →  ~/.pi/agent/extensions/*.ts
~/.pi/agent/tools/*.ts       →  ~/.pi/agent/extensions/*.ts
.pi/hooks/*.ts               →  .pi/extensions/*.ts
.pi/tools/*.ts               →  .pi/extensions/*.ts

Extension discovery rules (in extensions/ directories):

Direct files: extensions/*.ts or *.js → loaded directly
Subdirectory with index: extensions/myext/index.ts → loaded as single extension
Subdirectory with package.json: extensions/myext/package.json with "pi" field → loads declared paths

// extensions/my-package/package.json
{
  "name": "my-extension-package",
  "dependencies": { "zod": "^3.0.0" },
  "pi": {
    "extensions": ["./src/main.ts", "./src/tools.ts"]
  }
}

No recursion beyond one level. Complex packages must use the package.json manifest. Dependencies are resolved via jiti, and extensions can be published to and installed from npm.

Type renames:

HookAPI → ExtensionAPI
HookContext → ExtensionContext
HookCommandContext → ExtensionCommandContext
HookUIContext → ExtensionUIContext
CustomToolAPI → ExtensionAPI (merged)
CustomToolContext → ExtensionContext (merged)
CustomToolUIContext → ExtensionUIContext
CustomTool → ToolDefinition
CustomToolFactory → ExtensionFactory
HookMessage → CustomMessage

Import changes:

// Before (hook)
import type { HookAPI, HookContext } from "@mariozechner/pi-coding-agent";
export default function (pi: HookAPI) { ... }

// Before (custom tool)
import type { CustomToolFactory } from "@mariozechner/pi-coding-agent";
const factory: CustomToolFactory = (pi) => ({ name: "my_tool", ... });
export default factory;

// After (both are now extensions)
import type { ExtensionAPI } from "@mariozechner/pi-coding-agent";
export default function (pi: ExtensionAPI) {
  pi.on("tool_call", async (event, ctx) => { ... });
  pi.registerTool({ name: "my_tool", ... });
}

Custom tools now have full context access. Tools registered via pi.registerTool() now receive the same ctx object that event handlers receive. Previously, custom tools had limited context. Now all extension code shares the same capabilities:

pi.registerTool() - Register tools the LLM can call
pi.registerCommand() - Register commands like /mycommand
pi.registerShortcut() - Register keyboard shortcuts (shown in /hotkeys)
pi.registerFlag() - Register CLI flags (shown in --help)
pi.registerMessageRenderer() - Custom TUI rendering for message types
pi.on() - Subscribe to lifecycle events (tool_call, session_start, etc.)
pi.sendMessage() - Inject messages into the conversation
pi.appendEntry() - Persist custom data in session (survives restart/branch)
pi.exec() - Run shell commands
pi.getActiveTools() / pi.setActiveTools() - Dynamic tool enable/disable
pi.getAllTools() - List all available tools
pi.events - Event bus for cross-extension communication
ctx.ui.confirm() / select() / input() - User prompts
ctx.ui.notify() - Toast notifications
ctx.ui.setStatus() - Persistent status in footer (multiple extensions can set their own)
ctx.ui.setWidget() - Widget display above editor
ctx.ui.setTitle() - Set terminal window title
ctx.ui.custom() - Full TUI component with keyboard handling
ctx.ui.editor() - Multi-line text editor with external editor support
ctx.sessionManager - Read session entries, get branch history

Settings changes:

// Before
{
  "hooks": ["./my-hook.ts"],
  "customTools": ["./my-tool.ts"]
}

// After
{
  "extensions": ["./my-extension.ts"]
}

CLI changes:

# Before
pi --hook ./safety.ts --tool ./todo.ts

# After
pi --extension ./safety.ts -e ./todo.ts

Prompt Templates Migration

"Slash commands" (markdown files defining reusable prompts invoked via /name) are renamed to "prompt templates" to avoid confusion with extension-registered commands.

Automatic migration: The commands/ directory is automatically renamed to prompts/ on startup (if prompts/ doesn't exist). Works for both regular directories and symlinks.

Directory changes:

~/.pi/agent/commands/*.md    →  ~/.pi/agent/prompts/*.md
.pi/commands/*.md            →  .pi/prompts/*.md

SDK type renames:

FileSlashCommand → PromptTemplate
LoadSlashCommandsOptions → LoadPromptTemplatesOptions

SDK function renames:

discoverSlashCommands() → discoverPromptTemplates()
loadSlashCommands() → loadPromptTemplates()
expandSlashCommand() → expandPromptTemplate()
getCommandsDir() → getPromptsDir()

SDK option renames:

CreateAgentSessionOptions.slashCommands → .promptTemplates
AgentSession.fileCommands → .promptTemplates
PromptOptions.expandSlashCommands → .expandPromptTemplates

SDK Migration

Discovery functions:

discoverAndLoadHooks() → discoverAndLoadExtensions()
discoverAndLoadCustomTools() → merged into discoverAndLoadExtensions()
loadHooks() → loadExtensions()
loadCustomTools() → merged into loadExtensions()

Runner and wrapper:

HookRunner → ExtensionRunner
wrapToolsWithHooks() → wrapToolsWithExtensions()
wrapToolWithHooks() → wrapToolWithExtensions()

CreateAgentSessionOptions:

.hooks → removed (use .additionalExtensionPaths for paths)
.additionalHookPaths → .additionalExtensionPaths
.preloadedHooks → .preloadedExtensions
.customTools type changed: Array<{ path?; tool: CustomTool }> → ToolDefinition[]
.additionalCustomToolPaths → merged into .additionalExtensionPaths
.slashCommands → .promptTemplates

AgentSession:

.hookRunner → .extensionRunner
.fileCommands → .promptTemplates
.sendHookMessage() → .sendCustomMessage()

Session Migration

Automatic. Session version bumped from 2 to 3. Existing sessions are migrated on first load:

Message role "hookMessage" → "custom"

[0.34.1] - 2026-01-04

Changed

Expanded keybinding documentation to list all 32 supported symbol keys with notes on ctrl+symbol behavior (#450 by @kaofelix)

[0.34.0] - 2026-01-04

Added

Hook API: before_agent_start handlers can now return systemPromptAppend to dynamically append text to the system prompt for that turn. Multiple hooks' appends are concatenated.
Hook API: before_agent_start handlers can now return multiple messages (all are injected, not just the first)
New example hook: tools.ts - Interactive /tools command to enable/disable tools with session persistence
New example hook: pirate.ts - Demonstrates systemPromptAppend to make the agent speak like a pirate
Tool registry now contains all built-in tools (read, bash, edit, write, grep, find, ls) even when --tools limits the initially active set. Hooks can enable any tool from the registry via pi.setActiveTools().
System prompt now automatically rebuilds when tools change via setActiveTools(), updating tool descriptions and guidelines to match the new tool set
Hook errors now display full stack traces for easier debugging

Changed

Removed image placeholders after copy & paste, replaced with inserting image file paths directly. (#442 by @mitsuhiko)

Fixed

Fixed potential text decoding issues in bash executor by using streaming TextDecoder instead of Buffer.toString()
External editor (Ctrl-G) now shows full pasted content instead of [paste #N ...] placeholders (#444 by @aliou)

[0.33.0] - 2026-01-04

Breaking Changes

Key detection functions removed from @mariozechner/pi-tui: All isXxx() key detection functions (isEnter(), isEscape(), isCtrlC(), etc.) have been removed. Use matchesKey(data, keyId) instead (e.g., matchesKey(data, "enter"), matchesKey(data, "ctrl+c")). This affects hooks and custom tools that use ctx.ui.custom() with keyboard input handling. (#405)

Added

Clipboard image paste support via Ctrl+V. Images are saved to a temp file and attached to the message. Works on macOS, Windows, and Linux (X11). (#419)
Configurable keybindings via ~/.pi/agent/keybindings.json. All keyboard shortcuts (editor navigation, deletion, app actions like model cycling, etc.) can now be customized. Supports multiple bindings per action. (#405 by @hjanuschka)
/quit and /exit slash commands to gracefully exit the application. Unlike double Ctrl+C, these properly await hook and custom tool cleanup handlers before exiting. (#426 by @ben-vargas)

Fixed

Subagent example README referenced incorrect filename subagent.ts instead of index.ts (#427 by @Whamp)

[0.32.3] - 2026-01-03

Fixed

--list-models no longer shows Google Vertex AI models without explicit authentication configured
JPEG/GIF/WebP images not displaying in terminals using Kitty graphics protocol (Kitty, Ghostty, WezTerm). The protocol requires PNG format, so non-PNG images are now converted before display.
Version check URL typo preventing update notifications from working (#423 by @skuridin)
Large images exceeding Anthropic's 5MB limit now retry with progressive quality/size reduction (#424 by @mitsuhiko)

[0.32.2] - 2026-01-03

Added

$ARGUMENTS syntax for custom slash commands as alternative to $@ for all arguments joined. Aligns with patterns used by Claude, Codex, and OpenCode. Both syntaxes remain fully supported. (#418 by @skuridin)

Changed

Slash commands and hook commands now work during streaming: Previously, using a slash command or hook command while the agent was streaming would crash with "Agent is already processing". Now:
- Hook commands execute immediately (they manage their own LLM interaction via pi.sendMessage())
- File-based slash commands are expanded and queued via steer/followUp
- steer() and followUp() now expand file-based slash commands and error on hook commands (hook commands cannot be queued)
- prompt() accepts new streamingBehavior option ("steer" or "followUp") to specify queueing behavior during streaming
- RPC prompt command now accepts optional streamingBehavior field (#420)

Fixed

Slash command argument substitution now processes positional arguments ($1, $2, etc.) before all-arguments ($@, $ARGUMENTS) to prevent recursive substitution when argument values contain dollar-digit patterns like $100. (#418 by @skuridin)

[0.32.1] - 2026-01-03

Added

Shell commands without context contribution: use !!command to execute a bash command that is shown in the TUI and saved to session history but excluded from LLM context. Useful for running commands you don't want the AI to see. (#414)

[0.32.0] - 2026-01-03

Breaking Changes

Queue API replaced with steer/followUp: The queueMessage() method has been split into two methods with different delivery semantics (#403):
- steer(text): Interrupts the agent mid-run (Enter while streaming). Delivered after current tool execution.
- followUp(text): Waits until the agent finishes (Alt+Enter while streaming). Delivered only when agent stops.
Settings renamed: queueMode setting renamed to steeringMode. Added new followUpMode setting. Old settings.json files are migrated automatically.
AgentSession methods renamed:
- queueMessage() → steer() and followUp()
- queueMode getter → steeringMode and followUpMode getters
- setQueueMode() → setSteeringMode() and setFollowUpMode()
- queuedMessageCount → pendingMessageCount
- getQueuedMessages() → getSteeringMessages() and getFollowUpMessages()
- clearQueue() now returns { steering: string[], followUp: string[] }
- hasQueuedMessages() → hasPendingMessages()
Hook API signature changed: pi.sendMessage() second parameter changed from triggerTurn?: boolean to options?: { triggerTurn?, deliverAs? }. Use deliverAs: "followUp" for follow-up delivery. Affects both hooks and internal sendHookMessage() method.
RPC API changes:
- queue_message command → steer and follow_up commands
- set_queue_mode command → set_steering_mode and set_follow_up_mode commands
- RpcSessionState.queueMode → steeringMode and followUpMode
Settings UI: "Queue mode" setting split into "Steering mode" and "Follow-up mode"

[0.31.1] - 2026-01-02

Fixed

Model selector no longer allows negative index when pressing arrow keys before models finish loading (#398 by @mitsuhiko)
Type guard functions (isBashToolResult, etc.) now exported at runtime, not just in type declarations (#397)

[0.31.0] - 2026-01-02

This release introduces session trees for in-place branching, major API changes to hooks and custom tools, and structured compaction with file tracking.

