fix(grok): improve skills, plans, usage, and turn reliability

[t3dotgg]

Grok threads miss skills, reasoning controls, proposed plans, and usage data. Some turns stay on Working after ACP stops responding.

Continues #7638 by @1xpixi. The original commits and authorship are preserved, including work from @SunkenInTime, @ahmed-besic, @mbrow190, @lnieuwenhuis, and @gbarros-dev. This PR retains the original Grok changes and adds fixes found during review:

• Keep "Always allow this session" scoped to the matching operation. A different command still asks. A changed shell-command description does not invalidate the approval.
• Use the provider instance's environment when finding Grok plan files.
• Fix the shared ACP parser's unchecked array access so the server typecheck passes.
• Add focused tests and an opt-in real-Grok turn test.

Verification

• 194 focused tests passed, including Grok, shared ACP, Cursor, and usage handling.
• Five real CLI tests passed against Grok 1.0.5, covering authentication, model discovery, reasoning selection, and a streamed model response.
• Real Grok adapter checks passed in a private test home: approve a command for the session, repeat it without another prompt, request approval for a different command, deny it, and capture a proposed plan without implementing it.
• Parsed token and cost records written by the real Grok CLI.
• Server and web typechecks passed. Changed-file lint and formatting passed.
• Mobile typecheck reports 56 navigation errors. Clean main at d3c24a14b produces identical output. No new mobile type errors.

No live T3 state or normal Grok state was changed. Fresh browser verification and before/after captures have not been run. The screenshots below are the contributor's existing evidence from #7638, not captures of this follow-up branch.

Original UI evidence

Grok usage, from @1xpixi's Windows desktop build:

Grok reasoning control, from the same build:

Model: GPT-5.6 Sol. Harness: Codex in T3 Code.

---

Note

Medium Risk
Touches Grok turn lifecycle, permission mapping, and ACP parsing used beyond Grok; behavior changes (stall detection, rate-limit errors, session approval scoping) can affect live threads if mis-tuned.

Overview
Grok integration gains skill discovery (grok inspect --json), ACP-driven reasoning effort in the model picker and per-turn session/set_model metadata, and plan mode support that surfaces plan.md writes and _x.ai/exit_plan_mode as proposed-plan events while unblocking the native approval gate.

The Grok adapter adds a turn liveness watchdog (long silence after ACP progress fails the turn; longer window while tools run; refreshes on steer, plans, approvals, and user-input), maps usage limits from xAI rate_limit completions, and improves permissions: session-scoped auto-approval by operation identity, fallback when Grok omits allow_always, and CLI argv that mirror T3 runtime modes (--permission-mode / --always-approve).

Shared ACP changes emit in-progress tool updates, bound oversized tool output (including whitespace-only buffers), and coalesce streaming command output using content length—not just detail. The mobile usage chart adds Grok as a provider band with theme-aware colors. The ACP mock agent grows scenario flags to exercise these paths in tests.

^{Reviewed by Cursor Bugbot for commit 01cc3ac. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add Grok turn watchdog, plan mode, usage tracking, and reasoning effort

• Adds an inactivity watchdog to GrokAdapter.ts that cancels and settles stalled ACP turns (10 min default, 30 min with active tools), with liveness tracking around approvals and user input.
• Detects enter_plan_mode and exit_plan_mode tool calls, emits turn.proposed.completed events when Grok writes plan.md, and intercepts x.ai/exit_plan_mode to capture plan content without native gating.
• Adds Grok Build usage support across the stack: bumps usage contract to v5 with backward-compatible merge for v4, parses updates.jsonl transcripts via parseGrokLine in usageTranscripts.ts, and adds Grok series to web and mobile usage charts.
• Propagates per-turn reasoning effort to setSessionModel metadata in GrokAcpSupport.ts, validates tokens, and skips redundant calls.
• Discovers Grok skills at runtime via grok inspect --json in GrokSkills.ts with a 4 s timeout and soft-fail to empty.
• Improves tool-call update coalescing in AcpRuntimeModel.ts to emit on status changes and track combined output growth (detail/content/rawOutput) instead of detail length alone.
• Risk: USAGE_CONTRACT_VERSION bumps from 4 to 5 in usage.ts; older clients below USAGE_MERGE_COMPATIBLE_SINCE (4) will be marked stale. selectGrokPermissionOptionId falls back to allow_once when acceptForSession is missing, changing approval behavior for sessions where Grok omits that option.

^{Macroscope summarized 01cc3ac.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: c29d928a-c7b2-41ac-8687-bc7e3fca5c02

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.3 KiB	+20 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+2 B (+0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.4 KiB	+18 B (+0.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.6 KiB	+44 B (+0.1%)	66.4 KiB	✅
Codex	Live turn messages	10	11	+1 (+10.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−5 B (−0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−5 B (−0.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.4 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	11	10	−1 (−9.1%)	21	✅

Baseline: d3c24a1 · PR result: 01cc3ac · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[t3dotgg]

Grok threads miss skills, reasoning controls, proposed plans, and usage data. Some turns stay on Working after ACP stops responding.

