OpenAI Introduces AGENTS.md for Codex Code Review, Boosting Recall from 58% to 98%

OpenAI has added support for custom repository rules via AGENTS.md in Codex Code Review, allowing teams to encode reviewer knowledge explicitly. In evaluations, rule-based reviews achieved 98% recall on target findings, up from 58.3% baseline, addressing the key gap of missing context in AI code review. OpenAI recommends starting with checks that reviewers repeatedly explain, keeping rules concise and scoped.