Most team review checklists are buried in a wiki and ignored. They're too long to remember. The fix is to turn them into a living prompt your AI uses on every PR.
Start by extracting the actual rules from your checklist. Dump the fluff. Focus on the concrete, automatable items. Then structure it for an assistant.
Example from my team's old checklist:
* Security: No secrets in code? Use env vars.
* Logging: No console.log in production code? Use structured logger.
* Errors: Are all promises caught? Are errors meaningful?
* Dependencies: Are new packages justified? Any major version bumps?
The prompt skeleton:
```
You are a senior engineer reviewing this code. Apply these mandatory rules:
- Security: Flag any hardcoded credentials, keys, or secrets. Suggest environment variables.
- Logging: Flag any `console.log` statements. Suggest the team's structured logging library.
- Error Handling: Flag uncaught promises or empty catch blocks. Ensure errors have context.
- Dependencies: For any new `import` or `require`, ask if the dependency is necessary.
Review the provided code changes and list violations. Be specific with line numbers.
```
This forces the checklist into the review cycle. The AI becomes a consistent, blunt first pass. Engineers can then focus on architecture and logic, not chasing missing semicolons.