What problem does it solve? Review bots like Codex and CodeRabbit leave ambiguous signals on GitHub pull requests—reactions, empty review arrays, and comments—that are easy to misread as a clean verdict. This Skill teaches how to correctly read one PR review round so an unreviewed commit is never mistaken for an approved one. ## Core Features & Use Cases - Signal interpretation: Distinguishes the three clean-round signals (review submission, no-finding comment, thumbs-up reaction) and explains why an empty reviews array does not mean "not reviewed". - Commit-bound verdicts: Only counts a verdict when the bot-carried SHA (commit_id field or "Reviewed commit:" in the body) matches the current head SHA, never relying on timestamps or reactions. - Skip and nudge rules: Detects why a bot did not attach (bot-authored PR, fewer than 10 stars), when to wait versus escalate after 5 minutes of silence, and why only a human account may request a re-review. - Use Case: After pushing a fix to a stacked PR, an agent checks whether Codex actually reviewed the new head commit before reporting the round as clean to the author. ## Quick Start Read this Skill before opening a PR, marking it ready, or interpreting Codex or CodeRabbit review results on any pull request in this repository.