Session Tree

Sessions now use a tree structure with id/parentId fields. This enables in-place branching: navigate to any previous point with /tree, continue from there, and switch between branches while preserving all history in a single file.

Existing sessions are automatically migrated (v1 → v2) on first load. No manual action required.

New entry types: BranchSummaryEntry (context from abandoned branches), CustomEntry (hook state), CustomMessageEntry (hook-injected messages), LabelEntry (bookmarks).

See docs/session.md for the file format and SessionManager API.

Hooks Migration

The hooks API has been restructured with more granular events and better session access.

Type renames:

HookEventContext → HookContext
HookCommandContext is now a new interface extending HookContext with session control methods

Event changes:

The monolithic session event is now split into granular events: session_start, session_before_switch, session_switch, session_before_branch, session_branch, session_before_compact, session_compact, session_shutdown
session_before_switch and session_switch events now include reason: "new" | "resume" to distinguish between /new and /resume
New session_before_tree and session_tree events for /tree navigation (hook can provide custom branch summary)
New before_agent_start event: inject messages before the agent loop starts
New context event: modify messages non-destructively before each LLM call
Session entries are no longer passed in events. Use ctx.sessionManager.getEntries() or ctx.sessionManager.getBranch() instead

API changes:

pi.send(text, attachments?) → pi.sendMessage(message, triggerTurn?) (creates CustomMessageEntry)
New pi.appendEntry(customType, data?) for hook state persistence (not in LLM context)
New pi.registerCommand(name, options) for custom slash commands (handler receives HookCommandContext)
New pi.registerMessageRenderer(customType, renderer) for custom TUI rendering
New ctx.isIdle(), ctx.abort(), ctx.hasQueuedMessages() for agent state (available in all events)
New ctx.ui.editor(title, prefill?) for multi-line text editing with Ctrl+G external editor support
New ctx.ui.custom(component) for full TUI component rendering with keyboard focus
New ctx.ui.setStatus(key, text) for persistent status text in footer (multiple hooks can set their own)
New ctx.ui.theme getter for styling text with theme colors
ctx.exec() moved to pi.exec()
ctx.sessionFile → ctx.sessionManager.getSessionFile()
New ctx.modelRegistry and ctx.model for API key resolution

HookCommandContext (slash commands only):

ctx.waitForIdle() - wait for agent to finish streaming
ctx.newSession(options?) - create new sessions with optional setup callback
ctx.branch(entryId) - branch from a specific entry
ctx.navigateTree(targetId, options?) - navigate the session tree

These methods are only on HookCommandContext (not HookContext) because they can deadlock if called from event handlers that run inside the agent loop.

Removed:

hookTimeout setting (hooks no longer have timeouts; use Ctrl+C to abort)
resolveApiKey parameter (use ctx.modelRegistry.getApiKey(model))

See docs/hooks.md and examples/hooks/ for the current API.

Custom Tools Migration

The custom tools API has been restructured to mirror the hooks pattern with a context object.

Type renames:

CustomAgentTool → CustomTool
ToolAPI → CustomToolAPI
ToolContext → CustomToolContext
ToolSessionEvent → CustomToolSessionEvent

Execute signature changed:

// Before (v0.30.2)
execute(toolCallId, params, signal, onUpdate)

// After
execute(toolCallId, params, onUpdate, ctx, signal?)

The new ctx: CustomToolContext provides sessionManager, modelRegistry, model, and agent state methods:

ctx.isIdle() - check if agent is streaming
ctx.hasQueuedMessages() - check if user has queued messages (skip interactive prompts)
ctx.abort() - abort current operation (fire-and-forget)

Session event changes:

CustomToolSessionEvent now only has reason and previousSessionFile
Session entries are no longer in the event. Use ctx.sessionManager.getBranch() or ctx.sessionManager.getEntries() to reconstruct state
Reasons: "start" | "switch" | "branch" | "tree" | "shutdown" (no separate "new" reason; /new triggers "switch")
dispose() method removed. Use onSession with reason: "shutdown" for cleanup

See docs/custom-tools.md and examples/custom-tools/ for the current API.

SDK Migration

Type changes:

CustomAgentTool → CustomTool
AppMessage → AgentMessage
sessionFile returns string | undefined (was string | null)
model returns Model | undefined (was Model | null)
Attachment type removed. Use ImageContent from @oh-my-pi/pi-ai instead. Add images directly to message content arrays.

AgentSession API:

branch(entryIndex: number) → branch(entryId: string)
getUserMessagesForBranching() returns { entryId, text } instead of { entryIndex, text }
reset() → newSession(options?) where options has optional parentSession for lineage tracking
newSession() and switchSession() now return Promise<boolean> (false if cancelled by hook)
New navigateTree(targetId, options?) for in-place tree navigation

Hook integration:

New sendHookMessage(message, triggerTurn?) for hook message injection

SessionManager API:

Method renames: saveXXX() → appendXXX() (e.g., appendMessage, appendCompaction)
branchInPlace() → branch()
reset() → newSession(options?) with optional parentSession for lineage tracking
createBranchedSessionFromEntries(entries, index) → createBranchedSession(leafId)
SessionHeader.branchedFrom → SessionHeader.parentSession
saveCompaction(entry) → appendCompaction(summary, firstKeptEntryId, tokensBefore, details?)
getEntries() now excludes the session header (use getHeader() separately)
getSessionFile() returns string | undefined (undefined for in-memory sessions)
New tree methods: getTree(), getBranch(), getLeafId(), getLeafEntry(), getEntry(), getChildren(), getLabel()
New append methods: appendCustomEntry(), appendCustomMessageEntry(), appendLabelChange()
New branch methods: branch(entryId), branchWithSummary()

ModelRegistry (new):

ModelRegistry is a new class that manages model discovery and API key resolution. It combines built-in models with custom models from models.json and resolves API keys via AuthStorage.

import { discoverAuthStorage, discoverModels } from "@oh-my-pi/pi-coding-agent";

const authStorage = discoverAuthStorage(); // ~/.pi/agent/auth.json
const modelRegistry = discoverModels(authStorage); // + ~/.pi/agent/models.json

// Get all models (built-in + custom)
const allModels = modelRegistry.getAll();

// Get only models with valid API keys
const available = await modelRegistry.getAvailable();

// Find specific model
const model = modelRegistry.find("anthropic", "claude-sonnet-4-20250514");

// Get API key for a model
const apiKey = await modelRegistry.getApiKey(model);

This replaces the old resolveApiKey callback pattern. Hooks and custom tools access it via ctx.modelRegistry.

Renamed exports:

messageTransformer → convertToLlm
SessionContext alias LoadedSession removed

See docs/sdk.md and examples/sdk/ for the current API.

RPC Migration

Session commands:

reset command → new_session command with optional parentSession field

Branching commands:

branch command: entryIndex → entryId
get_branch_messages response: entryIndex → entryId

Type changes:

Messages are now AgentMessage (was AppMessage)
prompt command: attachments field replaced with images field using ImageContent format

Compaction events:

auto_compaction_start now includes reason field ("threshold" or "overflow")
auto_compaction_end now includes willRetry field
compact response includes full CompactionResult (summary, firstKeptEntryId, tokensBefore, details)

See docs/rpc.md for the current protocol.

Structured Compaction

Compaction and branch summarization now use a structured output format:

Clear sections: Goal, Progress, Key Information, File Operations
File tracking: readFiles and modifiedFiles arrays in details, accumulated across compactions
Conversations are serialized to text before summarization to prevent the model from "continuing" them

The before_compact and before_tree hook events allow custom compaction implementations. See docs/compaction.md.

Interactive Mode

/tree command:

Navigate the full session tree in-place
Search by typing, page with ←/→
Filter modes (Ctrl+O): default → no-tools → user-only → labeled-only → all
Press l to label entries as bookmarks
Selecting a branch switches context and optionally injects a summary of the abandoned branch

Entry labels:

Bookmark any entry via /tree → select → l
Labels appear in tree view and persist as LabelEntry

Theme changes (breaking for custom themes):

Custom themes must add these new color tokens or they will fail to load:

selectedBg: background for selected/highlighted items in tree selector and other components
customMessageBg: background for hook-injected messages (CustomMessageEntry)
customMessageText: text color for hook messages
customMessageLabel: label color for hook messages (the [customType] prefix)

Total color count increased from 46 to 50. See docs/theme.md for the full color list and copy values from the built-in dark/light themes.

Settings:

enabledModels: allowlist models in settings.json (same format as --models CLI)

Added

ctx.ui.setStatus(key, text) for hooks to display persistent status text in the footer (#385 by @prateekmedia)
ctx.ui.theme getter for styling status text and other output with theme colors
/share command to upload session as a secret GitHub gist and get a shareable URL via shittycodingagent.ai (#380)
HTML export now includes a tree visualization sidebar for navigating session branches (#375)
HTML export supports keyboard shortcuts: Ctrl+T to toggle thinking blocks, Ctrl+O to toggle tool outputs
HTML export supports theme-configurable background colors via optional export section in theme JSON (#387 by @mitsuhiko)
HTML export syntax highlighting now uses theme colors and matches TUI rendering
Snake game example hook: Demonstrates ui.custom(), registerCommand(), and session persistence. See examples/hooks/snake.ts.
thinkingText theme token: Configurable color for thinking block text. (#366 by @paulbettner)

Changed

Entry IDs: Session entries now use short 8-character hex IDs instead of full UUIDs
API key priority: ANTHROPIC_OAUTH_TOKEN now takes precedence over ANTHROPIC_API_KEY
HTML export template split into separate files (template.html, template.css, template.js) for easier maintenance

