ci-investigator

CI investigator (core-be)

Produce a short root-cause summary for one failing check (not full PR babysit — see pr-babysit for the full loop).

Steps

1. Identify the failing job:

   gh pr checks
   gh run view <run-id> --log-failed
   

2. Map the job to local commands:

   CI job name	Workflow	Local reproduction
   PR CI (lint, security, …)	pr-ci.yml	pnpm ci:quality (local aggregate; CI splits jobs)
   Tests	reusable-vitest-postgres-redis.yml	pnpm compose:up → pnpm db:migrate → pnpm test
   API smoke (local only)	CD post-deploy or local verify	pnpm verify:base or pnpm test:api-smoke
   Chaos (Toxiproxy)	reusable-chaos-toxiproxy.yml	pnpm chaos:up → pnpm chaos:provision → pnpm test:chaos
   Docker	reusable-docker-build-trivy.yml	node tooling/ci/check-dockerfile-sync.mjs + docker build
   API Docs (post-merge publish)	reusable-openapi-postman-publish.yml	pnpm docs:all / pnpm docs:check
   API Docs (PR-time check)	pr-docs-lane.yml	pnpm docs:check

3. Reproduce the first failing step locally when possible.

4. Classify the failure:

   • Code — fix in src/, tests, or migrations
   • Drift — run generator once (pnpm routes:catalog, pnpm tool:sync-env-example --fix)
   • Flake — note evidence; suggest re-run
   • Out of scope — needs base-branch merge or infra change; do not hack CI

Output format

## CI failure: <job name>

**Root cause:** …

**Evidence:** (log line or local command output)

**Fix:** …

**Commands:** `pnpm …`

Related skills

• pr-babysit — fix and push until green
• before-commit-guard — local pre-commit mirror of quality checks
• contract-test-maintainer / chaos-test-maintainer — specialized test slices