- Added a `mode` property to `CompactOptions` to allow fine-grained control over compaction strategies.
- Implemented `soft`, `remote`, and `snapcompact` submode overrides for the `/compact` command.
- Integrated `parseCompactArgs` to enable robust subcommand routing and validation, including focus instruction rejection for specific modes.
- Established a `CompactMode` registry to manage compaction strategies and verify remote availability.
- Validated every stage of a pipeline against `simple_command_is_safe` instead of only the first stage to prevent improper segmentation of compound shell constructs.
- Guarded segment re-execution by verifying that each `Display`-reconstructed command parses back to the expected pipeline shape.
- Configured segmented-chain execution to fall back to an unsegmented, whole-command path whenever a reconstructed segment diverges from the original AST.
- Resolved a syntax error during command execution by preventing `Display` from stripping terminators from compound commands like `while` and `for` loops.
- Removed qrcode and qrcode-view subcommands from the collab command.
- Updated showCollabLink to always trigger QR code display.
- Refactored internal verb logic to simplify session sharing flow.
Drop the runtime qrcode + @types/qrcode packages in favor of a zero-dependency
byte-mode QR encoder (versions 1-40, EC L/M/Q/H, auto version + mask selection)
with a half-block ANSI renderer. Cross-validated byte-for-byte against the qrcode
reference library and decoded end-to-end with jsQR. Adds encoder regression tests.
- Added collab.webUrl and rendered browser links as web UI wrappers whose fragments carry relay links.
- Added one-shot /collab qrcode and /collab qrcode-view commands with terminal QR rendering.
- Updated coding-agent and collab-web parsers to prefer parseable wrapper fragments while preserving legacy links.
- Added regression tests and changelog entries for split-host collab links and QR commands.
- Migrated the `pi` AI dialect from legacy XML-style elements to a compact, token-frugal sigil-delimited format using `§` for calls, `""` for body fences, and `¤` for thinking.
- Implemented robust parsing for the new format, including support for incremental streaming of tool arguments and automatic escalation of body fences to prevent content collisions.
- Updated documentation, settings configurations, and test suites across the `ai` and `coding-agent` packages to ensure consistency with the new dialect rules.
- Standardized moderation category terminology in `stats` to improve clarity in reporting metrics.
getOpenRouterRouteSuffix() used strict parseThinkingLevel(), which never
recognizes the max->xhigh alias, so openrouter/<id>:max was consumed as an
OpenRouter route suffix and cloned into a literal <id>:max model id with the
reasoning level dropped. This hit every exact-selector funnel
(parseModelPattern -> resolveCliModel/--model, resolveModelRoleValue/modelRoles
+ model picker, SDK default role) for the dominant aggregator provider.
Exclude max via parseThinkingSuffix(.., MAX_THINKING_SUFFIX_OPTIONS) so the
pattern falls through to the existing max-aware selector split. Literal :max
ids stay safe (none exist under openrouter; nanogpt literals win via exact
lookup before this path). Adds an openrouter/<id>:max regression test.
Pretty-printed HTML lists (e.g. <ul>\n <li>...) leaked source
indentation before bullets and inserted blank rows between items
because whitespace-only text between block-level tags was appended
literally. Skip insignificant formatting whitespace and pin it with
a regression test.
The dispose() disconnect added by this PR awaited mcpManager.disconnectAll()
unbounded. An owned manager holding an HTTP/SSE server whose session-
termination DELETE hangs would block dispose for the full MCP request timeout
(30s default, unbounded when OMP_MCP_TIMEOUT_MS=0), stalling /exit and
print-mode shutdown on a broken remote endpoint.
Wrap the disconnect in withTimeout(..., 3_000) — mirroring the bounded
async-job teardown two lines above and the startup bound from issue #2100.
stdio close (the subprocess reap this PR targets) completes well within the
bound; a slow transport close is left to finish detached.
Adds a regression test driving the real MCPManager.disconnectAll() with a
stalled transport close.
The quoted-path fix entry was appended under the already-released [16.0.6] section (creating a duplicate ### Fixed heading); released sections are immutable per repo convention. Move it to [Unreleased] and normalize the bullet.
