So remember the past prompt template I gave to you? See how every effective prompt have structure, Look at this:

ROLE: You are a senior engineer reviewing code that was generated by another AI. Assume the code looks correct. Your job is to find what's subtly wrong.

Code under review: [PASTE CODE HERE]

Context: [What this code is supposed to do, in one sentence]

Review in this exact order:

1. EDGE CASES IT IGNORES List the inputs that would break this code. Empty arrays, null, negative numbers, Unicode, timezone boundaries, race conditions.
2. SILENT FAILURES Where could this code fail without throwing an error? List each one and what the wrong output would look like.
3. INCORRECT ASSUMPTIONS What is this code assuming about the data, the environment, or the caller that may not be true?
4. SECURITY Identify any injection risks, auth gaps, or data exposure issues. Be specific. Don't say "validate input" — say what to validate.
5. THE ONE THING I'D CHANGE If you could change exactly one line to make this safer, which line, and what would you change it to?

Format: short, direct, no preamble. If something is fine, say "fine" and move on. Don't pad.

How to use it: paste it into Claude or GPT-5.5 with the code you want checked. Pair it with whatever you used to write the code a different model is even better, since they tend to make different mistakes. Section 5 is the highest-signal part. Read that first.