Give AI-generated code a proper review
Use a structured review to find real failures, sharpen the implementation, and document what you verified.
Make it yours.
A little context makes a better starting point.
Kept on your device. Prompts run in the AI tool you choose.
Define the review boundary
Give the reviewer the expected behavior and a precise diff. Context makes the review more useful than a generic request to find bugs.
Review this change: A new workflow detail page with prompt personalization and progress tracking. My main concern is The copied prompt must match the edited inputs and saved progress must survive a refresh. First ask for or inspect the relevant diff, repository instructions, and expected behavior. Summarize the user-visible flow and identify the data boundaries. Focus on correctness, regressions, security boundaries, and accessibility. Do not invent requirements or report style preferences as defects.
Before you move on
The complete guide.
Read every step, prompt, and review check in one place. The examples use the context below; use the interactive workflow above to make them your own.
Example context
- What changed?
- A new workflow detail page with prompt personalization and progress tracking
- Main concern
- The copied prompt must match the edited inputs and saved progress must survive a refresh
1. Define the review boundary
Give the reviewer the expected behavior and a precise diff. Context makes the review more useful than a generic request to find bugs.
Example prompt Review this change: A new workflow detail page with prompt personalization and progress tracking. My main concern is The copied prompt must match the edited inputs and saved progress must survive a refresh. First ask for or inspect the relevant diff, repository instructions, and expected behavior. Summarize the user-visible flow and identify the data boundaries. Focus on correctness, regressions, security boundaries, and accessibility. Do not invent requirements or report style preferences as defects.
Review checklist
- Identify the exact change being reviewed
- State the expected behavior
- Include the affected callers or routes
2. Try to reproduce failures
A useful finding includes a trigger, the wrong result, and the reason it matters. Ask for evidence before changing the implementation.
Example prompt For the change A new workflow detail page with prompt personalization and progress tracking, investigate realistic failure cases related to The copied prompt must match the edited inputs and saved progress must survive a refresh. For each suspected issue, trace the execution path and try to reproduce it with an appropriate test or browser check. Report only actionable findings with file location, triggering input or sequence, observed behavior, expected behavior, and impact. Clearly label anything you could not verify. Avoid tests that only repeat implementation details.
Review checklist
- Reproduce each claimed issue
- Check empty and invalid inputs
- Distinguish observed bugs from open questions
3. Fix and verify
Keep fixes focused on the confirmed failures. Verify the original flow again after the change.
Example prompt Fix the confirmed findings in A new workflow detail page with prompt personalization and progress tracking, keeping the scope focused. Preserve existing project patterns and public behavior outside the identified bugs. Add a regression check when it meaningfully protects the behavior. Re-run relevant checks and the original flow, especially The copied prompt must match the edited inputs and saved progress must survive a refresh. Summarize the fixes, commands run, observed results, and any unresolved limitations.
Review checklist
- Apply focused fixes
- Repeat the failing scenario
- Record the verification results
What you’re working toward.
A focused review with reproducible findings, targeted fixes, and a clear verification record.