Claude CodeIntermediateClaude Code
A 4-dimension checklist before merging AI code
Generated code that 'looks fine' and passes the tests can still leak data: an explicit checklist turns good intentions into a repeatable gate.
Define with your team one check per dimension: correctness (tests and edge cases), security (no secrets, validated inputs, least privilege), maintainability (conventions, no unexplained complexity) and human understanding (whoever merges can explain what it does and why). Anything that can be automated (regression tests, evals) goes into CI.
Example
## AI-code PR checklist - [ ] Correctness: tests pass and cover edge cases - [ ] Security: no secrets; inputs validated; minimum permissions - [ ] Maintainability: follows the team's conventions - [ ] Understanding: the author explains what it does and why, even with untested inputs