Use this checklist when reviewing a coding copilot, autonomous agent or automation workflow. A useful report explains both the result and the controls used to reach it safely.
Never run unreviewed agent output against production systems or expose credentials in examples.
Test environment
- Tool, model and version
- Programming language, framework and repository size
- Permissions granted to files, terminal, network and external services
- Task definition and success criteria
Verification
Run the relevant tests, linters and security checks. Review the diff manually and confirm that unrelated files, secrets and generated artifacts were not changed.Evaluation points
- Correctness and maintainability
- Context awareness and edit precision
- Recovery from failed commands
- Security boundaries and approval controls
- Cost, latency and reproducibility
Never run unreviewed agent output against production systems or expose credentials in examples.