feat(orchestration): opt-in bounded recovery search (#1020)#1108
Conversation
Add bounded read-only recovery attempts for compiled plans so safe state-refresh candidates can repair stale or blocked steps under explicit opt-in budgets.\n\nConstraint: Live browser recovery must be opt-in, budgeted, and limited to read-only candidates.\nRejected: General MCTS/live browser branching | unsafe for authenticated sessions and unnecessary for the current harness goal.\nConfidence: medium\nScope-risk: moderate\nDirective: Do not widen candidate execution beyond read-only tools without an explicit destructive-action gate and live verification.\nTested: npm test -- --runTestsByPath tests/orchestration/plan-executor-bounded-recovery.test.ts; npm run build\nNot-tested: Full execute_plan MCP transcript against Chrome fixture.
Qodo reviews are paused for this user.Troubleshooting steps vary by plan Learn more → On a Teams plan? Using GitHub Enterprise Server, GitLab Self-Managed, or Bitbucket Data Center? |
|
Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits. |
|
Warning You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again! |
Merge rationale (stack consolidation)Intent. Closes #1020 — adds an opt-in bounded recovery mode for compiled plan execution that consults the ranked recovery candidates added by #1088. Why this is correct.
CI. Targets the recovery-candidate-ranking feature branch; CI workflow only runs on |
1 participant
Progress / Review status
Auto-refreshed 2026-05-13 — owner comments cleaned up to reduce review noise.
feat/1020-bounded-recovery-search→feat/1018-recovery-candidate-ranking5cc0227— Recover failed plans with opt-in safe candidatesOwner comment cleanup: 0 issue + 0 inline review comments deleted. Outstanding feedback from automated/external reviewers above is unchanged.
Summary\n- Adds opt-in bounded recovery for compiled plan execution.\n- Executes only read-only candidate tools under strict candidate/tool-call/time budgets.\n- Returns structured recovery attempts in execute_plan results.\n\nCloses #1020.\n\nStacked on #1088 because it consumes ranked recovery candidates.\n\n## Validation\n- npm test -- --runTestsByPath tests/orchestration/plan-executor-bounded-recovery.test.ts\n- npm run build\n\n## Live OpenChrome verification\n- Not run in this PR session; requires execute_plan transcript against the stale-ref fixture added in #1073.\n\n## Safety\n- Disabled unless execute_plan receives boundedRecovery: true.\n- Only read_only candidates are executed; side-effect candidates are recorded as blocked.\n- Small default budgets: maxCandidates=3, maxToolCalls=2, perCandidateTimeoutMs=5000.