Resolved Amazon Bedrock application inference profile ARNs through the provider-specific model resolver and routed Bedrock requests to the ARN region.
Fixes#3004
Adds the missing regression for the credentialId-based invalidation (commit d9ae691): stick a session to an account at index>=1, remove a lower-index account so indices shift, restart, and assert the session re-resolves to the SAME account by id rather than the account now occupying the stale recorded index. Guards against the index-only resurrection the review flagged.
Threading printThoughts only into runPrintMode is too late when
hideThinkingBlock is set (settings or --hide-thinking): the session is
built with hideThinkingSummary=true, so the provider omits reasoning
summaries and --print-thoughts prints nothing. Override hideThinkingBlock
to false for single-shot print mode before session creation, gated on
print mode; an explicit --hide-thinking still wins.
After plan approval the executor delivers the plan-mode-reference exactly
once and sets `#planReferenceSent = true`. Both compaction paths — `compact()`
and `#runAutoCompaction()` — replace the conversation history that carried
that reference but never cleared the flag, so `#buildPlanReferenceMessage()`
short-circuited to null on every subsequent turn and the executor permanently
lost the plan it was working on (exactly the long-session failure reported).
Clear `#planReferenceSent` right after `replaceMessages()` in both paths so the
next turn re-reads the plan from disk and re-injects it. The reset is a no-op
for ordinary sessions: the default plan path (PLAN.md in session-local scratch)
has no file on disk, so `#buildPlanReferenceMessage()` still returns null there.
Adds a deterministic regression test (short-circuited compaction, mock stream)
that fails before this change and passes after, plus a guard proving normal
sessions get no spurious plan injection.
Fixes#1246
GitHub Copilot's `anthropic-messages` proxy (api.githubcopilot.com) forwards
to signature-enforcing Anthropic and returns full thinking signatures, but the
compat builder classified it as a non-signing reasoning endpoint via the
generic `reasoning && !official` default (`replayUnsignedThinking: true`).
When a checkpoint/branch-return turn is an abandoned tool-use turn (adaptive
Opus emits a tool call then ends on `stop`/`end_turn`), `transformMessages`
correctly strips its end_turn-bound, unreplayable signature. On a
`replayUnsignedThinking` endpoint the encoder then re-emitted that block as
`{ type: "thinking", signature: "" }`. An empty signature is rejected by the
signature-enforcing backend with `400 Invalid signature`, which corrupts the
session and re-trips on every full history re-send (e.g. after toggling MCP
servers).
Exclude github-copilot from `replayUnsignedThinking` so unsigned/stripped
thinking degrades to text exactly like the official Anthropic API — wire-valid
and lossless of the tool_use pairing. Z.AI / DeepSeek / other 3p reasoning
endpoints (#2005) and cross-model preservation (#2257/#2265) are unaffected.
Tests:
- packages/catalog/test/anthropic-copilot-signing-compat.test.ts: copilot
(incl. enterprise copilot-api.* hosts) -> replayUnsignedThinking false;
generic 3p reasoning -> true; official -> false. Fails before / passes after.
- packages/ai/test/anthropic-copilot-checkpoint-thinking-signature.test.ts:
a signing copilot model never emits an empty-signature thinking block for a
historical checkpoint turn (demotes to text, keeps tool_use), and still
replays a clean signed historical thinking block natively.
- Refined the advisor's scope to prioritize concrete technical risks and user advocacy while discouraging process-related critiques.
- Explicitly barred the advisor from intervening on user intent, process questions, or issues already handled by developer tooling.
- Updated `advise` tool documentation to clarify its purpose in preventing wasteful or incorrect work.
A login entry can store credentials under a different provider id via
storeCredentialsAs (e.g. openai-codex-device => openai-codex). Filtering
only on provider.id left such alias logins visible after disabling the
underlying model provider. Surface storeCredentialsAs on OAuthProviderInfo
and hide a login entry when either its own id or its storeCredentialsAs
target is in disabledProviders.
Addresses Codex review on #2906.
can1357