Codex Code Review is the code-review capability of OpenAI Codex. OpenAI documents turning it on for a GitHub repository, and a review can also be requested directly with @codex review. Custom Code Review rules for Codex announced that it can use custom repository rules written in AGENTS.md to catch issues specific to a codebase and to point authors to the guidance behind each finding.
OpenAI presents the feature as a response to review becoming the bottleneck once coding agents produce more code: repository rules let context that normally lives with a few experienced reviewers — an API contract to preserve, data to keep out of logs — sit next to the code it applies to, where contributors or coding agents new to that part of a repository can benefit from it.
Capabilities
Rules live in AGENTS.md, the same file that guides Codex’s coding tasks. Repository-wide rules go at the root and service-specific rules in the relevant directory; during review, Codex applies the guidance that covers the changed files and cites it in the finding. OpenAI’s example is a breaking-change rule in the Codex repository that protects an app-server notification consumed by Codex Cloud: a rename that still compiles would draw a finding explaining that consumers listen for the existing name and that the author should keep it or add a backward-compatible event.
OpenAI reports that in its primary eval suite, rule-guided variants recovered 98% of the required custom findings against 58.3% for a baseline control, and that it also tested whether clean changes avoided unnecessary findings, whether ordinary bugs outside the rules were still caught, and whether findings named the relevant guidance, location and priority.
Adoption & Ecosystem
OpenAI’s guidance is to keep deterministic and mechanical checks in tests, linters and CI and to reserve repository rules for judgment calls, starting with two or three rules for consequential, non-obvious invariants and refining them from what a representative pull request produces. It describes the tool as an additional reviewer rather than an enforcement mechanism: tests, branch protections and required approvals continue to provide hard enforcement. It sits among other AI review tools covered in this wiki under Agentic Code Review.