babysit

name	babysit
description	Watch a PR's CI checks and iteratively fix failures until green. Use after opening or pushing to a PR, or when CI is failing and you want autonomous fix-push cycles.
user-invocable	true

Babysit PR Checks

Monitor CI checks on the current branch's PR and iteratively fix failures until all checks pass.

Constraints

• Max 3 fix-push iterations. After 3 attempts, stop and report what's still failing. Ask the user how to proceed.
• Stop early if: the fix is large/risky, you can't diagnose the failure, or the failure is unrelated to changes on this branch (e.g. flaky infrastructure).
• Push freely within the iteration limit — no need to ask before each push.

Workflow

1. Confirm there's a PR

gh pr view --json url,statusCheckRollup -q .url

If no PR exists on the current branch, tell the user and exit.

2. Wait for checks to complete

Poll check status until all checks have completed (no pending/in_progress):

gh pr checks --watch --fail-fast

If all checks pass, report success and exit.

3. On failure — diagnose

Identify which checks failed:

gh pr checks

For each failed check, get the logs:

gh run view <run-id> --log-failed

Read the failure output carefully. Determine root cause before attempting a fix.

4. Fix and push

• Make the fix locally.
• Run any fast local validation that's relevant (typecheck, lint, unit tests) to confirm the fix before pushing.
• Commit and push:

git add <files>
git commit -m "fix CI: <brief description>"
git push

5. Repeat

Go back to step 2. Track iteration count. After 3 fix-push cycles, stop regardless of check status.

Reporting

After each iteration, briefly state:

• What failed
• What you fixed
• Current iteration count (e.g. "attempt 2/3")

When to bail out

Stop and ask the user if:

• The failure looks unrelated to the PR's changes (infra flake, unrelated test)
• The required fix is large or architecturally significant
• You're unsure about the correct fix
• The same check keeps failing after your fix (you're going in circles)