July 2026 Summaries

Codex Code Review, a tool developed by OpenAI, uses custom repository rules in AGENTS.md to enhance the code review process by providing context-specific guidance, thereby addressing recurring issues like preserving API contracts and avoiding breaking changes. By implementing concise rules, the system helps reviewers quickly identify key concerns without becoming bottlenecks, especially as the volume of pull requests increases. At OpenAI, this approach has significantly boosted productivity, with the weekly PR volume more than doubling since Q4. The repository-rules interface allows teams to encode critical review guidance, ensuring that even new contributors or coding agents working on unfamiliar codebases can adhere to established practices. Through rigorous testing, Codex Code Review has shown effectiveness in identifying necessary rule violations while avoiding unnecessary findings, as seen with its 98% success rate in recovering custom findings during evaluations. This framework allows coding agents to focus on substantial changes without getting bogged down by routine issues, thereby facilitating smoother and faster development processes.