Fixed

HTML export now properly sanitizes user messages containing HTML tags like <style> that could break DOM rendering
Crash when displaying bash output containing Unicode format characters like U+0600-U+0604 (#372 by @HACKE-RC)
Footer shows full session stats: Token usage and cost now include all messages, not just those after compaction. (#322)
Status messages spam chat log: Rapidly changing settings (e.g., thinking level via Shift+Tab) would add multiple status lines. Sequential status updates now coalesce into a single line. (#365 by @paulbettner)
Toggling thinking blocks during streaming shows nothing: Pressing Ctrl+T while streaming would hide the current message until streaming completed.
Resuming session resets thinking level to off: Initial model and thinking level were not saved to session file, causing --resume/--continue to default to off. (#342 by @aliou)
Hook tool_result event ignores errors from custom tools: The tool_result hook event was never emitted when tools threw errors, and always had isError: false for successful executions. Now emits the event with correct isError value in both success and error cases. (#374 by @nicobailon)
Edit tool fails on Windows due to CRLF line endings: Files with CRLF line endings now match correctly when LLMs send LF-only text. Line endings are normalized before matching and restored to original style on write. (#355 by @Pratham-Dubey)
Edit tool fails on files with UTF-8 BOM: Files with UTF-8 BOM marker could cause "text not found" errors since the LLM doesn't include the invisible BOM character. BOM is now stripped before matching and restored on write. (#394 by @prathamdby)
Use bash instead of sh on Unix: Fixed shell commands using /bin/sh instead of /bin/bash on Unix systems. (#328 by @dnouri)
OAuth login URL clickable: Made OAuth login URLs clickable in terminal. (#349 by @Cursivez)
Improved error messages: Better error messages when apiKey or model are missing. (#346 by @ronyrus)
Session file validation: findMostRecentSession() now validates session headers before returning, preventing non-session JSONL files from being loaded
Compaction error handling: generateSummary() and generateTurnPrefixSummary() now throw on LLM errors instead of returning empty strings
Compaction with branched sessions: Fixed compaction incorrectly including entries from abandoned branches, causing token overflow errors. Compaction now uses sessionManager.getPath() to work only on the current branch path, eliminating 80+ lines of duplicate entry collection logic between prepareCompaction() and compact()
enabledModels glob patterns: --models and enabledModels now support glob patterns like github-copilot/* or *sonnet*. Previously, patterns were only matched literally or via substring search. (#337)

[0.30.2] - 2025-12-26

Changed

Consolidated migrations: Moved auth migration from AuthStorage.migrateLegacy() to new migrations.ts module.

[0.30.1] - 2025-12-26

Fixed

Sessions saved to wrong directory: In v0.30.0, sessions were being saved to ~/.pi/agent/ instead of ~/.pi/agent/sessions/<encoded-cwd>/, breaking --resume and /resume. Misplaced sessions are automatically migrated on startup. (#320 by @aliou)
Custom system prompts missing context: When using a custom system prompt string, project context files (AGENTS.md), skills, date/time, and working directory were not appended. (#321)

[0.30.0] - 2025-12-25

Breaking Changes

SessionManager API: The second parameter of create(), continueRecent(), and list() changed from agentDir to sessionDir. When provided, it specifies the session directory directly (no cwd encoding). When omitted, uses default (~/.pi/agent/sessions/<encoded-cwd>/). open() no longer takes agentDir. (#313)

Added

--session-dir flag: Use a custom directory for sessions instead of the default ~/.pi/agent/sessions/<encoded-cwd>/. Works with -c (continue) and -r (resume) flags. (#313 by @scutifer)
Reverse model cycling and model selector: Shift+Ctrl+P cycles models backward, Ctrl+L opens model selector (retaining text in editor). (#315 by @mitsuhiko)

[0.29.1] - 2025-12-25

Added

Automatic custom system prompt loading: Pi now auto-loads SYSTEM.md files to replace the default system prompt. Project-local .pi/SYSTEM.md takes precedence over global ~/.pi/agent/SYSTEM.md. CLI --system-prompt flag overrides both. (#309)
Unified /settings command: New settings menu consolidating thinking level, theme, queue mode, auto-compact, show images, hide thinking, and collapse changelog. Replaces individual /thinking, /queue, /theme, /autocompact, and /show-images commands. (#310)

Fixed

Custom tools/hooks with typebox subpath imports: Fixed jiti alias for @sinclair/typebox to point to package root instead of entry file, allowing imports like @sinclair/typebox/compiler to resolve correctly. (#311 by @kim0)

[0.29.0] - 2025-12-25

Breaking Changes

Renamed /clear to /new: The command to start a fresh session is now /new. Hook event reasons before_clear/clear are now before_new/new. Merry Christmas @mitsuhiko! (#305)

Added

Auto-space before pasted file paths: When pasting a file path (starting with /, ~, or .) after a word character, a space is automatically prepended. (#307 by @mitsuhiko)
Word navigation in input fields: Added Ctrl+Left/Right and Alt+Left/Right for word-by-word cursor movement. (#306 by @kim0)
Full Unicode input: Input fields now accept Unicode characters beyond ASCII. (#306 by @kim0)

Fixed

Readline-style Ctrl+W: Now skips trailing whitespace before deleting the preceding word, matching standard readline behavior. (#306 by @kim0)

[0.28.0] - 2025-12-25

Changed

Credential storage refactored: API keys and OAuth tokens are now stored in ~/.pi/agent/auth.json instead of oauth.json and settings.json. Existing credentials are automatically migrated on first run. (#296)
SDK API changes (#296):
- Added AuthStorage class for credential management (API keys and OAuth tokens)
- Added ModelRegistry class for model discovery and API key resolution
- Added discoverAuthStorage() and discoverModels() discovery functions
- createAgentSession() now accepts authStorage and modelRegistry options
- Removed configureOAuthStorage(), defaultGetApiKey(), findModel(), discoverAvailableModels()
- Removed getApiKey callback option (use AuthStorage.setRuntimeApiKey() for runtime overrides)
- Use getModel() from @oh-my-pi/pi-ai for built-in models, modelRegistry.find() for custom models + built-in models
- See updated SDK documentation and README
Settings changes: Removed apiKeys from settings.json. Use auth.json instead. (#296)

Fixed

Duplicate skill warnings for symlinks: Skills loaded via symlinks pointing to the same file are now silently deduplicated instead of showing name collision warnings. (#304 by @mitsuhiko)

[0.27.9] - 2025-12-24

Fixed

Model selector and --list-models with settings.json API keys: Models with API keys configured in settings.json (but not in environment variables) now properly appear in the /model selector and --list-models output. (#295)

[0.27.8] - 2025-12-24

Fixed

API key priority: OAuth tokens now take priority over settings.json API keys. Previously, an API key in settings.json would trump OAuth, causing users logged in with a plan (unlimited tokens) to be billed via PAYG instead.

[0.27.7] - 2025-12-24

Fixed

Thinking tag leakage: Fixed Claude mimicking literal </thinking> tags in responses. Unsigned thinking blocks (from aborted streams) are now converted to plain text without <thinking> tags. The TUI still displays them as thinking blocks. (#302 by @nicobailon)

[0.27.6] - 2025-12-24

Added

Compaction hook improvements: The before_compact session event now includes:
- previousSummary: Summary from the last compaction (if any), so hooks can preserve accumulated context
- messagesToKeep: Messages that will be kept after the summary (recent turns), in addition to messagesToSummarize
- resolveApiKey: Function to resolve API keys for any model (checks settings, OAuth, env vars)
- Removed apiKey string in favor of resolveApiKey for more flexibility
SessionManager API cleanup:
- Renamed loadSessionFromEntries() to buildSessionContext() (builds LLM context from entries, handling compaction)
- Renamed loadEntries() to getEntries() (returns defensive copy of all session entries)
- Added buildSessionContext() method to SessionManager

[0.27.5] - 2025-12-24

Added

HTML export syntax highlighting: Code blocks in markdown and tool outputs (read, write) now have syntax highlighting using highlight.js with theme-aware colors matching the TUI.
HTML export improvements: Render markdown server-side using marked (tables, headings, code blocks, etc.), honor user's chosen theme (light/dark), add image rendering for user messages, and style code blocks with TUI-like language markers. (@scutifer)

Fixed

Ghostty inline images in tmux: Fixed terminal detection for Ghostty when running inside tmux by checking GHOSTTY_RESOURCES_DIR env var. (#299 by @nicobailon)

[0.27.4] - 2025-12-24

Fixed

Symlinked skill directories: Skills in symlinked directories (e.g., ~/.pi/agent/skills/my-skills -> /path/to/skills) are now correctly discovered and loaded.

[0.27.3] - 2025-12-24

Added

API keys in settings.json: Store API keys in ~/.pi/agent/settings.json under the apiKeys field (e.g., { "apiKeys": { "anthropic": "sk-..." } }). Settings keys take priority over environment variables. (#295)

Fixed

Allow startup without API keys: Interactive mode no longer throws when no API keys are configured. Users can now start the agent and use /login to authenticate. (#288)
--system-prompt file path support: The --system-prompt argument now correctly resolves file paths (like --append-system-prompt already did). (#287 by @scutifer)

[0.27.2] - 2025-12-23

Added

Skip conversation restore on branch: Hooks can return { skipConversationRestore: true } from before_branch to create the branched session file without restoring conversation messages. Useful for checkpoint hooks that restore files separately. (#286 by @nicobarray)

[0.27.1] - 2025-12-22

Fixed

Skill discovery performance: Skip node_modules directories when recursively scanning for skills. Fixes ~60ms startup delay when skill directories contain npm dependencies.

Added

Startup timing instrumentation: Set PI_TIMING=1 to see startup performance breakdown (interactive mode only).

[0.27.0] - 2025-12-22

Breaking

Session hooks API redesign: Merged branch event into session event. BranchEvent, BranchEventResult types and pi.on("branch", ...) removed. Use pi.on("session", ...) with reason: "before_branch" | "branch" instead. AgentSession.branch() returns { cancelled } instead of { skipped }. AgentSession.reset() and switchSession() now return boolean (false if cancelled by hook). RPC commands reset, switch_session, and branch now include cancelled in response data. (#278)

Added

Session lifecycle hooks: Added before_* variants (before_switch, before_clear, before_branch) that fire before actions and can be cancelled with { cancel: true }. Added shutdown reason for graceful exit handling. (#278)

Fixed

File tab completion display: File paths no longer get cut off early. Folders now show trailing / and removed redundant "directory"/"file" labels to maximize horizontal space. (#280)
Bash tool visual line truncation: Fixed bash tool output in collapsed mode to use visual line counting (accounting for line wrapping) instead of logical line counting. Now consistent with bash-execution.ts behavior. Extracted shared truncateToVisualLines utility. (#275)

[0.26.1] - 2025-12-22

Fixed

SDK tools respect cwd: Core tools (bash, read, edit, write, grep, find, ls) now properly use the cwd option from createAgentSession(). Added tool factory functions (createBashTool, createReadTool, etc.) for SDK users who specify custom cwd with explicit tools. (#279)

[0.26.0] - 2025-12-22

Added

SDK for programmatic usage: New createAgentSession() factory with full control over model, tools, hooks, skills, session persistence, and settings. Philosophy: "omit to discover, provide to override". Includes 12 examples and comprehensive documentation. (#272)
Project-specific settings: Settings now load from both ~/.pi/agent/settings.json (global) and <cwd>/.pi/settings.json (project). Project settings override global with deep merge for nested objects. Project settings are read-only (for version control). (#276)
SettingsManager static factories: SettingsManager.create(cwd?, agentDir?) for file-based settings, SettingsManager.inMemory(settings?) for testing. Added applyOverrides() for programmatic overrides.
SessionManager static factories: SessionManager.create(), SessionManager.open(), SessionManager.continueRecent(), SessionManager.inMemory(), SessionManager.list() for flexible session management.

