codex-pr-criticlisted

## When to Use This skill is invoked for each PR in the batch selected by `codex-pr-plan`. It deep-reads the PR and extracts structured quality signals. ## Non-Negotiable Rules | # | Rule | Violation Example | Correct Pattern | |---|------|-------------------|-----------------| | 1 | Read the full PR body, not just title | Scoring based on title alone | `gh pr view -R openai/codex {number} --json body,title,reviews,comments` | | 2 | Read ALL review comments | Only reading the first review | Fetch reviews AND review comments (inline) | | 3 | Distinguish review types | Treating all comments as equal | Separate: approval, changes_requested, comment-only, inline nit | | 4 | Note the outcome | Ignoring whether PR was merged or abandoned | merged_at != null means merged; otherwise rejected/abandoned | | 5 | Don't editorialize prematurely | Writing "this is a bad PR" | Record signals objectively; patterns emerge during distillation | ## Data Collection Per PR For each PR, collect these fields: ```bash # PR metadata gh pr view -R openai/codex {number} --json number,title,body,state,mergedAt,closedAt,author,additions,deletions,changedFiles,labels,createdAt,baseRefName,headRefName # Reviews (approval/rejection decisions) gh api "repos/openai/codex/pulls/{number}/reviews" --jq '.[] | {user: .user.login, state: .state, body: (.body[:300])}' # Review comments (inline code comments) gh api "repos/openai/codex/pulls/{number}/comments?per_page=100" --jq '.[] | {user: .user.login, bod