ToolArgsRevealController.setTarget initialized new entries with revealed=0, so the first message_update returned { __partialJson: "" } even when the provider had already parsed a complete chunk. For renderers without exposeRawPartialJson (e.g. write), the throttled re-parse + cached displayArgs short-circuited every subsequent setTarget, leaving the preview body blank until tool_execution_end.
Seed revealed with the full incoming partialJson length on entry creation (clamped to a surrogate-safe boundary). The first frame now carries the parsed path/content immediately; subsequent message_updates extend target and the reveal ticks pace only the newly arrived bytes — no field is ever truncated because the seeded prefix is the longest the entry has seen so far.
Fixes#3881
fix(compaction): cap snapcompact frame payloads (#3866)
Bound rebuilt snapcompact image payloads by a per-request base64 byte
budget so long sessions stop re-sending multi-megabyte standing image
archives on every provider request; auto-compaction falls back to
context-full summaries when snapcompact output is too large.
Resolved snapcompact.ts conflict against the main font-rendering refactor
by keeping both renderabilityProbeText and the frame-budget helpers.
Fixed historyBlocks to emit the omitted-frame notice before the kept
(newer) images, since the byte budget drops the oldest frames — keeping
reconstructed blocks oldest-to-newest (addresses Codex P2 review).
Fixes#3792
- Added eight new Go-specific rules to the discovery package.
- Registered the new Go rules in the default rule source index.
- Covered the new Go AST matching conditions with test cases in `builtin-defaults.test.ts`.
Forwarded persisted provider stream timeout settings into model requests so slow local LLM streams can widen or disable first-event and idle watchdogs without environment variables.
Fixes#3878
Normalized string-encoded JSON arrays in grep/search path handling so direct execute paths match validated tool-call behavior.
Added regression coverage for direct GrepTool.execute paths supplied as a JSON-array-shaped string.
Fixes#3873
The yield tool's per-call schema validator was skipped entirely for incremental
yields (`type: ["<label>"]`), so when a subagent emitted a non-conforming value
for a known section (e.g. DeepSeek-v4-pro returning "Correct"/"correct."/"approved"
for the reviewer's `overall_correctness` enum), the call succeeded locally and
the model got no retry feedback. The mismatch only surfaced post-mortem in
`finalizeSubprocessOutput` as a fatal `schema_violation` — the parent agent
lost the entire result, with no recourse for the subagent to fix it.
Build a per-label sub-validator map alongside the full-schema validator: each
entry validates one section's `data` against its top-level property's sub-schema
(items schema for array-typed labels like `findings`). The yield tool runs this
map for incremental yields and routes failures through the same MAX_SCHEMA_RETRIES
budget the terminal path uses, so the model sees up to three corrective retries
and the existing schema-override safety net accepts the value with
SUBAGENT_WARNING_SCHEMA_OVERRIDDEN after exhaustion. Unknown labels remain
unconstrained so scratchpad/streaming sections still pass.
Fixes#3870
When legacy snapcompact archives exceed the per-request byte budget, retain frames from the newest end of the archived middle and restore oldest-to-newest order for the kept subset.
Apply the snapcompact frame byte-budget cap even when the active model has no known context window, avoiding 80-frame archives on custom vision models.
Bounded persisted snapcompact image archives by base64 byte size so large sessions stop re-sending multi-megabyte frame walls on every provider request.
Auto snapcompact now falls back to context-full summaries when rendered frame payloads exceed the byte budget, and legacy oversized archives omit over-budget frames during LLM context rebuilds.
Fixes#3792
- Updated the default browser User-Agent string to emulate a modern version of Chrome.
- Added typical browser headers to the outgoing fetch request, including Sec-Ch-Ua, Sec-Fetch flags, and Referer.
- Added a blank "b" parameter to the form body to match native DuckDuckGo HTML search behavior.
StdinBuffer held a bare `\x1b\x1b` chunk and timer-flushed it as one
sequence. `parseKey("\x1b\x1b")` returns undefined, so CustomEditor
fell through to the base editor and never fired the configured `onEscape` —
the double-escape gesture and the second-press single-Esc handler both went
dead whenever the terminal batched the two presses into one stdin read.
