Leave an agent hunting a bug overnight against a repeatable test
Idea + starter prompt, not run by us yet
For: Developers with a bug they cannot explain and no time to dig through it by hand
The author sets Codex loose overnight, for several hours, on a bug with no clear cause. The key is a unit test that repeats the bug every time through the app's main entry point. The agent adds debug prints and tests ideas about the cause on its own.
Steps
- Write a test that fails the same way every time, starting from the app's normal entry point.
- Work on a separate branch so nothing reaches your main code.
- Paste the starter prompt and let it run while you are away.
- In the morning, read its notes and check the cause it found before keeping any change.
Starter prompt
The test [test name] fails every time. Run it with [command]. I do not know why. Work on branch [branch name] only. Add debug prints, form ideas about the cause and test them one by one against this test. Keep a running log in [notes file] of each idea and what you found. Do not push, merge or delete anything. When you stop, write a summary of the most likely cause and the evidence for it.
Original post: Hacker News · found via agentusecases.ai. Written up in our own words. We haven't run it, so test it on made-up data first.