Prove the bug before you fix it

A review caught that when a cleanup couldn’t be confirmed, the turn still said “completed”. So the next queued command could start while the old process might still be running.

Before fixing it, the agent wrote a test that failed. Then it made the fix and the same test passed. That’s the difference between “I think it’s fixed” and “here’s proof”.

The habit

Ask for the failing test first, make it reproduce the problem, then fix the cause. If the test passes before the fix, it isn’t testing your bug.

The test stays behind afterward, so the same bug can’t quietly come back.

Try it