Split an exact bare `\x1b\x1b` into two ESC events only after the
flush window proves no follower arrived. If a follower does arrive, emit the
first ESC and restart parsing at the second ESC so legacy Alt chords
(`\x1bd`, `\x1b\x7f`) remain one downstream keypress. Meta-CSI/SS3
chords (`\x1b\x1b[A`, `\x1b\x1bO…`) still emit as one combined
sequence.
EventController.tool_execution_update re-armed the working loader when a
transient overlay (auto-compaction / auto-retry / handoff) had torn it down
mid-tool; tool_execution_end did not. A subagent (`task`) call only fires
_end, so a task result landing after such an overlay left the UI looking
idle even though the session was still streaming. Mirror the reconciler
call in #handleToolExecutionEnd.
Fixes#3857
`tool_execution_end` for a long-running tool (`task` subagent, async bash
poll, …) is the next streaming event on the parent session when an inner
transient overlay (auto-snapcompact, auto-context-full, auto-retry) nulled the
working loader between the tool's start and end. The overlay-end handlers are
the only loader restorers keyed off the missing reference; if the subagent's
`tool_execution_end` lands while the overlay is still active (or its end
handler errored before re-arming), the spinner stays gone for the rest of the
parent turn even though the agent keeps streaming.
`#handleToolExecutionEnd` now calls `#ensureWorkingLoaderWhileStreaming()`
at the top, mirroring `tool_execution_update` so the working loader survives
a subagent completing inside the overlay window.
Refs #3858
StdinBuffer held a bare `\x1b\x1b` chunk (or emitted it as one when followed by
a non-CSI byte). `parseKey("\x1b\x1b")` returns undefined, so CustomEditor
fell through to the base editor and never fired the configured `onEscape` —
the double-escape gesture and the second-press single-Esc handler both went
dead whenever the terminal batched the two presses into one stdin read.
Split a bare `\x1b\x1b` into two ESC events at the buffer layer, mirroring
the existing split for ESC + SGR mouse report. Meta-CSI/SS3 chords
(`\x1b\x1b[A`, `\x1b\x1bO…`) still emit as one combined sequence.
EventController.tool_execution_update re-armed the working loader when a
transient overlay (auto-compaction / auto-retry / handoff) had torn it down
mid-tool; tool_execution_end did not. A subagent (`task`) call only fires
_end, so a task result landing after such an overlay left the UI looking
idle even though the session was still streaming. Mirror the reconciler
call in #handleToolExecutionEnd.
Fixes#3857
- Migrated global service tier settings to a per-model-family architecture (OpenAI, Anthropic, Google).
- Implemented `ServiceTierByFamily` mapping to allow independent configuration and resolution per provider.
- Added automatic migration logic for legacy service tier and fast-mode application settings.
- Updated telemetry, session management, and task execution to support provider-specific tier resolution.
- Introduced `isProbablyBinary` utility to sniff file headers for NUL bytes or invalid UTF-8 sequences.
- Updated `ReadTool` to use the binary sniffer, preventing mojibake corruption in output when reading non-text files.
- Refined `file-mentions` auto-reads to skip binary files and mark them as `binary` in the message transcript.
- Added comprehensive unit tests for binary detection logic, covering NUL bytes, truncated multibyte characters, and path-based file sniffing.
AgentSession.switchSession() eagerly called buildDisplaySessionContext()
before setSessionFile, walking the previous session's branch and expanding
every compaction entry's snapcompact archive and openaiRemoteCompaction
replacementHistory into messages. For huge pre-fix sessions that materialized
GBs of data and OOMed in-TUI /resume even after the streaming loader fix.
The snapshot is only needed for same-session reloads, where
#didSessionMessagesChange compares the pre/post message arrays to detect
rollback edits. Different-session switches skip the call entirely; the
error-recovery path rebuilds the previous context on demand from the
restored state so MCP-selection restoration still has its inputs.
Added a regression test (test/agent-session-switch-prev-context.test.ts)
that spies on sessionManager.buildSessionContext across switchSession and
asserts the expected call count and target file per branch.
