What problem does it solve? Codex review rounds on pull requests often loop: you fix the one flagged line, and Codex flags the same mistake elsewhere on the next pass. This Skill generalizes each Codex comment into an error pattern, sweeps the entire branch for other instances, fixes and tests them all, and drafts a reply so each review round actually converges. ## Core Features & Use Cases - Comment scoping and fetching: Locates the PR from a URL or the current branch, then collects open Codex comments across inline review comments, review summaries, and conversation comments using the GitHub CLI, identifying Codex by author login rather than bot type. - Pattern generalization and branch-wide audit: Converts each flagged line into a named error pattern (e.g., swallowed exceptions, hardcoded paths), diffs against the PR base ref, and searches the whole branch with grep/rg to fix every instance while guarding against false positives. - Verified, gated replies: Runs the project's tests and linters, drafts a per-comment reply summarizing fixes and audit tallies, gates the @codex review re-trigger on evidence of Codex access, and posts only after user confirmation. - Use Case: Codex flags a bare except: pass at one line of your PR. The skill fixes it, finds three more swallowed exceptions across the branch, runs pytest, and drafts a reply documenting the full audit for your approval. ## Quick Start Ask the assistant to address the Codex review comments on the current pull request and audit the whole branch for the same error patterns.
