Full Code Review – Code review prompt for ChatGPT and Claude
Free with your account
- Free with an account
- Copy filled in or download
- Updates included
Claude Opus 5.5
This prompt reviews a code change in full. Paste what the change should do, your stack and the diff or file; you get a verdict and every real issue ranked by severity, each with a fix.
What you get
- A verdict: approve, approve with changes or request changes, with the reason
- Every real issue, most severe first, with the input or sequence that triggers it
- A fix for each finding, as code when it is short
- Findings that depend on code you did not paste, marked Needs check, plus open questions
How to use it
- 1Paste the purpose, stack and diff
Say what the change should do and how your stack stores key values such as money or dates, then paste the diff with enough code around it.
- 2Read the verdict and the top findings
Start with the Critical and High items. The count line shows how much sits below them, so you can decide what to fix now.
- 3Confirm, then fix
Look up each Needs check item in your codebase, prove the serious findings with a failing test, then apply the fixes.
Better results
- Ask for every issue with a severity rather than only the serious ones. Current models follow "only report high-severity issues" literally and drop real findings; sorting a long list yourself is quicker than finding what was left out.
- Paste the callers, models and config the change depends on, or the review marks those findings Needs check instead of settling them.
- Follow up with: "Write a failing test for finding 1." A test that fails before the fix proves the bug is real.
Questions
How do I use ChatGPT for code review?
Paste the diff or file with what the change should do and your stack, ask for every real issue with a severity, the input that triggers it and a fix, then verify each finding. The prompt above does this in one message, so you only fill in the blanks.
Can I use this prompt in GitHub Copilot?
Yes. Paste it into Copilot Chat in your editor with the diff filled in; it uses plain text and XML tags and no model-specific commands. In an agent such as Copilot or Claude Code, you can point it at the changed files instead of pasting them.
Can I use this as a code review agent prompt?
Yes, with small changes. Keep the role, task, constraints and output format as the agent's instructions, and let the agent supply the diff, for example from git diff. Keep the line that asks for every issue with a severity, so the agent does not filter findings for you.
Updated 6 Oct 2026
