Continues #7638 by @1xpixi. The original commits and authorship are preserved, including work from @SunkenInTime, @ahmed-besic, @mbrow190, @lnieuwenhuis, and @gbarros-dev. This PR retains the original Grok changes and adds fixes found during review:

• Keep "Always allow this session" scoped to the matching operation. A different command still asks. A changed shell-command description does not invalidate the approval.
• Use the provider instance's environment when finding Grok plan files.
• Fix the shared ACP parser's unchecked array access so the server typecheck passes.
• Add focused tests and an opt-in real-Grok turn test.

Verification

• 194 focused tests passed, including Grok, shared ACP, Cursor, and usage handling.
• Five real CLI tests passed against Grok 1.0.5, covering authentication, model discovery, reasoning selection, and a streamed model response.
• Real Grok adapter checks passed in a private test home: approve a command for the session, repeat it without another prompt, request approval for a different command, deny it, and capture a proposed plan without implementing it.
• Parsed token and cost records written by the real Grok CLI.
• Server and web typechecks passed. Changed-file lint and formatting passed.
• Mobile typecheck reports 56 navigation errors. Clean main at d3c24a14b produces identical output. No new mobile type errors.

No live T3 state or normal Grok state was changed. Fresh browser verification and before/after captures have not been run. The screenshots below are the contributor's existing evidence from #7638, not captures of this follow-up branch.

Original UI evidence

Grok usage, from @1xpixi's Windows desktop build:

Grok reasoning control, from the same build:

Model: GPT-5.6 Sol. Harness: Codex in T3 Code.

---

Note

Medium Risk
Touches Grok turn lifecycle, permission mapping, and ACP parsing used beyond Grok; behavior changes (stall detection, rate-limit errors, session approval scoping) can affect live threads if mis-tuned.

Overview
Grok integration gains skill discovery (grok inspect --json), ACP-driven reasoning effort in the model picker and per-turn session/set_model metadata, and plan mode support that surfaces plan.md writes and _x.ai/exit_plan_mode as proposed-plan events while unblocking the native approval gate.

The Grok adapter adds a turn liveness watchdog (long silence after ACP progress fails the turn; longer window while tools run; refreshes on steer, plans, approvals, and user-input), maps usage limits from xAI rate_limit completions, and improves permissions: session-scoped auto-approval by operation identity, fallback when Grok omits allow_always, and CLI argv that mirror T3 runtime modes (--permission-mode / --always-approve).

Shared ACP changes emit in-progress tool updates, bound oversized tool output (including whitespace-only buffers), and coalesce streaming command output using content length—not just detail. The mobile usage chart adds Grok as a provider band with theme-aware colors. The ACP mock agent grows scenario flags to exercise these paths in tests.

^{Reviewed by Cursor Bugbot for commit 01cc3ac. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add Grok turn watchdog, plan mode, usage tracking, and reasoning effort

• Adds an inactivity watchdog to GrokAdapter.ts that cancels and settles stalled ACP turns (10 min default, 30 min with active tools), with liveness tracking around approvals and user input.
• Detects enter_plan_mode and exit_plan_mode tool calls, emits turn.proposed.completed events when Grok writes plan.md, and intercepts x.ai/exit_plan_mode to capture plan content without native gating.
• Adds Grok Build usage support across the stack: bumps usage contract to v5 with backward-compatible merge for v4, parses updates.jsonl transcripts via parseGrokLine in usageTranscripts.ts, and adds Grok series to web and mobile usage charts.
• Propagates per-turn reasoning effort to setSessionModel metadata in GrokAcpSupport.ts, validates tokens, and skips redundant calls.
• Discovers Grok skills at runtime via grok inspect --json in GrokSkills.ts with a 4 s timeout and soft-fail to empty.
• Improves tool-call update coalescing in AcpRuntimeModel.ts to emit on status changes and track combined output growth (detail/content/rawOutput) instead of detail length alone.
• Risk: USAGE_CONTRACT_VERSION bumps from 4 to 5 in usage.ts; older clients below USAGE_MERGE_COMPATIBLE_SINCE (4) will be marked stale. selectGrokPermissionOptionId falls back to allow_once when acceptForSession is missing, changing approval behavior for sessions where Grok omits that option.

^{Macroscope summarized 01cc3ac.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: c29d928a-c7b2-41ac-8687-bc7e3fca5c02

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.3 KiB	+20 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+2 B (+0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.4 KiB	+18 B (+0.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.6 KiB	+44 B (+0.1%)	66.4 KiB	✅
Codex	Live turn messages	10	11	+1 (+10.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−5 B (−0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−5 B (−0.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.4 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	11	10	−1 (−9.1%)	21	✅

Baseline: d3c24a1 · PR result: 01cc3ac · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[t3dotgg]

Grok threads miss skills, reasoning controls, proposed plans, and usage data. Some turns stay on Working after ACP stops responding.