[0.25.4] - 2025-12-22

Fixed

Syntax highlighting stderr spam: Fixed cli-highlight logging errors to stderr when markdown contains malformed code fences (e.g., missing newlines around closing backticks). Now validates language identifiers before highlighting and falls back silently to plain text. (#274)

[0.25.3] - 2025-12-21

Added

Gemini 3 preview models: Added gemini-3-pro-preview and gemini-3-flash-preview to the google-gemini-cli provider. (#264 by @LukeFost)
External editor support: Press Ctrl+G to edit your message in an external editor. Uses $VISUAL or $EDITOR environment variable. On successful save, the message is replaced; on cancel, the original is kept. (#266 by @aliou)
Process suspension: Press Ctrl+Z to suspend pi and return to the shell. Resume with fg as usual. (#267 by @aliou)
Configurable skills directories: Added granular control over skill sources with enableCodexUser, enableClaudeUser, enableClaudeProject, enablePiUser, enablePiProject toggles, plus customDirectories and ignoredSkills settings. (#269 by @nicobailon)
Skills CLI filtering: Added --skills <patterns> flag for filtering skills with glob patterns. Also added includeSkills setting and glob pattern support for ignoredSkills. (#268)

[0.25.2] - 2025-12-21

Fixed

Image shifting in tool output: Fixed an issue where images in tool output would shift down (due to accumulating spacers) each time the tool output was expanded or collapsed via Ctrl+O.

[0.25.1] - 2025-12-21

Fixed

Gemini image reading broken: Fixed the read tool returning images causing flaky/broken responses with Gemini models. Images in tool results are now properly formatted per the Gemini API spec.
Tab completion for absolute paths: Fixed tab completion producing //tmp instead of /tmp/. Also fixed symlinks to directories (like /tmp) not getting a trailing slash, which prevented continuing to tab through subdirectories.

[0.25.0] - 2025-12-20

Added

Interruptible tool execution: Queuing a message while tools are executing now interrupts the current tool batch. Remaining tools are skipped with an error result, and your queued message is processed immediately. Useful for redirecting the agent mid-task. (#259 by @steipete)
Google Gemini CLI OAuth provider: Access Gemini 2.0/2.5 models for free via Google Cloud Code Assist. Login with /login and select "Google Gemini CLI". Uses your Google account with rate limits.
Google Antigravity OAuth provider: Access Gemini 3, Claude (sonnet/opus thinking models), and GPT-OSS models for free via Google's Antigravity sandbox. Login with /login and select "Antigravity". Uses your Google account with rate limits.

Changed

Model selector respects --models scope: The /model command now only shows models specified via --models flag when that flag is used, instead of showing all available models. This prevents accidentally selecting models from unintended providers. (#255)

Fixed

Connection errors not retried: Added "connection error" to the list of retryable errors so Anthropic connection drops trigger auto-retry instead of silently failing. (#252)
Thinking level not clamped on model switch: Fixed TUI showing xhigh thinking level after switching to a model that doesn't support it. Thinking level is now automatically clamped to model capabilities. (#253)
Cross-model thinking handoff: Fixed error when switching between models with different thinking signature formats (e.g., GPT-OSS to Claude thinking models via Antigravity). Thinking blocks without signatures are now converted to text with <thinking> delimiters.

[0.24.5] - 2025-12-20

Fixed

Input buffering in iTerm2: Fixed Ctrl+C, Ctrl+D, and other keys requiring multiple presses in iTerm2. The cell size query response parser was incorrectly holding back keyboard input.

[0.24.4] - 2025-12-20

Fixed

Arrow keys and Enter in selector components: Fixed arrow keys and Enter not working in model selector, session selector, OAuth selector, and other selector components when Caps Lock or Num Lock is enabled. (#243)

[0.24.3] - 2025-12-19

Fixed

Footer overflow on narrow terminals: Fixed footer path display exceeding terminal width when resizing to very narrow widths, causing rendering crashes. /arminsayshi

[0.24.2] - 2025-12-20

Fixed

More Kitty keyboard protocol fixes: Fixed Backspace, Enter, Home, End, and Delete keys not working with Caps Lock enabled. The initial fix in 0.24.1 missed several key handlers that were still using raw byte detection. Now all key handlers use the helper functions that properly mask out lock key bits. (#243)

[0.24.1] - 2025-12-19

Added

OAuth and model config exports: Scripts using AgentSession directly can now import getAvailableModels, getApiKeyForModel, findModel, login, logout, and getOAuthProviders from @oh-my-pi/pi-coding-agent to reuse OAuth token storage and model resolution. (#245)
xhigh thinking level for gpt-5.2 models: The thinking level selector and shift+tab cycling now show xhigh option for gpt-5.2 and gpt-5.2-codex models (in addition to gpt-5.1-codex-max). (#236 by @theBucky)

Fixed

Hooks wrap custom tools: Custom tools are now executed through the hook wrapper, so tool_call/tool_result hooks can observe, block, and modify custom tool executions (consistent with hook type docs). (#248 by @nicobailon)
Hook onUpdate callback forwarding: The onUpdate callback is now correctly forwarded through the hook wrapper, fixing custom tool progress updates. (#238 by @nicobailon)
Terminal cleanup on Ctrl+C in session selector: Fixed terminal not being properly restored when pressing Ctrl+C in the session selector. (#247 by @aliou)
OpenRouter models with colons in IDs: Fixed parsing of OpenRouter model IDs that contain colons (e.g., openrouter:meta-llama/llama-4-scout:free). (#242 by @aliou)
Global AGENTS.md loaded twice: Fixed global AGENTS.md being loaded twice when present in both ~/.pi/agent/ and the current directory. (#239 by @aliou)
Kitty keyboard protocol on Linux: Fixed keyboard input not working in Ghostty on Linux when Num Lock is enabled. The Kitty protocol includes Caps Lock and Num Lock state in modifier values, which broke key detection. Now correctly masks out lock key bits when matching keyboard shortcuts. (#243)
Emoji deletion and cursor movement: Backspace, Delete, and arrow keys now correctly handle multi-codepoint characters like emojis. Previously, deleting an emoji would leave partial bytes, corrupting the editor state. (#240)

[0.24.0] - 2025-12-19

Added

Subagent orchestration example: Added comprehensive custom tool example for spawning and orchestrating sub-agents with isolated context windows. Includes scout/planner/reviewer/worker agents and workflow commands for multi-agent pipelines. (#215 by @nicobailon)
getMarkdownTheme() export: Custom tools can now import getMarkdownTheme() from @oh-my-pi/pi-coding-agent to use the same markdown styling as the main UI.
pi.exec() signal and timeout support: Custom tools and hooks can now pass { signal, timeout } options to pi.exec() for cancellation and timeout handling. The result includes a killed flag when the process was terminated.
Kitty keyboard protocol support: Shift+Enter, Alt+Enter, Shift+Tab, Ctrl+D, and all Ctrl+key combinations now work in Ghostty, Kitty, WezTerm, and other modern terminals. (#225 by @kim0)
Dynamic API key refresh: OAuth tokens (GitHub Copilot, Anthropic OAuth) are now refreshed before each LLM call, preventing failures in long-running agent loops where tokens expire mid-session. (#223 by @kim0)
/hotkeys command: Shows all keyboard shortcuts in a formatted table.
Markdown table borders: Tables now render with proper top and bottom borders.

Changed

Subagent example improvements: Parallel mode now streams updates from all tasks. Chain mode shows all completed steps during streaming. Expanded view uses proper markdown rendering with syntax highlighting. Usage footer shows turn count.
Skills standard compliance: Skills now adhere to the Agent Skills standard. Validates name (must match parent directory, lowercase, max 64 chars), description (required, max 1024 chars), and frontmatter fields. Warns on violations but remains lenient. Prompt format changed to XML structure. Removed {baseDir} placeholder in favor of relative paths. (#231)

Fixed

JSON mode stdout flush: Fixed race condition where pi --mode json could exit before all output was written to stdout, causing consumers to miss final events.
Symlinked tools, hooks, and slash commands: Discovery now correctly follows symlinks when scanning for custom tools, hooks, and slash commands. (#219, #232 by @aliou)

Breaking Changes

Custom tools now require index.ts entry point: Auto-discovered custom tools must be in a subdirectory with an index.ts file. The old pattern ~/.pi/agent/tools/mytool.ts must become ~/.pi/agent/tools/mytool/index.ts. This allows multi-file tools to import helper modules. Explicit paths via --tool or settings.json still work with any .ts file.
Hook tool_result event restructured: The ToolResultEvent now exposes full tool result data instead of just text. (#233)
- Removed: result: string field
- Added: content: (TextContent | ImageContent)[] - full content array
- Added: details: unknown - tool-specific details (typed per tool via discriminated union on toolName)
- ToolResultEventResult.result renamed to ToolResultEventResult.text (removed), use content instead
- Hook handlers returning { result: "..." } must change to { content: [{ type: "text", text: "..." }] }
- Built-in tool details types exported: BashToolDetails, ReadToolDetails, GrepToolDetails, FindToolDetails, LsToolDetails, TruncationResult
- Type guards exported for narrowing: isBashToolResult, isReadToolResult, isEditToolResult, isWriteToolResult, isGrepToolResult, isFindToolResult, isLsToolResult

[0.23.4] - 2025-12-18

Added

Syntax highlighting: Added syntax highlighting for markdown code blocks, read tool output, and write tool content. Uses cli-highlight with theme-aware color mapping and VS Code-style syntax colors. (#214 by @svkozak)
Intra-line diff highlighting: Edit tool now shows word-level changes with inverse highlighting when a single line is modified. Multi-line changes show all removed lines first, then all added lines.

Fixed

Gemini tool result format: Fixed tool result format for Gemini 3 Flash Preview which strictly requires { output: value } for success and { error: value } for errors. Previous format using { result, isError } was rejected by newer Gemini models. (#213, #220)
Google baseUrl configuration: Google provider now respects baseUrl configuration for custom endpoints or API proxies. (#216, #221 by @theBucky)
Google provider FinishReason: Added handling for new IMAGE_RECITATION and IMAGE_OTHER finish reasons. Upgraded @google/genai to 1.34.0.

[0.23.3] - 2025-12-17

Fixed

Check for compaction before submitting user prompt, not just after agent turn ends. This catches cases where user aborts mid-response and context is already near the limit.

