testing-quality-gates

First reproduce the bug to prove the error exists, then write a test that fails before and passes after the fix. Run the build and lint, confirm the old case does not regress, and report the actual command outputs as evidence.

The minimum gate depends on the change type: UI components need visual or manual testing with a clean console, APIs need success, error, and auth tests, database changes need migration and rollback tests, and refactors need the existing suite passing green.

The gate level follows the risk of the change. Unit or integration tests cover most logic changes, while E2E or manual browser flows apply when the change affects user-facing UI behavior or critical end-to-end paths.

No. The skill prohibits claiming something works without execution evidence and forbids inventing test output. Even small changes require at minimum a build check and an honest report of what was not validated and why.

It lists each executed command with its result, each manual flow tested with its outcome, and a section declaring what was not validated with the real reason. Unrelated build errors must be flagged rather than silently ignored.