Continues #7638 by @1xpixi. The original commits and authorship are preserved, including work from @SunkenInTime, @ahmed-besic, @mbrow190, @lnieuwenhuis, and @gbarros-dev. This PR retains the original Grok changes and adds fixes found during review:

• Keep "Always allow this session" scoped to the matching operation. A different command still asks. A changed shell-command description does not invalidate the approval.
• Use the provider instance's environment when finding Grok plan files.
• Fix the shared ACP parser's unchecked array access so the server typecheck passes.
• Add focused tests and an opt-in real-Grok turn test.

Verification

• 194 focused tests passed, including Grok, shared ACP, Cursor, and usage handling.
• Five real CLI tests passed against Grok 1.0.5, covering authentication, model discovery, reasoning selection, and a streamed model response.
• Real Grok adapter checks passed in a private test home: approve a command for the session, repeat it without another prompt, request approval for a different command, deny it, and capture a proposed plan without implementing it.
• Parsed token and cost records written by the real Grok CLI.
• Server and web typechecks passed. Changed-file lint and formatting passed.
• Mobile typecheck reports 56 navigation errors. Clean main at d3c24a14b produces identical output. No new mobile type errors.

No live T3 state or normal Grok state was changed. Fresh browser verification and before/after captures have not been run. The screenshots below are the contributor's existing evidence from #7638, not captures of this follow-up branch.

Original UI evidence

Grok usage, from @1xpixi's Windows desktop build:

Grok reasoning control, from the same build:

Model: GPT-5.6 Sol. Harness: Codex in T3 Code.

---

Note

Medium Risk
Touches Grok turn lifecycle, permission mapping, and ACP parsing used beyond Grok; behavior changes (stall detection, rate-limit errors, session approval scoping) can affect live threads if mis-tuned.

Overview
Grok integration gains skill discovery (grok inspect --json), ACP-driven reasoning effort in the model picker and per-turn session/set_model metadata, and plan mode support that surfaces plan.md writes and _x.ai/exit_plan_mode as proposed-plan events while unblocking the native approval gate.

The Grok adapter adds a turn liveness watchdog (long silence after ACP progress fails the turn; longer window while tools run; refreshes on steer, plans, approvals, and user-input), maps usage limits from xAI rate_limit completions, and improves permissions: session-scoped auto-approval by operation identity, fallback when Grok omits allow_always, and CLI argv that mirror T3 runtime modes (--permission-mode / --always-approve).

Shared ACP changes emit in-progress tool updates, bound oversized tool output (including whitespace-only buffers), and coalesce streaming command output using content length—not just detail. The mobile usage chart adds Grok as a provider band with theme-aware colors. The ACP mock agent grows scenario flags to exercise these paths in tests.

^{Reviewed by Cursor Bugbot for commit 01cc3ac. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add Grok turn watchdog, plan mode, usage tracking, and reasoning effort

• Adds an inactivity watchdog to GrokAdapter.ts that cancels and settles stalled ACP turns (10 min default, 30 min with active tools), with liveness tracking around approvals and user input.
• Detects enter_plan_mode and exit_plan_mode tool calls, emits turn.proposed.completed events when Grok writes plan.md, and intercepts x.ai/exit_plan_mode to capture plan content without native gating.
• Adds Grok Build usage support across the stack: bumps usage contract to v5 with backward-compatible merge for v4, parses updates.jsonl transcripts via parseGrokLine in usageTranscripts.ts, and adds Grok series to web and mobile usage charts.
• Propagates per-turn reasoning effort to setSessionModel metadata in GrokAcpSupport.ts, validates tokens, and skips redundant calls.
• Discovers Grok skills at runtime via grok inspect --json in GrokSkills.ts with a 4 s timeout and soft-fail to empty.
• Improves tool-call update coalescing in AcpRuntimeModel.ts to emit on status changes and track combined output growth (detail/content/rawOutput) instead of detail length alone.
• Risk: USAGE_CONTRACT_VERSION bumps from 4 to 5 in usage.ts; older clients below USAGE_MERGE_COMPATIBLE_SINCE (4) will be marked stale. selectGrokPermissionOptionId falls back to allow_once when acceptForSession is missing, changing approval behavior for sessions where Grok omits that option.

^{Macroscope summarized 01cc3ac.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: c29d928a-c7b2-41ac-8687-bc7e3fca5c02

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.3 KiB	+20 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+2 B (+0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.4 KiB	+18 B (+0.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.6 KiB	+44 B (+0.1%)	66.4 KiB	✅
Codex	Live turn messages	10	11	+1 (+10.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−5 B (−0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−5 B (−0.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.4 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	11	10	−1 (−9.1%)	21	✅

Baseline: d3c24a1 · PR result: 01cc3ac · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[t3dotgg]

Grok threads miss skills, reasoning controls, proposed plans, and usage data. Some turns stay on Working after ACP stops responding.

