| name | Fix CI Failure |
| description | Investigates and fixes a CI check failure on a pull request |
| user-invocable | true |
| allowed-tools | bash,read,edit,write,grep,glob |
You are fixing a CI check failure on a pull request. The failing check
details are provided below.
Instructions
- Examine the failing check output to understand what went wrong
- If the failure is in a test:
- Read the test file and the code it tests
- Determine if the test needs updating or the code has a bug
- Fix accordingly
- If the failure is a lint/build error:
- Read the error output
- Fix the source file
- Run the failing check locally to verify the fix before pushing.
This is critical -- do not push until you have confidence the fix works.
Run the relevant test suite, linter, or build command locally.
- Commit with a message like "fix: (CI)"
- Only push after local verification passes
Important
- Focus only on the failing check -- don't fix unrelated issues
- If the failure seems like a flaky test or infrastructure issue, report it
rather than making code changes
- If you can't determine the cause, report what you found
- Do not push a fix unless you've verified it locally. A bad push
triggers another CI run and potentially another fix attempt, creating
a feedback loop.
Concurrency Warning
Another agent may be working in this same worktree concurrently (e.g.,
addressing review comments while you fix CI). Before committing:
- Run
git status to check for unexpected changes
- If there are uncommitted changes you didn't make, do a
git stash before
your work and git stash pop after, or coordinate via sequential commits
- Pull before pushing to avoid conflicts