Changed

Improved system prompt documentation section with clearer pointers to specific doc files for custom models, themes, skills, hooks, custom tools, and RPC.
Cleaned up documentation:
- theme.md: Added missing color tokens (thinkingXhigh, bashMode)
- skills.md: Rewrote with better framing and examples
- hooks.md: Fixed timeout/error handling docs, added import aliases section
- custom-tools.md: Added intro with use cases and comparison table
- rpc.md: Added missing hook_error event documentation
- README.md: Complete settings table, condensed philosophy section, standardized OAuth docs
Hooks loader now supports same import aliases as custom tools (@sinclair/typebox, @oh-my-pi/pi-ai, @oh-my-pi/pi-tui, @oh-my-pi/pi-coding-agent).

Breaking Changes

Hooks: turn_end event's toolResults type changed from AppMessage[] to ToolResultMessage[]. If you have hooks that handle turn_end events and explicitly type the results, update your type annotations.

[0.23.2] - 2025-12-17

Fixed

Fixed Claude models via GitHub Copilot re-answering all previous prompts in multi-turn conversations. The issue was that assistant message content was sent as an array instead of a string, which Copilot's Claude adapter misinterpreted. Also added missing Openai-Intent: conversation-edits header and fixed X-Initiator logic to check for any assistant/tool message in history. (#209)
Detect image MIME type via file magic (read tool and @file attachments), not filename extension.
Fixed markdown tables overflowing terminal width. Tables now wrap cell contents to fit available width instead of breaking borders mid-row. (#206 by @kim0)

[0.23.1] - 2025-12-17

Fixed

Fixed TUI performance regression caused by Box component lacking render caching. Built-in tools now use Text directly (like v0.22.5), and Box has proper caching for custom tool rendering.
Fixed custom tools failing to load from ~/.pi/agent/tools/ when pi is installed globally. Module imports (@sinclair/typebox, @oh-my-pi/pi-tui, @oh-my-pi/pi-ai) are now resolved via aliases.

[0.23.0] - 2025-12-17

Added

Custom tools: Extend pi with custom tools written in TypeScript. Tools can provide custom TUI rendering, interact with users via pi.ui (select, confirm, input, notify), and maintain state across sessions via onSession callback. See docs/custom-tools.md and examples/custom-tools/. (#190)
Hook and tool examples: Added examples/hooks/ and examples/custom-tools/ with working examples. Examples are now bundled in npm and binary releases.

Breaking Changes

Hooks: Replaced session_start and session_switch events with unified session event. Use event.reason ("start" | "switch" | "clear") to distinguish. Event now includes entries array for state reconstruction.

[0.22.5] - 2025-12-17

Fixed

Fixed --session flag not saving sessions in print mode (-p). The session manager was never receiving events because no subscriber was attached.

[0.22.4] - 2025-12-17

Added

--list-models [search] CLI flag to list available models with optional fuzzy search. Shows provider, model ID, context window, max output, thinking support, and image support. Only lists models with configured API keys. (#203)

Fixed

Fixed tool execution showing green (success) background while still running. Now correctly shows gray (pending) background until the tool completes.

[0.22.3] - 2025-12-16

Added

Streaming bash output: Bash tool now streams output in real-time during execution. The TUI displays live progress with the last 5 lines visible (expandable with ctrl+o). (#44)

Changed

Tool output display: When collapsed, tool output now shows the last N lines instead of the first N lines, making streaming output more useful.
Updated @oh-my-pi/pi-ai with X-Initiator header support for GitHub Copilot, ensuring agent calls are not deducted from quota. (#200 by @kim0)

Fixed

Fixed editor text being cleared during compaction. Text typed while compaction is running is now preserved. (#179)
Improved RGB to 256-color mapping for terminals without truecolor support. Now correctly uses grayscale ramp for neutral colors and preserves semantic tints (green for success, red for error, blue for pending) instead of mapping everything to wrong cube colors.
/think off now actually disables thinking for all providers. Previously, providers like Gemini with "dynamic thinking" enabled by default would still use thinking even when turned off. (#180 by @markusylisiurunen)

[0.22.2] - 2025-12-15

Changed

Updated @oh-my-pi/pi-ai with interleaved thinking enabled by default for Anthropic Claude 4 models.

[0.22.1] - 2025-12-15

Dedicated to Peter's shoulder (@steipete)

Changed

Updated @oh-my-pi/pi-ai with interleaved thinking support for Anthropic models.

[0.22.0] - 2025-12-15

Added

GitHub Copilot support: Use GitHub Copilot models via OAuth login (/login -> "GitHub Copilot"). Supports both github.com and GitHub Enterprise. Models are sourced from models.dev and include Claude, GPT, Gemini, Grok, and more. All models are automatically enabled after login. (#191 by @cau1k)

Fixed

Model selector fuzzy search now matches against provider name (not just model ID) and supports space-separated tokens where all tokens must match

[0.21.0] - 2025-12-14

Added

Inline image rendering: Terminals supporting Kitty graphics protocol (Kitty, Ghostty, WezTerm) or iTerm2 inline images now render images inline in tool output. Aspect ratio is preserved by querying terminal cell dimensions on startup. Toggle with /show-images command or terminal.showImages setting. Falls back to text placeholder on unsupported terminals or when disabled. (#177 by @nicobailon)
Gemini 3 Pro thinking levels: Thinking level selector now works with Gemini 3 Pro models. Minimal/low map to Google's LOW, medium/high map to Google's HIGH. (#176 by @markusylisiurunen)

Fixed

Fixed read tool failing on macOS screenshot filenames due to Unicode Narrow No-Break Space (U+202F) in timestamp. Added fallback to try macOS variant paths and consolidated duplicate expandPath functions into shared path-utils.ts. (#181 by @nicobailon)
Fixed double blank lines rendering after markdown code blocks (#173 by @markusylisiurunen)

[0.20.1] - 2025-12-13

Added

Exported skills API: loadSkillsFromDir, formatSkillsForPrompt, and related types are now exported for use by other packages (e.g., mom).

[0.20.0] - 2025-12-13

Breaking Changes

Pi skills now use SKILL.md convention: Pi skills must now be named SKILL.md inside a directory, matching Codex CLI format. Previously any *.md file was treated as a skill. Migrate by renaming ~/.pi/agent/skills/foo.md to ~/.pi/agent/skills/foo/SKILL.md.

Added

Display loaded skills on startup in interactive mode

[0.19.1] - 2025-12-12

Fixed

Documentation: Added skills system documentation to README (setup, usage, CLI flags, settings)

[0.19.0] - 2025-12-12

Added

Skills system: Auto-discover and load instruction files on-demand. Supports Claude Code (~/.claude/skills/*/SKILL.md), Codex CLI (~/.codex/skills/), and Pi-native formats (~/.pi/agent/skills/, .pi/skills/). Skills are listed in system prompt with descriptions, agent loads them via read tool when needed. Supports {baseDir} placeholder. Disable with --no-skills or skills.enabled: false in settings. (#169)
Version flag: Added --version / -v flag to display the current version and exit. (#170)

[0.18.2] - 2025-12-11

Added

Auto-retry on transient errors: Automatically retries requests when providers return overloaded, rate limit, or server errors (429, 500, 502, 503, 504). Uses exponential backoff (2s, 4s, 8s). Shows retry status in TUI with option to cancel via Escape. Configurable in settings.json via retry.enabled, retry.maxRetries, retry.baseDelayMs. RPC mode emits auto_retry_start and auto_retry_end events. (#157)
HTML export line numbers: Read tool calls in HTML exports now display line number ranges (e.g., file.txt:10-20) when offset/limit parameters are used, matching the TUI display format. Line numbers appear in yellow color for better visibility. (#166)

Fixed

Branch selector now works with single message: Previously the branch selector would not open when there was only one user message. Now it correctly allows branching from any message, including the first one. This is needed for checkpoint hooks to restore state from before the first message. (#163)
In-memory branching for --no-session mode: Branching now works correctly in --no-session mode without creating any session files. The conversation is truncated in memory.
Git branch indicator now works in subdirectories: The footer's git branch detection now walks up the directory hierarchy to find the git root, so it works when running pi from a subdirectory of a repository. (#156)

[0.18.1] - 2025-12-10

Added

Mistral provider: Added support for Mistral AI models. Set MISTRAL_API_KEY environment variable to use.

Fixed

Fixed print mode (-p) not exiting after output when custom themes are present (theme watcher now properly stops in print mode) (#161)

[0.18.0] - 2025-12-10

Added

Hooks system: TypeScript modules that extend agent behavior by subscribing to lifecycle events. Hooks can intercept tool calls, prompt for confirmation, modify results, and inject messages from external sources. Auto-discovered from ~/.pi/agent/hooks/*.ts and .pi/hooks/*.ts. Thanks to @nicobailon for the collaboration on the design and implementation. (#145, supersedes #158)
pi.send() API: Hooks can inject messages into the agent session from external sources (file watchers, webhooks, CI systems). If streaming, messages are queued; otherwise a new agent loop starts immediately.
--hook <path> CLI flag: Load hook files directly for testing without modifying settings.
Hook events: session_start, session_switch, agent_start, agent_end, turn_start, turn_end, tool_call (can block), tool_result (can modify), branch.
Hook UI primitives: ctx.ui.select(), ctx.ui.confirm(), ctx.ui.input(), ctx.ui.notify() for interactive prompts from hooks.
Hooks documentation: Full API reference at docs/hooks.md, shipped with npm package.

[0.17.0] - 2025-12-09

Changed

Simplified compaction flow: Removed proactive compaction (aborting mid-turn when threshold approached). Compaction now triggers in two cases only: (1) overflow error from LLM, which compacts and auto-retries, or (2) threshold crossed after a successful turn, which compacts without retry.
Compaction retry uses Agent.continue(): Auto-retry after overflow now uses the new continue() API instead of re-sending the user message, preserving exact context state.
Merged turn prefix summary: When a turn is split during compaction, the turn prefix summary is now merged into the main history summary instead of being stored separately.

Added

isCompacting property on AgentSession: Check if auto-compaction is currently running.
Session compaction indicator: When resuming a compacted session, displays "Session compacted N times" status message.

Fixed

Block input during compaction: User input is now blocked while auto-compaction is running to prevent race conditions.
Skip error messages in usage calculation: Context size estimation now skips both aborted and error messages, as neither have valid usage data.

[0.16.0] - 2025-12-09

Breaking Changes

New RPC protocol: The RPC mode (--mode rpc) has been completely redesigned with a new JSON protocol. The old protocol is no longer supported. See docs/rpc.md for the new protocol documentation and test/rpc-example.ts for a working example. Includes RpcClient TypeScript class for easy integration. (#91)

Changed

README restructured: Reorganized documentation from 30+ flat sections into 10 logical groups. Converted verbose subsections to scannable tables. Consolidated philosophy sections. Reduced size by ~60% while preserving all information.

