What Happens When a Gate Fails

Quality gate failures are normal. They're caught early so nothing broken reaches main. Claude handles the vast majority of failures automatically without involving you.

Claude's Automatic Response

When a gate fails, Claude reads the failure details from the PR comment posted by the summary gate. Claude diagnoses the issue, fixes the code, and pushes an update to the PR branch. The gates rerun automatically on the new commit.

If the fix is straightforward (a type error, a lint rule violation, a Prettier formatting issue, or a test that broke after a code change), Claude handles it without asking for your input. You'll see the PR update and the gates go green.

When Claude Asks for Your Help

Claude will surface the issue to you if:

• The failure reveals an ambiguity in the requirements, for example a security rule that conflicts with a feature you asked for

• The fix requires a product decision, not just a technical one

• The gate has failed multiple times on the same issue without a clear resolution path

In these cases, Claude explains the problem and the options, and waits for your direction before making further changes.

The No-Bad-Code Guarantee

Nothing reaches main that hasn't passed all four gates. If Claude can't fix a failure, it stops and tells you why rather than bypassing the checks or lowering the bar to get the PR through.

The gates are not optional and cannot be skipped. They exist to protect the integrity of your codebase, not to slow things down.