A reviewer should receive the expected behavior and diff, but not inherit the implementer’s untested assumptions.
The decision behind the technique
Independent review reduces self-confirmation and lets instructions focus on a narrow risk domain. The reviewer’s job is to find reproducible issues, not to praise the patch.
A reviewer should receive the expected behavior and diff, but not inherit the implementer’s untested assumptions.
A practical step-by-step workflow
Use this sequence as a default, then tighten or expand it only when the evidence changes.
- Choose one specialty and its threat model
- Forbid edits during the review
- Require severity, evidence and a tight location
- Reproduce the finding before asking for a fix
Failure boundary and finish line
A specialist reviewer does not replace the general integration checks for the final user flow.
Finish when another person can reproduce the check against the exact final state.
Frequently asked questions
What should I do first?+
Choose one specialty and its threat model
What is the main failure mode?+
A specialist reviewer does not replace the general integration checks for the final user flow.
OFFICIAL SOURCES
Verify against the source
OpenAI subagent guidanceOpenAI code review guidanceThis guide provides general information, not an OpenAI allowance promise or support decision. Your account and official pages govern actual usage, benefits and recovery times.