[0.15.0] - 2025-12-09

Changed

Major code refactoring: Restructured codebase for better maintainability and separation of concerns. Moved files into organized directories (core/, modes/, utils/, cli/). Extracted AgentSession class as central session management abstraction. Split main.ts and tui-renderer.ts into focused modules. See DEVELOPMENT.md for the new code map. (#153)

[0.14.2] - 2025-12-08

Added

/debug command now includes agent messages as JSONL in the output

Fixed

Fix crash when bash command outputs binary data (e.g., curl downloading a video file)

[0.14.1] - 2025-12-08

Fixed

Fix build errors with tsgo 7.0.0-dev.20251208.1 by properly importing ReasoningEffort type

[0.14.0] - 2025-12-08

Breaking Changes

Custom themes require new color tokens: Themes must now include thinkingXhigh and bashMode color tokens. The theme loader provides helpful error messages listing missing tokens. See built-in themes (dark.json, light.json) for reference values.

Added

OpenAI compatibility overrides in models.json: Custom models using openai-completions API can now specify a compat object to override provider quirks (supportsStore, supportsDeveloperRole, supportsReasoningEffort, maxTokensField). Useful for LiteLLM, custom proxies, and other non-standard endpoints. (#133, thanks @fink-andreas for the initial idea and PR)
xhigh thinking level: Added xhigh thinking level for OpenAI codex-max models. Cycle through thinking levels with Shift+Tab; xhigh appears only when using a codex-max model. (#143)
Collapse changelog setting: Add "collapseChangelog": true to ~/.pi/agent/settings.json to show a condensed "Updated to vX.Y.Z" message instead of the full changelog after updates. Use /changelog to view the full changelog. (#148)
Bash mode: Execute shell commands directly from the editor by prefixing with ! (e.g., !ls -la). Output streams in real-time, is added to the LLM context, and persists in session history. Supports multiline commands, cancellation (Escape), truncation for large outputs, and preview/expand toggle (Ctrl+O). Also available in RPC mode via {"type":"bash","command":"..."}. (#112, original implementation by @markusylisiurunen)

[0.13.2] - 2025-12-07

Changed

Tool output truncation: All tools now enforce consistent truncation limits with actionable notices for the LLM. (#134)
- Limits: 2000 lines OR 50KB (whichever hits first), never partial lines
- read: Shows [Showing lines X-Y of Z. Use offset=N to continue]. If first line exceeds 50KB, suggests bash command
- bash: Tail truncation with temp file. Shows [Showing lines X-Y of Z. Full output: /tmp/...]
- grep: Pre-truncates match lines to 500 chars. Shows match limit and line truncation notices
- find/ls: Shows result/entry limit notices
- TUI displays truncation warnings in yellow at bottom of tool output (visible even when collapsed)

[0.13.1] - 2025-12-06

Added

Flexible Windows shell configuration: The bash tool now supports multiple shell sources beyond Git Bash. Resolution order: (1) custom shellPath in settings.json, (2) Git Bash in standard locations, (3) any bash.exe on PATH. This enables Cygwin, MSYS2, and other bash environments. Configure with ~/.pi/agent/settings.json: {"shellPath": "C:\\cygwin64\\bin\\bash.exe"}.

Fixed

Windows binary detection: Fixed Bun compiled binary detection on Windows by checking for URL-encoded %7EBUN in addition to $bunfs and ~BUN in import.meta.url. This ensures the binary correctly locates supporting files (package.json, themes, etc.) next to the executable.

[0.12.15] - 2025-12-06

Fixed

Editor crash with emojis/CJK characters: Fixed crash when pasting or typing text containing wide characters (emojis like ✅, CJK characters) that caused line width to exceed terminal width. The editor now uses grapheme-aware text wrapping with proper visible width calculation.

[0.12.14] - 2025-12-06

Added

Double-Escape Branch Shortcut: Press Escape twice with an empty editor to quickly open the /branch selector for conversation branching.

[0.12.13] - 2025-12-05

Changed

Faster startup: Version check now runs in parallel with TUI initialization instead of blocking startup for up to 1 second. Update notifications appear in chat when the check completes.

[0.12.12] - 2025-12-05

Changed

Footer display: Token counts now use M suffix for millions (e.g., 10.2M instead of 10184k). Context display shortened from 61.3% of 200k to 61.3%/200k.

Fixed

Multi-key sequences in inputs: Inputs like model search now handle multi-key sequences identically to the main prompt editor. (#122 by @markusylisiurunen)
Line wrapping escape codes: Fixed underline style bleeding into padding when wrapping long URLs. ANSI codes now attach to the correct content, and line-end resets only turn off underline (preserving background colors). (#109)

Added

Fuzzy search models and sessions: Implemented a simple fuzzy search for models and sessions (e.g., codexmax now finds gpt-5.1-codex-max). (#122 by @markusylisiurunen)
Prompt History Navigation: Browse previously submitted prompts using Up/Down arrow keys when the editor is empty. Press Up to cycle through older prompts, Down to return to newer ones or clear the editor. Similar to shell history and Claude Code's prompt history feature. History is session-scoped and stores up to 100 entries. (#121 by @nicobailon)
/resume Command: Switch to a different session mid-conversation. Opens an interactive selector showing all available sessions. Equivalent to the --resume CLI flag but can be used without restarting the agent. (#117 by @hewliyang)

[0.12.11] - 2025-12-05

Changed

Compaction UI: Simplified collapsed compaction indicator to show warning-colored text with token count instead of styled banner. Removed redundant success message after compaction. (#108)

Fixed

Print mode error handling: -p flag now outputs error messages and exits with code 1 when requests fail, instead of silently producing no output.
Branch selector crash: Fixed TUI crash when user messages contained Unicode characters (like ✔ or ›) that caused line width to exceed terminal width. Now uses proper truncateToWidth instead of substring.
Bash output escape sequences: Fixed incomplete stripping of terminal escape sequences in bash tool output. stripAnsi misses some sequences like standalone String Terminator (ESC \), which could cause rendering issues when displaying captured TUI output.
Footer overflow crash: Fixed TUI crash when terminal width is too narrow for the footer stats line. The footer now truncates gracefully instead of overflowing.

Added

authHeader option in models.json: Custom providers can set "authHeader": true to automatically add Authorization: Bearer <apiKey> header. Useful for providers that require explicit auth headers. (#81)
--append-system-prompt Flag: Append additional text or file contents to the system prompt. Supports both inline text and file paths. Complements --system-prompt for layering custom instructions without replacing the base system prompt. (#114 by @markusylisiurunen)
Thinking Block Toggle: Added Ctrl+T shortcut to toggle visibility of LLM thinking blocks. When toggled off, shows a static "Thinking..." label instead of full content. Useful for reducing visual clutter during long conversations. (#113 by @markusylisiurunen)

[0.12.10] - 2025-12-04

Added

Added gpt-5.1-codex-max model support

[0.12.9] - 2025-12-04

Added

/copy Command: Copy the last agent message to clipboard. Works cross-platform (macOS, Windows, Linux). Useful for extracting text from rendered Markdown output. (#105 by @markusylisiurunen)

[0.12.8] - 2025-12-04

Fix: Use CTRL+O consistently for compaction expand shortcut (not CMD+O on Mac)

[0.12.7] - 2025-12-04

Added

Context Compaction: Long sessions can now be compacted to reduce context usage while preserving recent conversation history. (#92, docs)
- /compact [instructions]: Manually compact context with optional custom instructions for the summary
- /autocompact: Toggle automatic compaction when context exceeds threshold
- Compaction summarizes older messages while keeping recent messages (default 20k tokens) verbatim
- Auto-compaction triggers when context reaches contextWindow - reserveTokens (default 16k reserve)
- Compacted sessions show a collapsible summary in the TUI (toggle with o key)
- HTML exports include compaction summaries as collapsible sections
- RPC mode supports {"type":"compact"} command and auto-compaction (emits compaction events)
Branch Source Tracking: Branched sessions now store branchedFrom in the session header, containing the path to the original session file. Useful for tracing session lineage.

[0.12.5] - 2025-12-03

Added

Forking/Rebranding Support: All branding (app name, config directory, environment variable names) is now configurable via piConfig in package.json. Forks can change piConfig.name and piConfig.configDir to rebrand the CLI without code changes. Affects CLI banner, help text, config paths, and error messages. (#95)

Fixed

Bun Binary Detection: Fixed Bun compiled binary failing to start after Bun updated its virtual filesystem path format from %7EBUN to $bunfs. (#95)

[0.12.4] - 2025-12-02

Added

RPC Termination Safeguard: When running as an RPC worker (stdin pipe detected), the CLI now exits immediately if the parent process terminates unexpectedly. Prevents orphaned RPC workers from persisting indefinitely and consuming system resources.

[0.12.3] - 2025-12-02

Fixed

Rate limit handling: Anthropic rate limit errors now trigger automatic retry with exponential backoff (base 10s, max 5 retries). Previously these errors would abort the request immediately.
Usage tracking during retries: Retried requests now correctly accumulate token usage from all attempts, not just the final successful one. Fixes artificially low token counts when requests were retried.

[0.12.2] - 2025-12-02

Changed

Removed support for gpt-4.5-preview and o3 models (not yet available)

[0.12.1] - 2025-12-02

Added

Models: Added support for OpenAI's new models:
- gpt-4.1 (128K context)
- gpt-4.1-mini (128K context)
- gpt-4.1-nano (128K context)
- o3 (200K context, reasoning model)
- o4-mini (200K context, reasoning model)

[0.12.0] - 2025-12-02

Added

-p, --print Flag: Run in non-interactive batch mode. Processes input message or piped stdin without TUI, prints agent response directly to stdout. Ideal for scripting, piping, and CI/CD integration. Exits after first response.
-P, --print-streaming Flag: Like -p, but streams response tokens as they arrive. Use --print-streaming --no-markdown for raw unformatted output.
--print-turn Flag: Continue processing tool calls and agent turns until the agent naturally finishes or requires user input. Combine with -p for complete multi-turn conversations.
--no-markdown Flag: Output raw text without Markdown formatting. Useful when piping output to tools that expect plain text.
Streaming Print Mode: Added internal printStreaming option for streaming output in non-TUI mode.
RPC Mode print Command: Send {"type":"print","content":"text"} to get formatted print output via print_output events.
Auto-Save in Print Mode: Print mode conversations are automatically saved to the session directory, allowing later resumption with --continue.
Thinking level options: Added --thinking-off, --thinking-minimal, --thinking-low, --thinking-medium, --thinking-high flags for directly specifying thinking level without the selector UI.

Changed

Simplified RPC Protocol: Replaced the prompt wrapper command with direct message objects. Send {"role":"user","content":"text"} instead of {"type":"prompt","message":"text"}. Better aligns with message format throughout the codebase.
RPC Message Handling: Agent now processes raw message objects directly, with timestamp auto-populated if missing.

[0.11.9] - 2025-12-02

Changed

Change Ctrl+I to Ctrl+P for model cycling shortcut to avoid collision with Tab key in some terminals

[0.11.8] - 2025-12-01

Fixed

Absolute glob patterns (e.g., /Users/foo/**/*.ts) are now handled correctly. Previously the leading / was being stripped, causing the pattern to be interpreted relative to the current directory.

