AGENTS.md 驱动 Codex 代码审查规则

codex code review now reads 𝗰𝘂𝘀𝘁𝗼𝗺 𝗿𝘂𝗹𝗲𝘀 straight from AGENTS.md. when a rule fires, the review comment cites the one you wrote. so if your file is a mess, that's a problem now. also a decent excuse to clean it up. reminded me of the CLAUDE.md file that topped github trending (karpathy's four coding principles, 𝟭𝟵𝟱𝗸 𝘀𝘁𝗮𝗿𝘀 now). it teaches your agent how to write code. flip it around and the same four rules review every PR: → 𝘁𝗵𝗶𝗻𝗸 𝗯𝗲𝗳𝗼𝗿𝗲 𝗰𝗼𝗱𝗶𝗻𝗴: the ticket reads more than one way and the diff silently picks one, flag it → 𝘀𝗶𝗺𝗽𝗹𝗶𝗰𝗶𝘁𝘆 𝗳𝗶𝗿𝘀𝘁: single-use abstractions, speculative features, error handling for cases that can't happen, flag it → 𝘀𝘂𝗿𝗴𝗶𝗰𝗮𝗹 𝗲𝗱𝗶𝘁𝘀: changed lines that don't serve the PR's goal, drive-by refactors next door, flag it → 𝗴𝗼𝗮𝗹-𝗱𝗿𝗶𝘃𝗲𝗻: a bugfix with no test that reproduces the bug, flag it paste the whole block into root AGENTS.md: ## Code Review Rules ### Think before coding if the linked issue allows multiple readings, the PR states which one it picked and why. flag silent picks. ### Simplicity first flag abstractions with a single call site, features nobody asked for, and error handling for scenarios that cannot happen. ### Surgical edits every changed line traces to the PR's stated goal. flag drive-by refactors and formatting churn in untouched logic. ### Goal-driven a bugfix ships with a test that reproduces the bug. flag fixes that ship without one, unless the PR says why a test is impractical. repo below 👇

Codex Code Review can now use custom repository rules in AGENTS.md. Start with a check your reviewers keep repeating. Keep it concise and scoped so Codex can catch repository-specific issues even when the relevant context is siloed. https://t.co/dMF1APvYYR