Continues #7638 by @1xpixi. The original commits and authorship are preserved, including work from @SunkenInTime, @ahmed-besic, @mbrow190, @lnieuwenhuis, and @gbarros-dev. This PR retains the original Grok changes and adds fixes found during review:

• Keep "Always allow this session" scoped to the matching operation. A different command still asks. A changed shell-command description does not invalidate the approval.
• Use the provider instance's environment when finding Grok plan files.
• Fix the shared ACP parser's unchecked array access so the server typecheck passes.
• Add focused tests and an opt-in real-Grok turn test.

Verification

• 194 focused tests passed, including Grok, shared ACP, Cursor, and usage handling.
• Five real CLI tests passed against Grok 1.0.5, covering authentication, model discovery, reasoning selection, and a streamed model response.
• Real Grok adapter checks passed in a private test home: approve a command for the session, repeat it without another prompt, request approval for a different command, deny it, and capture a proposed plan without implementing it.
• Parsed token and cost records written by the real Grok CLI.
• Server and web typechecks passed. Changed-file lint and formatting passed.
• Mobile typecheck reports 56 navigation errors. Clean main at d3c24a14b produces identical output. No new mobile type errors.

No live T3 state or normal Grok state was changed. Fresh browser verification and before/after captures have not been run. The screenshots below are the contributor's existing evidence from #7638, not captures of this follow-up branch.

Original UI evidence

Grok usage, from @1xpixi's Windows desktop build:

Grok reasoning control, from the same build:

Model: GPT-5.6 Sol. Harness: Codex in T3 Code.

---

Note

Medium Risk
Touches Grok turn lifecycle, permission mapping, and ACP parsing used beyond Grok; behavior changes (stall detection, rate-limit errors, session approval scoping) can affect live threads if mis-tuned.

Overview
Grok integration gains skill discovery (grok inspect --json), ACP-driven reasoning effort in the model picker and per-turn session/set_model metadata, and plan mode support that surfaces plan.md writes and _x.ai/exit_plan_mode as proposed-plan events while unblocking the native approval gate.

The Grok adapter adds a turn liveness watchdog (long silence after ACP progress fails the turn; longer window while tools run; refreshes on steer, plans, approvals, and user-input), maps usage limits from xAI rate_limit completions, and improves permissions: session-scoped auto-approval by operation identity, fallback when Grok omits allow_always, and CLI argv that mirror T3 runtime modes (--permission-mode / --always-approve).

Shared ACP changes emit in-progress tool updates, bound oversized tool output (including whitespace-only buffers), and coalesce streaming command output using content length—not just detail. The mobile usage chart adds Grok as a provider band with theme-aware colors. The ACP mock agent grows scenario flags to exercise these paths in tests.

^{Reviewed by Cursor Bugbot for commit 01cc3ac. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add Grok turn watchdog, plan mode, usage tracking, and reasoning effort

• Adds an inactivity watchdog to GrokAdapter.ts that cancels and settles stalled ACP turns (10 min default, 30 min with active tools), with liveness tracking around approvals and user input.
• Detects enter_plan_mode and exit_plan_mode tool calls, emits turn.proposed.completed events when Grok writes plan.md, and intercepts x.ai/exit_plan_mode to capture plan content without native gating.
• Adds Grok Build usage support across the stack: bumps usage contract to v5 with backward-compatible merge for v4, parses updates.jsonl transcripts via parseGrokLine in usageTranscripts.ts, and adds Grok series to web and mobile usage charts.
• Propagates per-turn reasoning effort to setSessionModel metadata in GrokAcpSupport.ts, validates tokens, and skips redundant calls.
• Discovers Grok skills at runtime via grok inspect --json in GrokSkills.ts with a 4 s timeout and soft-fail to empty.
• Improves tool-call update coalescing in AcpRuntimeModel.ts to emit on status changes and track combined output growth (detail/content/rawOutput) instead of detail length alone.
• Risk: USAGE_CONTRACT_VERSION bumps from 4 to 5 in usage.ts; older clients below USAGE_MERGE_COMPATIBLE_SINCE (4) will be marked stale. selectGrokPermissionOptionId falls back to allow_once when acceptForSession is missing, changing approval behavior for sessions where Grok omits that option.

^{Macroscope summarized 01cc3ac.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: c29d928a-c7b2-41ac-8687-bc7e3fca5c02

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.3 KiB	+20 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+2 B (+0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.4 KiB	+18 B (+0.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.6 KiB	+44 B (+0.1%)	66.4 KiB	✅
Codex	Live turn messages	10	11	+1 (+10.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−5 B (−0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−5 B (−0.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.4 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	11	10	−1 (−9.1%)	21	✅

Baseline: d3c24a1 · PR result: 01cc3ac · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[t3dotgg]

Grok threads miss skills, reasoning controls, proposed plans, and usage data. Some turns stay on Working after ACP stops responding.