[0.11.7] - 2025-12-01

Fixed

Fix read path traversal vulnerability. Paths are now validated to prevent reading outside the working directory or its parents. The read tool can read from cwd, its ancestors (for config files), and all descendants. Symlinks are resolved before validation.

[0.11.6] - 2025-12-01

Fixed

Fix --system-prompt <path> allowing the path argument to be captured by the message collection, causing "file not found" errors.

[0.11.5] - 2025-11-30

Fixed

Fixed fatal error "Cannot set properties of undefined (setting '0')" when editing empty files in the edit tool.
Simplified edit tool output: Shows only "Edited file.txt" for successful edits instead of verbose search/replace details.
Fixed fatal error in footer rendering when token counts contain NaN values due to missing usage data.

[0.11.4] - 2025-11-30

Fixed

Fixed chat rendering crash when messages contain preformatted/styled text (e.g., thinking traces with gray italic styling). The markdown renderer now preserves existing ANSI escape codes when they appear before inline elements.

[0.11.3] - 2025-11-29

Fixed

Fix file drop functionality for absolute paths

[0.11.2] - 2025-11-29

Fixed

Fixed TUI crash when pasting content containing tab characters. Tabs are now converted to 4 spaces before insertion.
Fixed terminal corruption after exit when shell integration sequences (OSC 133) appeared in bash output. These sequences are now stripped along with other ANSI codes.

[0.11.1] - 2025-11-29

Added

Added fd integration for file path autocompletion. Now uses fd for faster fuzzy file search

Fixed

Fixed keyboard shortcuts Ctrl+A, Ctrl+E, Ctrl+K, Ctrl+U, Ctrl+W, and word navigation (Option+Arrow) not working in VS Code integrated terminal and some other terminal emulators

[0.11.0] - 2025-11-29

Added

File-based Slash Commands: Create custom reusable prompts as .txt files in ~/.pi/slash-commands/. Files become /filename commands with first-line descriptions. Supports {{selection}} placeholder for referencing selected/attached content.
/branch Command: Create conversation branches from any previous user message. Opens a selector to pick a message, then creates a new session file starting from that point. Original message text is placed in the editor for modification.
Unified Content References: Both @path in messages and --file path CLI arguments now use the same attachment system with consistent MIME type detection.
Drag & Drop Files: Drop files onto the terminal to attach them to your message. Supports multiple files and both text and image content.

Changed

Model Selector with Search: The /model command now opens a searchable list. Type to filter models by name, use arrows to navigate, Enter to select.
Improved File Autocomplete: File path completion after @ now supports fuzzy matching and shows file/directory indicators.
Session Selector with Search: The --resume and --session flags now open a searchable session list with fuzzy filtering.
Attachment Display: Files added via @path are now shown as "Attached: filename" in the user message, separate from the prompt text.
Tab Completion: Tab key now triggers file path autocompletion anywhere in the editor, not just after @ symbol.

Fixed

Fixed autocomplete z-order issue where dropdown could appear behind chat messages
Fixed cursor position when navigating through wrapped lines in the editor
Fixed attachment handling for continued sessions to preserve file references

[0.10.6] - 2025-11-28

Changed

Show base64-truncated indicator for large images in tool output

Fixed

Fixed image dimensions not being read correctly from PNG/JPEG/GIF files
Fixed PDF images being incorrectly base64-truncated in display
Allow reading files from ancestor directories (needed for monorepo configs)

[0.10.5] - 2025-11-28

Added

Full multimodal support: attach images (PNG, JPEG, GIF, WebP) and PDFs to prompts using @path syntax or --file flag

Fixed

@-references now handle special characters in file names (spaces, quotes, unicode)
Fixed cursor positioning issues with multi-byte unicode characters in editor

[0.10.4] - 2025-11-28

Fixed

Removed padding on first user message in TUI to improve visual consistency.

[0.10.3] - 2025-11-28

Added

Added RPC mode (--rpc) for programmatic integration. Accepts JSON commands on stdin, emits JSON events on stdout. See RPC mode documentation for protocol details.

Changed

Refactored internal architecture to support multiple frontends (TUI, RPC) with shared agent logic.

[0.10.2] - 2025-11-26

Added

Added thinking level persistence. Default level stored in ~/.pi/settings.json, restored on startup. Per-session overrides saved in session files.
Added model cycling shortcut: Ctrl+I cycles through available models (or scoped models with -m flag).
Added automatic retry with exponential backoff for transient API errors (network issues, 500s, overload).
Cumulative token usage now shown in footer (total tokens used across all messages in session).
Added --system-prompt flag to override default system prompt with custom text or file contents.
Footer now shows estimated total cost in USD based on model pricing.

Changed

Replaced --models flag with -m/--model supporting multiple values. Specify models as provider/model@thinking (e.g., anthropic/claude-sonnet-4-20250514@high). Multiple -m flags scope available models for the session.
Thinking level border now persists visually after selector closes.
Improved tool result display with collapsible output (default collapsed, expand with Ctrl+O).

[0.10.1] - 2025-11-25

Added

Add custom model configuration via ~/.pi/models.json

[0.10.0] - 2025-11-25

Initial public release.

Added

Interactive TUI with streaming responses
Conversation session management with --continue, --resume, and --session flags
Multi-line input support (Shift+Enter or Option+Enter for new lines)
Tool execution: read, write, edit, bash, glob, grep, think
Thinking mode support for Claude with visual indicator and /thinking selector
File path autocompletion with @ prefix
Slash command autocompletion
/export command for HTML session export
/model command for runtime model switching
/session command for session statistics
Model provider support: Anthropic (Claude), OpenAI, Google (Gemini)
Git branch display in footer
Message queueing during streaming responses
OAuth integration for Gmail and Google Calendar access
HTML export with syntax highlighting and collapsible sections

[0.9.3] - 2025-11-24

Added

Added Anthropic Claude Opus 4.5 support

[0.9.2] - 2025-11-24

Fixed

Edit Tool Dollar Sign Bug: Fixed critical bug in the edit tool where String.replace() was interpreting $ as a special replacement pattern (e.g., $$, $&, $'). When trying to insert $ into code (like adding a dollar sign to a template literal), the replacement would silently fail and produce unchanged content, but the tool would incorrectly report success. Now uses indexOf + substring for raw string replacement without special character interpretation. Also added verification that content actually changed, rejecting with a clear error if the replacement produces identical content. (#53)

[0.9.0] - 2025-11-21

Added

/clear Command: New slash command to reset the conversation context and start a fresh session. Aborts any in-flight agent work, clears all messages, and creates a new session file. (#48)
Model Cycling with Thinking Levels: The --models flag now supports thinking level syntax (e.g., --models sonnet:high,haiku:low). When cycling models with Ctrl+P, the associated thinking level is automatically applied. The first model in the scope is used as the initial model when starting a new session. Both model and thinking level changes are now saved to session and settings for persistence. (#47)
--thinking Flag: New CLI flag to set thinking level directly (e.g., --thinking high). Valid values: off, minimal, low, medium, high. Takes highest priority over all other thinking level sources. (#45)

Breaking

Interactive Mode with Initial Prompt: Passing a prompt on the command line (e.g., pi "List files") now starts interactive mode with the prompt pre-submitted, instead of exiting after completion. Use --print or -p to get the previous non-interactive behavior (e.g., pi -p "List files"). This matches Claude CLI (-p) and Codex (exec) behavior. (#46)

Fixed

Slash Command Autocomplete: Fixed issue where pressing Enter on a highlighted slash command suggestion (e.g., typing /mod with /model highlighted) would submit the partial text instead of executing the selected command. Now Enter applies the completion and submits in one action. (#49)
Model Matching Priority: The --models flag now prioritizes exact matches over partial matches. Supports provider/modelId format (e.g., openrouter/openai/gpt-5.1-codex) for precise selection. Exact ID matches are tried before partial matching, so --models gpt-5.1-codex correctly selects gpt-5.1-codex instead of openai/gpt-5.1-codex-mini.
Markdown Link Rendering: Fixed links with identical text and href (e.g., https://github.com/badlogic/pi-mono/pull/48/files) being rendered twice. Now correctly compares raw text instead of styled text (which contains ANSI codes) when determining if link text matches href.

[0.8.5] - 2025-11-21

Fixed

Path Completion Hanging: Fixed catastrophic regex backtracking in path completion that caused the terminal to hang when text contained many / characters (e.g., URLs). Replaced complex regex with simple string operations. (#18)
Autocomplete Arrow Keys: Fixed issue where arrow keys would move both the autocomplete selection and the editor cursor simultaneously when the file selector list was shown.

[0.8.4] - 2025-11-21

Fixed

Read Tool Error Handling: Fixed issue where the read tool would return errors as successful text content instead of throwing. Now properly throws errors for file not found and offset out of bounds conditions.

[0.8.3] - 2025-11-21

Improved

Export HTML: Limited container width to 700px for better readability. Fixed message statistics to match /session command output with proper breakdown of User/Assistant/Tool Calls/Tool Results/Total messages.
Dark Theme: Increased visibility of editor border (darkGray from #303030 to #505050) and thinking minimal indicator (from #4e4e4e to #6e6e6e).

[0.8.0] - 2025-11-21

Added

Theme System: Full theming support with 44 customizable color tokens. Two built-in themes (dark, light) with auto-detection based on terminal background. Use /theme command to select themes interactively. Custom themes in ~/.pi/agent/themes/*.json support live editing - changes apply immediately when the file is saved. Themes use RGB hex values for consistent rendering across terminals. VS Code users: set terminal.integrated.minimumContrastRatio to 1 for proper color rendering. See Theme Documentation for details.

[0.7.29] - 2025-11-20

Improved

Read Tool Display: When the read tool is called with offset/limit parameters, the tool execution now displays the line range in a compact format (e.g., read src/main.ts:100-200 for offset=100, limit=100).

[0.7.28] - 2025-11-20

Added

Message Queuing: You can now send multiple messages while the agent is processing without waiting for the previous response to complete. Messages submitted during streaming are queued and processed based on your queue mode setting. Queued messages are shown in a pending area below the chat. Press Escape to abort and restore all queued messages to the editor. Use /queue to select between "one-at-a-time" (process queued messages sequentially, recommended) or "all" (process all queued messages at once). The queue mode setting is saved and persists across sessions. (#15)

[0.7.27] - 2025-11-20

Fixed

Slash Command Submission: Fixed issue where slash commands required two Enter presses to execute. Now pressing Enter on a slash command autocomplete suggestion immediately submits the command, while Tab still applies the completion for adding arguments. (#30)
Slash Command Autocomplete: Fixed issue where typing a typo then correcting it would not show autocomplete suggestions. Autocomplete now re-triggers when typing or backspacing in a slash command context. (#29)

[0.7.26] - 2025-11-20

Added

Tool Output Expansion: Press Ctrl+O to toggle between collapsed and expanded tool output display. Expands all tool call outputs (bash, read, write, etc.) to show full content instead of truncated previews. (#31)
Custom Headers: Added support for custom HTTP headers in models.json configuration. Headers can be specified at both provider and model level, with model-level headers overriding provider-level ones. This enables bypassing Cloudflare bot detection and other proxy requirements. (#39)

Fixed

Chutes AI Provider: Fixed 400 errors when using Chutes AI provider. Added compatibility fixes for store field exclusion, max_tokens parameter usage, and system prompt role handling. (#42 by @butelo)
Mistral/Chutes Syntax Error: Fixed syntax error in merged PR that used iif instead of if.
Anthropic OAuth Bug: Fixed bug where process.env.ANTHROPIC_API_KEY = undefined set the env var to string "undefined" instead of deleting it. Now uses delete operator.

