For the complete documentation index, see llms.txt. This page is also available as Markdown.

Recipe: Run a judge in CI without auto-merging on pass

Recipe — gate CI on a judge verdict without auto-merging on pass.

When to use: judge is reliable enough to fail on, but you want human review on pass too.

Steps

  1. CI runs the judge as a required check.

  2. Failure blocks merge (hard gate).

  3. Success is signal but doesn't auto-merge — keep human reviewer required.

Code

.github/workflows/eval.yml:

- run: python scripts/ci-eval.py
 # Exits 1 on regression; otherwise 0. PR rules require this check + a reviewer.

See also

Last updated

Was this helpful?