FRI, NOVEMBER 06, 2026
Independent · In‑Depth · Practitioner‑Tested
Claude Code Review

Claude Prompts for Reviewing Code You Did Not Write

Reviewing AI-generated code is now a bigger share of the job than writing it. These prompts are built for that: each one asks Claude to look for a specific class of problem and to say plainly when it finds nothing, rather than inventing a finding to look useful. Paste the diff or the file first, then the prompt.

⌨️ 6 prompts 🕐 Updated Nov 6, 2026 👁 6 views
💡 How to use these prompts: Replace everything in [BRACKETS] with your specific details before sending. Click Copy to copy any prompt to your clipboard instantly.
1
The honest first pass
Pull request review
Review this diff. For each file, list only defects you can point to a specific line for. Do not restate what the code does. Do not suggest style changes.\n\nFor each finding give: the line, what breaks, and the input or state that triggers it.\n\nIf a file has no defects, write the filename and NO FINDINGS. I would rather read NO FINDINGS five times than read five invented suggestions.
2
Error paths only
Error handling audit
Ignore the happy path entirely. Look only at what happens when things fail.\n\nFor every external call, file operation and parse in this code, tell me: what happens on failure, whether the failure is swallowed, and whether the caller can tell the difference between success and a silently handled error.\n\nFlag every catch block that does not either rethrow, log with context, or change control flow in a way the caller observes.
3
The boundary check
Input validation
For every function in this file that takes input from outside the module, list the assumptions it makes about that input which are not checked in code.\n\nFormat each as: function name, assumption, what happens if the assumption is false.\n\nTreat null, empty collection, zero, negative, very large, and wrong type as separate cases. Do not suggest adding validation yet. Just tell me what is currently assumed.
4
Concurrency read
Race condition hunt
Assume two or more of these code paths run at the same time on the same data.\n\nIdentify every piece of shared mutable state, every read-then-write sequence that is not atomic, and every place where ordering between two operations matters but is not enforced.\n\nFor each one, describe the interleaving that produces the wrong result. If you cannot describe a concrete interleaving, do not list it.
5
What the tests do not cover
Test coverage gaps
Here is the implementation and here are its tests.\n\nList the behaviours the implementation has that no test exercises. Rank them by how badly a regression there would hurt, not by how easy a test would be to write.\n\nSeparate genuinely untested behaviour from behaviour that is tested indirectly. Say which is which. Do not write the tests yet.
6
Explain it back
Understanding a change
Before I accept this change, explain to me what it does without reading the commit message or any comments. Use only the code.\n\nThen tell me which parts of your explanation you are confident about and which parts you inferred from naming rather than from logic.\n\nThe second list is what I need. Anything you had to infer from a variable name is a place the code is not saying what it means.