GitHub Fix CI
Use this skill for failing GitHub Actions checks on a PR.
Run all gh commands with elevated network access.
Expert Standard
- Think like a senior engineer debugging CI under time pressure.
- Give advice like an expert: identify the real root cause, not just the loudest error.
- Execute like an expert: reproduce the smallest useful failure, make the narrowest correct fix, and validate it.
Communication Style
- Explain failures in plain English.
- Turn each broken check into a simple story: what failed, why it failed, and what fixes it.
- Keep summaries short and practical.
- Avoid dumping raw logs when one or two clear lines explain the issue.
Scope
- In scope: GitHub Actions checks reachable through
gh - Out of scope: external CI providers; report the details URL only
Workflow
- Run
gh auth status. If unauthenticated, stop and ask the user to rungh auth login. - Resolve the PR from the current branch unless the user provides one.
- Inspect failing checks with:
python "<path-to-skill>/scripts/inspect_pr_checks.py" --repo "." --pr "<number-or-url>" - Summarize each failure with:
- check name
- failed step or strongest failure line
- likely root cause
- failure type: deterministic, flaky, environmental, or external
- run URL
- Propose a short fix plan and ask for approval.
- After approval, make the smallest coherent fix and run the most relevant validation.
- Summarize the diff, residual risk, and whether the checks should be re-run.
Rules
- Collapse multiple failing checks into one remediation path when they share a root cause.
- If the failure is flaky or external, say so instead of forcing a speculative code change.
- Use manual
ghcommands only if the bundled script is insufficient. - Keep the debugging expert-grade, but make the explanation understandable to a non-specialist.