Codex Audit Report Is Too Broad to Act On
You asked Codex for an audit; got a 50-bullet report mixing typos with architecture. Re-scope by dimension, cap output size, force file:line.
Articles tagged with #Coding agent
You asked Codex for an audit; got a 50-bullet report mixing typos with architecture. Re-scope by dimension, cap output size, force file:line.
Codex stops mid-patch — broken imports, half-converted types, untouched call sites. Bind "done" to a verifier, not a vibe.
Codex creates `utils.v2.ts` next to `utils.ts` instead of editing in place. Force in-place edits with explicit AGENTS.md rules + git as the safety net.
Codex skips the build, misreads the output, or trusts a truncated tail. Use machine-readable verifiers + exit codes, not prose summaries.
The reported bug is gone, two adjacent features regressed. Cap blast radius, enumerate call sites, prefer caller-side guards over shared-util changes.
New files land in `/src/` when your repo uses `/app/`; deps go to the root `package.json` in a monorepo; tests use a framework you don't have. Fix with AGENTS.md + canonical-example pointers.
Codex confidently uses a column that doesn't exist, an env var with the wrong shape, or a function signature it never read. Force quote-before-write grounding.
Codex uses `getUserById` when your codebase uses `findUserById`. Codify each convention as `rule + canonical example` in AGENTS.md.
"Patch hunk failed to apply" — file moved between Codex's read and write. Refresh state, normalize whitespace, or split the patch.
"Consider error handling" / "add tests" — bullets that work on any PR. Fix with specific review questions tied to file:line.
Functionally correct but reads like a stranger wrote it — async/await mixed with .then(), wrong import ordering, comment style off. Anchor to a canonical file + enforce with lint.
"Test happy path and error path" — useless in 2026. Force tests bound to the function's actual signature, edge cases, and bug history.