← ClaudeAtlas

ci-failurelisted

Diagnose a failing CI pipeline — pull the failing job's logs, find the first real error, decide whether it's a genuine regression, a flaky test, or an infrastructure problem, and propose the fix. Use when the user says CI/the build/a check/GitHub Actions is red, failing or flaky.
polytheseus/sre-skills · ★ 0 · DevOps & Infrastructure · score 72
Install: claude install-skill polytheseus/sre-skills
# CI failure diagnosis The goal is a **verdict** with evidence, followed by a fix: - **Regression**: the code change broke something. Fix the code (or the test, if the test was wrong). - **Flaky**: it fails intermittently with no relevant change. Quarantine it and file the flake with evidence. - **Infra/environment**: runner, network, registry, cache, secret, quota or dependency drift. Fix the pipeline, not the code. ## 1. Find the failing run GitHub Actions (adapt the commands for GitLab `glab`, CircleCI, Buildkite and so on): ```bash gh pr checks # on a PR branch gh run list --branch <branch> --limit 10 gh run view <run-id> # jobs and steps, with failures marked gh run view <run-id> --log-failed > /tmp/ci-failed.log wc -l /tmp/ci-failed.log ``` Logs are long, so don't read them top to bottom. Search for the **first** error. Later errors are usually cascades of the first one: ```bash grep -nE '(^|\s)(error|ERROR|Error:|FAIL|FAILED|panic:|Traceback|exit code [1-9]|Exception)' /tmp/ci-failed.log | head -40 ``` Then read about 40 lines of context around the first hit. ## 2. Classify Ask these in order: 1. **Does it fail the same way on a re-run?** `gh run rerun <run-id> --failed` (this costs CI minutes, so ask first if the pipeline is long or expensive). If it passes on re-run without any change, it's flaky or infra. 2. **Does it fail on the base branch too?** `gh run list --branch main --workflow