Continues #7638 by @1xpixi. The original commits and authorship are preserved, including work from @SunkenInTime, @ahmed-besic, @mbrow190, @lnieuwenhuis, and @gbarros-dev. This PR retains the original Grok changes and adds fixes found during review:

• Keep "Always allow this session" scoped to the matching operation. A different command still asks. A changed shell-command description does not invalidate the approval.
• Use the provider instance's environment when finding Grok plan files.
• Fix the shared ACP parser's unchecked array access so the server typecheck passes.
• Add focused tests and an opt-in real-Grok turn test.

Verification

• 194 focused tests passed, including Grok, shared ACP, Cursor, and usage handling.
• Five real CLI tests passed against Grok 1.0.5, covering authentication, model discovery, reasoning selection, and a streamed model response.
• Real Grok adapter checks passed in a private test home: approve a command for the session, repeat it without another prompt, request approval for a different command, deny it, and capture a proposed plan without implementing it.
• Parsed token and cost records written by the real Grok CLI.
• Server and web typechecks passed. Changed-file lint and formatting passed.
• Mobile typecheck reports 56 navigation errors. Clean main at d3c24a14b produces identical output. No new mobile type errors.

No live T3 state or normal Grok state was changed. Fresh browser verification and before/after captures have not been run. The screenshots below are the contributor's existing evidence from #7638, not captures of this follow-up branch.

Original UI evidence

Grok usage, from @1xpixi's Windows desktop build:

Grok reasoning control, from the same build:

Model: GPT-5.6 Sol. Harness: Codex in T3 Code.

---

Note

Medium Risk
Touches Grok turn lifecycle, permission mapping, and ACP parsing used beyond Grok; behavior changes (stall detection, rate-limit errors, session approval scoping) can affect live threads if mis-tuned.

Overview
Grok integration gains skill discovery (grok inspect --json), ACP-driven reasoning effort in the model picker and per-turn session/set_model metadata, and plan mode support that surfaces plan.md writes and _x.ai/exit_plan_mode as proposed-plan events while unblocking the native approval gate.

The Grok adapter adds a turn liveness watchdog (long silence after ACP progress fails the turn; longer window while tools run; refreshes on steer, plans, approvals, and user-input), maps usage limits from xAI rate_limit completions, and improves permissions: session-scoped auto-approval by operation identity, fallback when Grok omits allow_always, and CLI argv that mirror T3 runtime modes (--permission-mode / --always-approve).

Shared ACP changes emit in-progress tool updates, bound oversized tool output (including whitespace-only buffers), and coalesce streaming command output using content length—not just detail. The mobile usage chart adds Grok as a provider band with theme-aware colors. The ACP mock agent grows scenario flags to exercise these paths in tests.

^{Reviewed by Cursor Bugbot for commit 01cc3ac. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add Grok turn watchdog, plan mode, usage tracking, and reasoning effort

• Adds an inactivity watchdog to GrokAdapter.ts that cancels and settles stalled ACP turns (10 min default, 30 min with active tools), with liveness tracking around approvals and user input.
• Detects enter_plan_mode and exit_plan_mode tool calls, emits turn.proposed.completed events when Grok writes plan.md, and intercepts x.ai/exit_plan_mode to capture plan content without native gating.
• Adds Grok Build usage support across the stack: bumps usage contract to v5 with backward-compatible merge for v4, parses updates.jsonl transcripts via parseGrokLine in usageTranscripts.ts, and adds Grok series to web and mobile usage charts.
• Propagates per-turn reasoning effort to setSessionModel metadata in GrokAcpSupport.ts, validates tokens, and skips redundant calls.
• Discovers Grok skills at runtime via grok inspect --json in GrokSkills.ts with a 4 s timeout and soft-fail to empty.
• Improves tool-call update coalescing in AcpRuntimeModel.ts to emit on status changes and track combined output growth (detail/content/rawOutput) instead of detail length alone.
• Risk: USAGE_CONTRACT_VERSION bumps from 4 to 5 in usage.ts; older clients below USAGE_MERGE_COMPATIBLE_SINCE (4) will be marked stale. selectGrokPermissionOptionId falls back to allow_once when acceptForSession is missing, changing approval behavior for sessions where Grok omits that option.

^{Macroscope summarized 01cc3ac.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: c29d928a-c7b2-41ac-8687-bc7e3fca5c02

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.3 KiB	+20 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+2 B (+0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.4 KiB	+18 B (+0.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.6 KiB	+44 B (+0.1%)	66.4 KiB	✅
Codex	Live turn messages	10	11	+1 (+10.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−5 B (−0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−5 B (−0.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.4 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	11	10	−1 (−9.1%)	21	✅

Baseline: d3c24a1 · PR result: 01cc3ac · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[t3dotgg]

Grok threads miss skills, reasoning controls, proposed plans, and usage data. Some turns stay on Working after ACP stops responding.

