kaeawc/auto-mobile/ci-failure-triage

CI Failure Triage

Use this for a failed pull_request.yml run or a red roll-up gate. It is read-only unless the user also asks to fix a confirmed cause.

1. Run bash scripts/ci/classify-failure.sh <run-id> first. It lists each

failed/cancelled job, failed step, check-run annotations, and a matching scripts/ci/known-flakes.txt verdict.

1. The green-main ruleset requires 22 status checks; it is the authoritative

list (gh api repos/kaeawc/auto-mobile/rulesets, then rulesets/<id> and its required_status_checks rule). The roll-up gates (IDE Plugin, iOS Build, Shell Tests) only give a required context a stable name over matrix or conditionally skipped jobs; the other required checks are ordinary jobs that report directly, so read the failing job itself. Advisory means not in the ruleset (simulator, emulator, device-capture lanes): it never blocks a merge. The Linux unit run and its 100 ms budget step live in the required Node TypeScript Build and Test (ubuntu-latest) job (#10893), so a budget breach there blocks; classify it before rerunning (a loaded-runner overage is RERUN-DONT-FIX). Runs from before the non-required iOS, Android, Node Tests, and WebRTC roll-ups were removed still classify those contexts as CHECK-UPSTREAM-FIRST when only an advisory lane failed.

1. For CHECK-UPSTREAM-FIRST, inspect the named upstream row before retrying or

filing work. A red aggregator is not a test diagnosis.

1. When the verdict is INVESTIGATE with a named test file:line or missing

import path, act on that diagnosis. Treat a shard as signal-killed only when the output says killed by signal.

1. Do not re-fix KNOWN-NONFIX signatures. In particular, a Dependabot

sharp/jimp bump that breaks the pinned clean-room graph is designed to be rejected; close or ignore that Dependabot PR instead of changing runtime dependencies.

1. For integration-test or runtime-graph input changes, run

bash scripts/prepush-integration.sh before pushing. It runs only changed integration files and only runs the clean-room runtime audit when its inputs changed.

Use check-ci when the failure needs full exact-head PR-state and log analysis.