pb-ci-fix
Fix red CI or set up GitHub Actions end to end: read the failing run, find the root cause, patch and validate the workflow, push only after GO, and watch the run go green on the pushed commit. Use for
SKILL.mdskills/playbooks/pb-ci-fix/SKILL.md
GitHubFix red CI or set up GitHub Actions
What you get: a green GitHub Actions run on a pushed commit, with the cause and the fix written down.
Inputs
- repo — the current directory, or the one the user names
- failing_run_id (optional) — from the GitHub Actions page; otherwise the latest failed run
Steps
- github — repo →
gh run list --limit 5(gh missing, not authenticated, or no GitHub remote → log it, give thegh auth login/ remote hint, stop) andgh run view <id> --log-failedexcerpt in the run log — failing job and step named; no workflows exist → go to step 4 (setup path; skips steps 2, 3 and 6, and step 7 runs the new workflow's commands locally) - deja-memory — error text →
deja fixresult in the run log — prior fix found, or "none" - systematic-debugging — failing step → root cause and a local repro command — repro fails locally, or the reason it cannot run locally (a config-only fault: the validator or step commands are the repro for step 7)
- ci-pipeline (no workflows) or github-actions-generator (patch) — root cause →
.github/workflows/*.ymldiff — diff touches only the cause - github-actions-validator — workflows → validator report — zero errors after the mandatory rerun
- bdbresilience — only if step 3 classified the failure as transient → retry or timeout on that step only — no blanket retries
- verification-before-completion — repro command → fresh passing output pasted in the run log — output pasted, not claimed
- git commit
ci: <cause>— stage only the changed.github/workflows/*files (never the run log) → SHA in the run log — commit succeeds (any hook failure is reported, never bypassed with --no-verify) - [GO] git push — SHA → remote branch (name the branch in the WAITING FOR GO line) — The run stops here until the human types GO. Each retry cycle that pushes again needs its own fresh GO here.
- github — SHA →
<id>fromgh run list --commit <sha> --limit 1 --json databaseId(wait until it exists), thengh run watch <id> --exit-status—gh run list --commit <sha> --json conclusion -q '.[0].conclusion'equalssuccess; otherwise back to step 3, at most 2 cycles (each push behind a fresh GO at step 9), then stop and escalate
Run log: production_artifacts/pb-ci-fix-<date>.md
Rules
- Anything other than the literal GO (case-insensitive) is not a GO; a GO covers only that one step, one time.
- GO is by contract in every harness. The go-gate hook is only a backstop on Claude Code, OpenCode and agy (Codex: unverified; Cursor, Kimi: none). A missing hook is never permission to proceed.
- A failed check stops the run: write the failure into the run log and report. No silent retries beyond what a step names.
- Write one run-log line per step as it completes (
N. done|skipped|failed — artifact — check result) andWAITING FOR GO: <step>at each gate.