Continues #7638 by @1xpixi. The original commits and authorship are preserved, including work from @SunkenInTime, @ahmed-besic, @mbrow190, @lnieuwenhuis, and @gbarros-dev. This PR retains the original Grok changes and adds fixes found during review:

• Keep "Always allow this session" scoped to the matching operation. A different command still asks. A changed shell-command description does not invalidate the approval.
• Use the provider instance's environment when finding Grok plan files.
• Fix the shared ACP parser's unchecked array access so the server typecheck passes.
• Add focused tests and an opt-in real-Grok turn test.

Verification

• 194 focused tests passed, including Grok, shared ACP, Cursor, and usage handling.
• Five real CLI tests passed against Grok 1.0.5, covering authentication, model discovery, reasoning selection, and a streamed model response.
• Real Grok adapter checks passed in a private test home: approve a command for the session, repeat it without another prompt, request approval for a different command, deny it, and capture a proposed plan without implementing it.
• Parsed token and cost records written by the real Grok CLI.
• Server and web typechecks passed. Changed-file lint and formatting passed.
• Mobile typecheck reports 56 navigation errors. Clean main at d3c24a14b produces identical output. No new mobile type errors.

No live T3 state or normal Grok state was changed. Fresh browser verification and before/after captures have not been run. The screenshots below are the contributor's existing evidence from #7638, not captures of this follow-up branch.

Original UI evidence

Grok usage, from @1xpixi's Windows desktop build:

Grok reasoning control, from the same build:

Model: GPT-5.6 Sol. Harness: Codex in T3 Code.

---

Note

Medium Risk
Touches Grok turn lifecycle, permission mapping, and ACP parsing used beyond Grok; behavior changes (stall detection, rate-limit errors, session approval scoping) can affect live threads if mis-tuned.

Overview
Grok integration gains skill discovery (grok inspect --json), ACP-driven reasoning effort in the model picker and per-turn session/set_model metadata, and plan mode support that surfaces plan.md writes and _x.ai/exit_plan_mode as proposed-plan events while unblocking the native approval gate.

The Grok adapter adds a turn liveness watchdog (long silence after ACP progress fails the turn; longer window while tools run; refreshes on steer, plans, approvals, and user-input), maps usage limits from xAI rate_limit completions, and improves permissions: session-scoped auto-approval by operation identity, fallback when Grok omits allow_always, and CLI argv that mirror T3 runtime modes (--permission-mode / --always-approve).

Shared ACP changes emit in-progress tool updates, bound oversized tool output (including whitespace-only buffers), and coalesce streaming command output using content length—not just detail. The mobile usage chart adds Grok as a provider band with theme-aware colors. The ACP mock agent grows scenario flags to exercise these paths in tests.

^{Reviewed by Cursor Bugbot for commit 01cc3ac. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add Grok turn watchdog, plan mode, usage tracking, and reasoning effort

• Adds an inactivity watchdog to GrokAdapter.ts that cancels and settles stalled ACP turns (10 min default, 30 min with active tools), with liveness tracking around approvals and user input.
• Detects enter_plan_mode and exit_plan_mode tool calls, emits turn.proposed.completed events when Grok writes plan.md, and intercepts x.ai/exit_plan_mode to capture plan content without native gating.
• Adds Grok Build usage support across the stack: bumps usage contract to v5 with backward-compatible merge for v4, parses updates.jsonl transcripts via parseGrokLine in usageTranscripts.ts, and adds Grok series to web and mobile usage charts.
• Propagates per-turn reasoning effort to setSessionModel metadata in GrokAcpSupport.ts, validates tokens, and skips redundant calls.
• Discovers Grok skills at runtime via grok inspect --json in GrokSkills.ts with a 4 s timeout and soft-fail to empty.
• Improves tool-call update coalescing in AcpRuntimeModel.ts to emit on status changes and track combined output growth (detail/content/rawOutput) instead of detail length alone.
• Risk: USAGE_CONTRACT_VERSION bumps from 4 to 5 in usage.ts; older clients below USAGE_MERGE_COMPATIBLE_SINCE (4) will be marked stale. selectGrokPermissionOptionId falls back to allow_once when acceptForSession is missing, changing approval behavior for sessions where Grok omits that option.

^{Macroscope summarized 01cc3ac.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: c29d928a-c7b2-41ac-8687-bc7e3fca5c02

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.3 KiB	+20 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+2 B (+0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.4 KiB	+18 B (+0.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.6 KiB	+44 B (+0.1%)	66.4 KiB	✅
Codex	Live turn messages	10	11	+1 (+10.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−5 B (−0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−5 B (−0.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.4 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	11	10	−1 (−9.1%)	21	✅

Baseline: d3c24a1 · PR result: 01cc3ac · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[t3dotgg]

Grok threads miss skills, reasoning controls, proposed plans, and usage data. Some turns stay on Working after ACP stops responding.

