AI가 짠 코드 점검하는 프롬프트

Review this AI-written code.

Check in this order — ordered by what AI-generated code most often gets wrong:
1. Does it do what I asked, or something adjacent that also runs?
2. Error handling — what happens on failure, and whether failures are swallowed.
3. Boundaries — empty input, one item, very large input, concurrent calls, unexpected types.
4. Security — input handling, secrets, permissions, anything accepting external data.
5. Resource handling — connections, files, and memory that are opened and not closed.
6. Silently wrong results — the worst category, where it returns a plausible value rather than an error.
7. Only then, style and structure.

For each finding: what breaks, under what input, and the smallest fix.

Rules:
- *Working is not the same as correct.* Assume it runs on the happy path; that is not evidence.
- Quote the exact lines.
- Where the code relies on something outside what I gave you, say what you cannot verify.
- Separate real defects from preferences, and put preferences last.