Fixes#3846
Dirty isolated baselines can be accidentally committed by subagents that run git add -A. Fetching the raw isolation HEAD then cherry-picking the range would replay that baseline WIP into parent history.
Add a dirty-baseline replay path that rewrites each agent commit against the captured baseline tree, preserving the agent commit message and author while excluding staged, unstaged, and untracked changes that existed before isolation started. Clean baselines still use the raw git fetch path, and nested-only changes keep returning patches without creating an empty root branch.
Add a regression for baseline staged + untracked WIP committed by the agent, asserting the task branch contains only the agent file and parent WIP remains staged/untracked after merge.
Fixes#3842
When an isolated task agent commits its own changes before yielding, the
harness used to collapse the captured delta into one AI-summarized commit
and discard the agent's commit messages and authorship entirely. This
violated commit discipline for agentic swarms — multiple logical commits
("fix bug" + "add test") became a single opaque commit, and the
agent's commit object (which lived in isolation/.git/objects under
overlayfs/rcopy) was lost when cleanupIsolation tore down the overlay.
commitToBranch now detects when isolation HEAD moved past baseline.root
.headCommit. When it has, the function git-fetches the agent's HEAD into
the parent repo as omp/task/${taskId} so the commit objects survive
cleanupIsolation, and stamps the captured baselineSha onto the returned
CommitToBranchResult. mergeTaskBranches cherry-picks the inclusive range
baseSha..branchName when baseSha is provided, replaying each agent
commit verbatim with its original message and author. Any uncommitted
leftover (staged, unstaged, untracked) on top of the agent's last commit
becomes one trailing AI-summarized commit on the same branch.
Falls back to the legacy single-commit path when the agent never moved
HEAD (purely dirty working tree); existing patch-mode flow is untouched.
Fixes#3842
Applied isolated branch patches with three-way fallback when unrelated parent dirt appears in patch context.
Surfaced branch preparation failures instead of reporting no changes.
Fixes#3841
- Always remove surfaced irc:incoming records from the pending-aside queue; the inbox tool result already injects the body, so leaving them queued would auto-inject a duplicate at the next step.
- Updated the inbox tool to drain pending asides regardless of peek.
- Added a regression test asserting a peeked pending aside does not auto-inject.
- Drained running-session IRC asides through the inbox tool before the model step consumes them.
- Added a regression test for messages delivered while the recipient is already running.
Fixes#3834
Codex review on #3829: when an MCP server lives in a non-writable
source config such as opencode.json with enabled:false, the dashboard
re-enable had nowhere to write to — the writable mcp.json fallback
did not own the server, so setMcpServerEnabled fell through to the
denylist and the source's enabled:false kept the row disabled.
Added a parallel allowlist to the user-level mcp.json that overrides
a non-writable source's enabled:false flag without ever mutating the
foreign config:
- types + schema: new enabledServers array (mirrors disabledServers).
- config-writer: readEnabledServers + setServerForceEnabled helpers,
and setMcpServerEnabled now writes to enabledServers on enable
when no writable mcp.json owns the server, clears it whenever a
writable source becomes the source of truth, and always clears the
override on disable so a force-enabled server can be turned off.
- mcp/config (runtime loader) and state-manager (dashboard read):
honor enabledServers as an override on enabled:false, while still
letting disabledServers win.
- Added a regression test that walks the full lifecycle for an
opencode.json server: enabled:false is surfaced as disabled, the
dashboard re-enable force-enables via enabledServers without
touching opencode.json, then disable clears the override and
populates disabledServers.
Fixes#3827
Codex review on #3829: the dashboard re-enable path still missed MCP
servers loaded from supported non-primary native config files such as
.omp/.mcp.json or user .mcp.json. Those rows carry enabled:false from
their source file, so falling back to the user disabledServers denylist
could not make the row active again.
- setMcpServerEnabled now accepts the loaded row's sourcePath and checks
it before the primary project/user mcp.json paths.
- extension-dashboard passes the source path for writable MCP providers
(native and mcp-json), avoiding accidental edits to third-party tool
configs while still updating .omp/.mcp.json and standalone MCP JSON
sources.