Continues #7638 by @1xpixi. The original commits and authorship are preserved, including work from @SunkenInTime, @ahmed-besic, @mbrow190, @lnieuwenhuis, and @gbarros-dev. This PR retains the original Grok changes and adds fixes found during review:

• Keep "Always allow this session" scoped to the matching operation. A different command still asks. A changed shell-command description does not invalidate the approval.
• Use the provider instance's environment when finding Grok plan files.
• Fix the shared ACP parser's unchecked array access so the server typecheck passes.
• Add focused tests and an opt-in real-Grok turn test.

Verification

• 194 focused tests passed, including Grok, shared ACP, Cursor, and usage handling.
• Five real CLI tests passed against Grok 1.0.5, covering authentication, model discovery, reasoning selection, and a streamed model response.
• Real Grok adapter checks passed in a private test home: approve a command for the session, repeat it without another prompt, request approval for a different command, deny it, and capture a proposed plan without implementing it.
• Parsed token and cost records written by the real Grok CLI.
• Server and web typechecks passed. Changed-file lint and formatting passed.
• Mobile typecheck reports 56 navigation errors. Clean main at d3c24a14b produces identical output. No new mobile type errors.

No live T3 state or normal Grok state was changed. Fresh browser verification and before/after captures have not been run. The screenshots below are the contributor's existing evidence from #7638, not captures of this follow-up branch.

Original UI evidence

Grok usage, from @1xpixi's Windows desktop build:

Grok reasoning control, from the same build:

Model: GPT-5.6 Sol. Harness: Codex in T3 Code.

---

Note

Medium Risk
Touches Grok turn lifecycle, permission mapping, and ACP parsing used beyond Grok; behavior changes (stall detection, rate-limit errors, session approval scoping) can affect live threads if mis-tuned.

Overview
Grok integration gains skill discovery (grok inspect --json), ACP-driven reasoning effort in the model picker and per-turn session/set_model metadata, and plan mode support that surfaces plan.md writes and _x.ai/exit_plan_mode as proposed-plan events while unblocking the native approval gate.

The Grok adapter adds a turn liveness watchdog (long silence after ACP progress fails the turn; longer window while tools run; refreshes on steer, plans, approvals, and user-input), maps usage limits from xAI rate_limit completions, and improves permissions: session-scoped auto-approval by operation identity, fallback when Grok omits allow_always, and CLI argv that mirror T3 runtime modes (--permission-mode / --always-approve).

Shared ACP changes emit in-progress tool updates, bound oversized tool output (including whitespace-only buffers), and coalesce streaming command output using content length—not just detail. The mobile usage chart adds Grok as a provider band with theme-aware colors. The ACP mock agent grows scenario flags to exercise these paths in tests.

^{Reviewed by Cursor Bugbot for commit 01cc3ac. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add Grok turn watchdog, plan mode, usage tracking, and reasoning effort

• Adds an inactivity watchdog to GrokAdapter.ts that cancels and settles stalled ACP turns (10 min default, 30 min with active tools), with liveness tracking around approvals and user input.
• Detects enter_plan_mode and exit_plan_mode tool calls, emits turn.proposed.completed events when Grok writes plan.md, and intercepts x.ai/exit_plan_mode to capture plan content without native gating.
• Adds Grok Build usage support across the stack: bumps usage contract to v5 with backward-compatible merge for v4, parses updates.jsonl transcripts via parseGrokLine in usageTranscripts.ts, and adds Grok series to web and mobile usage charts.
• Propagates per-turn reasoning effort to setSessionModel metadata in GrokAcpSupport.ts, validates tokens, and skips redundant calls.
• Discovers Grok skills at runtime via grok inspect --json in GrokSkills.ts with a 4 s timeout and soft-fail to empty.
• Improves tool-call update coalescing in AcpRuntimeModel.ts to emit on status changes and track combined output growth (detail/content/rawOutput) instead of detail length alone.
• Risk: USAGE_CONTRACT_VERSION bumps from 4 to 5 in usage.ts; older clients below USAGE_MERGE_COMPATIBLE_SINCE (4) will be marked stale. selectGrokPermissionOptionId falls back to allow_once when acceptForSession is missing, changing approval behavior for sessions where Grok omits that option.

^{Macroscope summarized 01cc3ac.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: c29d928a-c7b2-41ac-8687-bc7e3fca5c02

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.3 KiB	+20 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+2 B (+0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.4 KiB	+18 B (+0.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.6 KiB	+44 B (+0.1%)	66.4 KiB	✅
Codex	Live turn messages	10	11	+1 (+10.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−5 B (−0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−5 B (−0.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.4 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	11	10	−1 (−9.1%)	21	✅

Baseline: d3c24a1 · PR result: 01cc3ac · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[t3dotgg]

Grok threads miss skills, reasoning controls, proposed plans, and usage data. Some turns stay on Working after ACP stops responding.

