Correct Grok model-family policy documentation

[TheGreatAxios]

Summary

Correct ARCHITECTURE.md Grok model-family policy docs: Grok shares the five-minute sub-agent stall timeout (not 90s) and tool-only nudge threshold; applyGrokFinishBias remains leaf-only and is excluded from orchestrators. Line reference updated to model-family-policy.ts:48.

Verification

• git diff --check exited 0
• bun run check: lint, typecheck, and build passed; full suite exited 1 with 5 failures unrelated to this docs-only edit (initTemporaryGitRepo fixture timeout; createOptimizedContextStore torn-base heal; createOptimizedContextStore unpublished rewrite; assembleChatAgent omits evidence archive; assembleChatAgent getWorkdir/getCompactor). Failures observed on this worktree after bun install; not caused by the ARCHITECTURE.md change.

Fixes CL-7659

[linear-code]

CL-7659

[TheGreatAxios]

Critic · Approve (intent)

Corrects ARCHITECTURE.md so Grok shares the five-minute stall timeout and default tool-only nudge; finish bias stays leaf-only.

GitHub blocked --approve (own PR). Verdict: ready to merge.

Findings

No findings.

Notes

• Docs-only: single file docs/ARCHITECTURE.md (+1/−1); line ref :48 matches DEFAULT_POLICY.
• Code truth: GROK_POLICY.subAgentStallTimeoutMs / toolOnlyTurnNudgeAt alias defaults (5 * 60_000, 25); applyGrokFinishBias && !orchestrator at resolve. Tests cover shared stall and leaf-vs-orchestrator bias.
• Kimi/default discussion preserved; no behavior change.

[TheGreatAxios]

Review · Approve

ARCHITECTURE.md Grok stall/timeout sentence matches GROK_POLICY sharing the default five-minute sub-agent stall timeout and tool-only nudge threshold.

No findings.

Notes

GitHub blocks approve on own PRs, so this is a comment with an approve verdict.

[TheGreatAxios]

Critic · Approve

GROK_POLICY.subAgentStallTimeoutMs equals DEFAULT_POLICY (5 minutes). applyGrokFinishBias remains leaf-only. Existing model-family-policy tests already lock this.

No findings.