- Added a regression test for a server loaded from .omp/.mcp.json with
enabled:false; re-enable flips that file to enabled:true and does not
write the denylist.
Fixes#3827
Codex review on #3829: when an MCP server's mcp.json entry carries
enabled:false, the dashboard toggle previously only removed the name
from the user-level disabledServers denylist. state-manager's new
`server.enabled === false` check (state-manager.ts:156) then still
marked the row disabled, leaving such servers impossible to re-enable
from /extensions.
Extracted setMcpServerEnabled() into mcp/config-writer.ts mirroring
/mcp enable | /mcp disable semantics:
- Server defined in project mcp.json -> update enabled on that entry.
- Else server defined in user mcp.json -> update enabled on that entry.
- Else (discovered third-party server) -> use the user-level disabledServers denylist.
- On re-enable, always clear any stale denylist entry.
extension-dashboard.ts routes mcp:* toggles through this helper. Added
four new regression tests covering: enabled:false re-enable, mixed
flag+denylist re-enable, disable on a config-resident server writing
enabled:false (not denylist), and discovered-server denylist round-trip.
Fixes#3827
Two read paths previously diverged on whether an MCP server was active or
disabled. /mcp list (slash-commands/helpers/mcp.ts:388) treats a server
as disabled when config.enabled === false OR the name is in the
user-level disabledServers denylist; the runtime MCP loader does the
same in mcp/config.ts:115. The /extensions dashboard only consulted
the dashboard-private settings.disabledExtensions array, so a server
disabled via /mcp disable or enabled:false kept showing as active.
Toggling MCP servers from the dashboard had the mirror problem: it only
wrote to settings.disabledExtensions, so /mcp list never noticed.
- state-manager: read user-level disabledServers from mcp.json once and
consider enabled:false / denylist membership when deriving each MCP
extension's state, matching /mcp list semantics.
- extension-dashboard: route mcp:* toggles through setServerDisabled
against the canonical mcp.json denylist, and clean any legacy
settings.disabledExtensions entry on re-enable so it doesn't keep the
server marked disabled.
- Added a regression test exercising both read signals and the
setServerDisabled round-trip the dashboard's MCP toggle now uses.
Fixes#3827
- Update `releaseProviderInFlightLease` to signal into the specific directory path associated with the lease rather than recomputing it from the root.
- Introduce `signalProviderInFlightWaitersInDir` to decoupling waking waiters from global provider path resolution.
- Remove redundant tests from `coding-agent`.
The initial PR update changed the MCP OAuth default prompt to 'login consent'.
The reporter verified Cloudflare's flow actually matches the reference MCP SDK
when no prompt parameter is sent: Cloudflare then reuses the existing account
grant and opens the scope/permission picker first. Forcing any prompt keeps the
flow on Cloudflare's account/consent page instead.
Match the reference SDK behavior: omit prompt by default and send
'prompt=consent' only when the requested scope contains offline_access, where
OIDC Core requires re-consent for offline access. Explicit oauth.prompt values,
including the empty-string omit escape hatch, still take precedence.
Also rename the dynamically registered MCP OAuth client from Codex to oh-my-pi
so Cloudflare consent screens show the current product name.
Fixes#3817
The MCP OAuth flow defaulted the authorization-request prompt parameter
to 'consent'. Per OpenID Connect Core 1.0 §3.1.2.1 that asks the
authorization server to re-prompt for consent only while reusing the
existing browser authentication session. Cloudflare's MCP OAuth server
(and other strict OIDC providers) honor that literally, so /mcp reauth
landed on the consent screen attached to whichever account the browser
cookie was for, leaving no way to switch the signed-in account.
Default to 'login consent' instead so the provider first re-prompts for
authentication (the page Claude Code shows on its reauth flow) and then
re-confirms consent, preserving the original intent of always
re-displaying the authorize screen. RFC 6749 §3.1 requires providers to
ignore prompt values they do not support, so the two-value form is safe
for non-OIDC servers. Existing per-server overrides via mcp.json's
`oauth.prompt` (including the empty-string escape hatch) are unchanged.
Fixes#3817
can1357