Continues #7638 by @1xpixi. The original commits and authorship are preserved, including work from @SunkenInTime, @ahmed-besic, @mbrow190, @lnieuwenhuis, and @gbarros-dev. This PR retains the original Grok changes and adds fixes found during review:

• Keep "Always allow this session" scoped to the matching operation. A different command still asks. A changed shell-command description does not invalidate the approval.
• Use the provider instance's environment when finding Grok plan files.
• Fix the shared ACP parser's unchecked array access so the server typecheck passes.
• Add focused tests and an opt-in real-Grok turn test.

Verification

• 194 focused tests passed, including Grok, shared ACP, Cursor, and usage handling.
• Five real CLI tests passed against Grok 1.0.5, covering authentication, model discovery, reasoning selection, and a streamed model response.
• Real Grok adapter checks passed in a private test home: approve a command for the session, repeat it without another prompt, request approval for a different command, deny it, and capture a proposed plan without implementing it.
• Parsed token and cost records written by the real Grok CLI.
• Server and web typechecks passed. Changed-file lint and formatting passed.
• Mobile typecheck reports 56 navigation errors. Clean main at d3c24a14b produces identical output. No new mobile type errors.

No live T3 state or normal Grok state was changed. Fresh browser verification and before/after captures have not been run. The screenshots below are the contributor's existing evidence from #7638, not captures of this follow-up branch.

Original UI evidence

Grok usage, from @1xpixi's Windows desktop build:

Grok reasoning control, from the same build:

Model: GPT-5.6 Sol. Harness: Codex in T3 Code.

---

Note

Medium Risk
Touches Grok turn lifecycle, permission mapping, and ACP parsing used beyond Grok; behavior changes (stall detection, rate-limit errors, session approval scoping) can affect live threads if mis-tuned.

Overview
Grok integration gains skill discovery (grok inspect --json), ACP-driven reasoning effort in the model picker and per-turn session/set_model metadata, and plan mode support that surfaces plan.md writes and _x.ai/exit_plan_mode as proposed-plan events while unblocking the native approval gate.

The Grok adapter adds a turn liveness watchdog (long silence after ACP progress fails the turn; longer window while tools run; refreshes on steer, plans, approvals, and user-input), maps usage limits from xAI rate_limit completions, and improves permissions: session-scoped auto-approval by operation identity, fallback when Grok omits allow_always, and CLI argv that mirror T3 runtime modes (--permission-mode / --always-approve).

Shared ACP changes emit in-progress tool updates, bound oversized tool output (including whitespace-only buffers), and coalesce streaming command output using content length—not just detail. The mobile usage chart adds Grok as a provider band with theme-aware colors. The ACP mock agent grows scenario flags to exercise these paths in tests.

^{Reviewed by Cursor Bugbot for commit 01cc3ac. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add Grok turn watchdog, plan mode, usage tracking, and reasoning effort

• Adds an inactivity watchdog to GrokAdapter.ts that cancels and settles stalled ACP turns (10 min default, 30 min with active tools), with liveness tracking around approvals and user input.
• Detects enter_plan_mode and exit_plan_mode tool calls, emits turn.proposed.completed events when Grok writes plan.md, and intercepts x.ai/exit_plan_mode to capture plan content without native gating.
• Adds Grok Build usage support across the stack: bumps usage contract to v5 with backward-compatible merge for v4, parses updates.jsonl transcripts via parseGrokLine in usageTranscripts.ts, and adds Grok series to web and mobile usage charts.
• Propagates per-turn reasoning effort to setSessionModel metadata in GrokAcpSupport.ts, validates tokens, and skips redundant calls.
• Discovers Grok skills at runtime via grok inspect --json in GrokSkills.ts with a 4 s timeout and soft-fail to empty.
• Improves tool-call update coalescing in AcpRuntimeModel.ts to emit on status changes and track combined output growth (detail/content/rawOutput) instead of detail length alone.
• Risk: USAGE_CONTRACT_VERSION bumps from 4 to 5 in usage.ts; older clients below USAGE_MERGE_COMPATIBLE_SINCE (4) will be marked stale. selectGrokPermissionOptionId falls back to allow_once when acceptForSession is missing, changing approval behavior for sessions where Grok omits that option.

^{Macroscope summarized 01cc3ac.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: c29d928a-c7b2-41ac-8687-bc7e3fca5c02

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.3 KiB	+20 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+2 B (+0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.4 KiB	+18 B (+0.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.6 KiB	+44 B (+0.1%)	66.4 KiB	✅
Codex	Live turn messages	10	11	+1 (+10.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−5 B (−0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−5 B (−0.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.4 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	11	10	−1 (−9.1%)	21	✅

Baseline: d3c24a1 · PR result: 01cc3ac · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.