What problem does it solve? Reviewing someone else's pull request requires reading repository conventions, verifying claims against actual code, and posting well-formatted feedback to GitHub without overstepping (e.g., approving on the user's behalf). This Skill structures that entire workflow so reviews are evidence-based, correctly formatted, and only posted after explicit user approval. ## Core Features & Use Cases - Convention-aware review: Reads the repository's own review rules, PR templates, and existing comment culture before judging the diff, using prefixes like [must], [iaio], and [nits]. - Adversarial self-verification: Verifies line numbers, builds, tests, and existing code patterns against the base branch, then attacks its own findings to eliminate false positives before posting. - One-shot GitHub posting: Submits the summary comment and all inline comments as a single review via the gh API, always as COMMENT (never APPROVE), with author mentions and AI-generation notes. - Use Case: A teammate asks you to review their PR. You run the skill with the PR number; it fetches the diff, reads the repo's conventions, verifies each finding, reports an approve/reject recommendation to you, and posts the review only after your GO. ## Quick Start Ask the AI to review PR 123 in the current repository and report the findings before posting anything to GitHub.