[0.7.25] - 2025-11-20

Added

Model Cycling: Press Ctrl+P to quickly cycle through models. Use --models CLI argument to scope to specific models (e.g., --models claude-sonnet,gpt-4o). Supports pattern matching and smart version selection (prefers aliases over dated versions). (#37 by @fightbulc)

[0.7.24] - 2025-11-20

Added

Thinking Level Cycling: Press Shift+Tab to cycle through thinking levels (off → minimal → low → medium → high) for reasoning-capable models. Editor border color changes to indicate current level (gray → blue → cyan → magenta).

[0.7.23] - 2025-11-20

Added

Update Notifications: Interactive mode now checks for new versions on startup and displays a notification if an update is available.

Changed

System Prompt: Updated system prompt to instruct agent to output plain text summaries directly instead of using cat or bash commands to display what it did.

Fixed

File Path Completion: Removed 10-file limit in tab completion selector. All matching files and directories now appear in the completion list.
Absolute Path Completion: Fixed tab completion for absolute paths (e.g., /Applications). Absolute paths in the middle of text (like "hey /") now complete correctly. Also fixed crashes when trying to stat inaccessible files (like macOS .VolumeIcon.icns) during directory traversal.

[0.7.22] - 2025-11-19

Fixed

Long Line Wrapping: Fixed crash when rendering long lines without spaces (e.g., file paths). Long words now break character-by-character to fit within terminal width.

[0.7.21] - 2025-11-19

Fixed

Terminal Flicker: Fixed flicker at bottom of viewport (especially editor component) in xterm.js-based terminals (VS Code, etc.) by using per-line clear instead of clear-to-end sequence.
Background Color Rendering: Fixed black cells appearing at end of wrapped lines when using background colors. Completely rewrote text wrapping and background application to properly handle ANSI reset codes.
Tool Output: Strip ANSI codes from bash/tool output before rendering to prevent conflicts with TUI styling.

[0.7.20] - 2025-11-18

Fixed

Message Wrapping: Fixed word-based text wrapping for long lines in chat messages. Text now properly wraps at word boundaries while preserving ANSI styling (colors, bold, italic, etc.) across wrapped lines. Background colors now extend to the full width of each line. Empty lines in messages now render correctly with full-width background.

[0.7.18] - 2025-11-18

Fixed

Bash Tool Error Handling: Bash tool now properly throws errors for failed commands (non-zero exit codes), timeouts, and aborted executions. This ensures tool execution components display with red background when bash commands fail.
Thinking Traces Styling: Thinking traces now maintain gray italic styling throughout, even when containing inline code blocks, bold text, or other inline formatting

[0.7.17] - 2025-11-18

Added

New Model: Added gemini-3-pro-preview to Google provider.
OAuth Authentication: Added /login and /logout commands for OAuth-based authentication with Claude Pro/Max subscriptions. Tokens are stored in ~/.pi/agent/oauth.json with 0600 permissions and automatically refreshed when expired. OAuth tokens take priority over API keys for Anthropic models.

Fixed

Anthropic Aborted Thinking: Fixed error when resubmitting assistant messages with incomplete thinking blocks (from aborted streams). Thinking blocks without valid signatures are now converted to text blocks with <thinking> delimiters, preventing API rejection.
Model Selector Loading: Fixed models not appearing in /model selector until user started typing. Models now load asynchronously and re-render when available.
Input Paste Support: Added bracketed paste mode support to Input component, enabling paste of long OAuth authorization codes.

[0.7.16] - 2025-11-17

Fixed

Tool Error Display: Fixed edit tool (and all other tools) not showing error state correctly in TUI. Failed tool executions now properly display with red background and show the error message. Previously, the isError flag from tool execution events was not being passed to the UI component, causing all tool results to show with green (success) background regardless of whether they succeeded or failed.

[0.7.15] - 2025-11-17

Fixed

Anthropic OAuth Support: Added support for ANTHROPIC_OAUTH_TOKEN environment variable. The agent now checks for OAuth tokens before falling back to API keys for Anthropic models, enabling OAuth-based authentication.

[0.7.14] - 2025-11-17

Fixed

Mistral API Compatibility: Fixed compatibility with Mistral API by excluding the store field and using max_tokens instead of max_completion_tokens, and avoiding the developer role in system prompts.
Error Display: Fixed error message display in assistant messages to include proper spacing before the error text.
Message Streaming: Fixed missing message_start event when no partial message chunks were received during streaming.

[0.7.13] - 2025-11-16

Fixed

TUI Editor: Fixed unicode input support for umlauts (äöü), emojis (😀), and other extended characters. Previously the editor only accepted ASCII characters (32-126). Now properly handles all printable unicode while still filtering out control characters. (#20)

[0.7.12] - 2025-11-16

Added

Custom Models and Providers: Support for custom models and providers via ~/.pi/agent/models.json configuration file. Add local models (Ollama, vLLM, LM Studio) or any OpenAI-compatible, Anthropic-compatible, or Google-compatible API. File is reloaded on every /model selector open, allowing live updates without restart. (#21)
Added gpt-5.1-codex model to OpenAI provider (400k context, 128k max output, reasoning-capable).

Changed

Breaking: No longer hardcodes Anthropic/Claude as default provider/model. Now prefers sensible defaults per provider (e.g., claude-sonnet-4-5 for Anthropic, gpt-5.1-codex for OpenAI), or requires explicit selection in interactive mode.
Interactive mode now allows starting without a model, showing helpful error on message submission instead of failing at startup.
Non-interactive mode (CLI messages, JSON, RPC) still fails early if no model or API key is available.
Model selector now saves selected model as default in settings.json.
models.json validation errors (syntax + schema) now surface with precise file/field info in both CLI and /model selector.
Agent system prompt now includes absolute path to its own README.md for self-documentation.

Fixed

Fixed crash when restoring a session with a custom model that no longer exists or lost credentials. Now gracefully falls back to default model, logs the reason, and appends a warning message to the restored chat.
Footer no longer crashes when no model is selected.

[0.7.11] - 2025-11-16

Changed

The /model selector now filters models based on available API keys. Only models for which API keys are configured in environment variables are shown. This prevents selecting models that would fail due to missing credentials. A yellow hint is displayed at the top of the selector explaining this behavior. (#19)

[0.7.10] - 2025-11-14

Added

/branch command for creating conversation branches. Opens a selector showing all user messages in chronological order. Selecting a message creates a new session with all messages before the selected one, and places the selected message in the editor for modification or resubmission. This allows exploring alternative conversation paths without losing the current session. (fixes #16)

[0.7.9] - 2025-11-14

Changed

Editor: updated keyboard shortcuts to follow Unix conventions:
- Ctrl+W deletes the previous word (stopping at whitespace or punctuation)
- Ctrl+U deletes from cursor to start of line (at line start, merges with previous line)
- Ctrl+K deletes from cursor to end of line (at line end, merges with next line)
- Option+Backspace in Ghostty now behaves like Ctrl+W (delete word backwards)
- Cmd+Backspace in Ghostty now behaves like Ctrl+U (delete to start of line)

[0.7.8] - 2025-11-13

Changed

Updated README.md with /changelog slash command documentation

[0.7.7] - 2025-11-13

Added

Automatic changelog display on startup in interactive mode. When starting a new session (not continuing/resuming), the agent will display all changelog entries since the last version you used. The last shown version is tracked in ~/.pi/agent/settings.json.
/changelog command to display the changelog in the TUI
OpenRouter Auto Router model support (#5)
Windows Git Bash support with automatic detection and process tree termination (#1)

Changed

BREAKING: Renamed project context file from AGENT.md to AGENTS.md. The system now looks for AGENTS.md or CLAUDE.md (with AGENTS.md preferred). Existing AGENT.md files will need to be renamed to AGENTS.md to continue working. (fixes #9)
BREAKING: Session file format changed to store provider and model ID separately instead of as a single provider/modelId string. Existing sessions will not restore the model correctly when resumed - you'll need to manually set the model again using /model. (fixes #4)
Improved Windows Git Bash detection logic with better error messages showing actual paths searched (#13)

Fixed

Fixed markdown list rendering bug where bullets were not displayed when list items contained inline code with cyan color formatting
Fixed context percentage showing 0% in footer when last assistant message was aborted (#12)
Fixed error message loss when turn_end event contains an error. Previously, errors in turn_end events (e.g., "Provider returned error" from OpenRouter Auto Router) were not captured in agent.state.error, making it appear as if the agent completed successfully. (#6)

[0.7.6] - 2025-11-13

Previous releases did not maintain a changelog.

1.5 MiB Raw Blame History Unescape Escape

Changelog

[Unreleased]

Fixed

[17.1.5] - 2026-07-27

Added

Changed

Fixed

[17.1.4] - 2026-07-26

Added

Changed

Fixed

[17.1.3] - 2026-07-24

Fixed

[17.1.2] - 2026-07-24

Added

Changed

Fixed

Removed

[17.1.1] - 2026-07-24

Added

Changed

Fixed

[17.1.0] - 2026-07-24

Breaking Changes

Added

Changed

Fixed

[17.0.9] - 2026-07-23

Added

Changed

Fixed

[17.0.8] - 2026-07-22

Added

Changed

Removed

[17.0.7] - 2026-07-21

Fixed

[17.0.6] - 2026-07-20

Added

Changed

Fixed

Removed

[17.0.5] - 2026-07-18

Added

Changed

Fixed

[17.0.4] - 2026-07-18

Fixed

[17.0.3] - 2026-07-17

Changed

Fixed

Removed

[17.0.2] - 2026-07-17

Added

Changed

Fixed

[17.0.1] - 2026-07-16

Changed

Fixed

Removed

[17.0.1] - 2026-07-16

Removed

Fixed

Added

[17.0.0] - 2026-07-15

Breaking Changes

Added

Changed

Fixed

[16.5.2] - 2026-07-14

Breaking Changes

Added

Changed

Fixed

[16.5.1] - 2026-07-14

Changed

[16.5.0] - 2026-07-13

Breaking Changes

Added

1.5 MiB

Raw Blame History