Objective
Route actionable Every Code PR failures back into the Every Code session while avoiding noisy transient infrastructure failures.
Finish Line
Actionable PR failures reach Every Code, transient preview flakes retry quietly
Current Status
State: Done. PR #398 merged and deployed; Every Code worker now routes app/build/test PR check failures back into the existing PR-feedback session path, retries preview infrastructure failures once without notifying Every Code, suppresses duplicate failure feedback, and marks pending synthetic failures ignored after recovery.
Next action: None for this slice.
Blocked by: None.
Last verified: 2026-05-07 with PR #398 CI/Security/CodeQL green and Deploy Launchplane run 25507960125 successful.
Scope
- Monitor preview/check failures for Every Code PRs.
- Retry known transient preview infra failures once when safe.
- Comment on the PR with compact failure details for persistent/actionable failures.
- Let existing PR feedback ingestion deliver those comments to live or resumed Every Code sessions.
Acceptance Criteria
- A transient preview infra failure can be retried once without notifying Every Code prematurely.
- Persistent preview infra failure is visible to operators but not misrouted as app-code work.
- App/build/test failures are commented on the PR with enough detail for Every Code to act.
- Duplicate comments are avoided for the same failing run/job/SHA.
- Failed state clears or updates after a later green run.
- Tests cover transient retry, persistent infra failure, app failure routing, duplicate suppression, and recovery.
Relationships
Sub-issue of #373.
Uses the PR feedback channel shipped for Every Code sessions.
Validation
Unit tests around failure classification and a live smoke with a controlled failing check or mocked workflow fixture.
Decisions
Do not route one-off infra/auth flakes to Every Code unless they persist after retry.
Open Questions
- Which check names should be treated as infra-only versus application-actionable by default?
- Should failed preview image publishing remove or keep the
preview label while recovering?
Objective
Route actionable Every Code PR failures back into the Every Code session while avoiding noisy transient infrastructure failures.
Finish Line
Actionable PR failures reach Every Code, transient preview flakes retry quietly
Current Status
State: Done. PR #398 merged and deployed; Every Code worker now routes app/build/test PR check failures back into the existing PR-feedback session path, retries preview infrastructure failures once without notifying Every Code, suppresses duplicate failure feedback, and marks pending synthetic failures ignored after recovery.
Next action: None for this slice.
Blocked by: None.
Last verified: 2026-05-07 with PR #398 CI/Security/CodeQL green and Deploy Launchplane run 25507960125 successful.
Scope
Acceptance Criteria
Relationships
Sub-issue of #373.
Uses the PR feedback channel shipped for Every Code sessions.
Validation
Unit tests around failure classification and a live smoke with a controlled failing check or mocked workflow fixture.
Decisions
Do not route one-off infra/auth flakes to Every Code unless they persist after retry.
Open Questions
previewlabel while recovering?