| name | ci-failure-fix |
| description | Fix failing tests or CI by reading logs, identifying the smallest failing surface, making a minimal change, and rerunning targeted verification. Use when given CI logs, failing commands, or broken checks. |
CI Failure Fix
Use this skill when the user asks to fix failing tests, lint, typecheck, build,
or CI.
Workflow
- Capture the exact failing command, job, or log snippet.
- Identify whether the failure is deterministic, flaky, environment-specific,
or external-provider-specific.
- Reproduce locally when feasible with the narrowest command.
- Make the smallest fix that targets the failure.
- Rerun the failing command.
- Broaden verification if the fix touches shared code.
- Report the failure, fix, command, and residual CI-only risks.
Guardrails
- Do not rewrite code broadly to satisfy a narrow failure.
- Do not weaken tests, type checks, lint rules, or CI gates without approval.
- Do not treat external providers as local proof if logs are unavailable.
- Do not chase unrelated warnings unless they block the requested check.
Output
CI Fix
- failing check:
- root cause:
- change:
- rerun:
- remaining risk:
