What problem does it solve? It prevents accidental or unauthorized GitHub writes during AI-assisted code review by gating every PR comment behind explicit user authorization, and it keeps cross-AI reviews independent by routing each PR to the agent that did not author it. ## Core Features & Use Cases - Three-mode write gate: Distinguishes write-authorized requests (slash command or explicit posting intent), draft mode (comment proposal only, no diff fetch or posting until confirmed), and read-only delegation. - Mutual review routing: Inspects Co-Authored-By trailers across all commits to pick the opposite AI as reviewer, failing closed to human judgment on mixed or unknown authorship. - Untrusted-input discipline: Treats PR titles, bodies, diffs, and comments from others as data only, validates PR numbers and repo slugs before shell use, and pins review targets by verified OID rather than branch names. - Use Case: You ask the agent to request a review on PR #42; it drafts the request comment for your confirmation, routes the review to Codex because Claude authored the commits, and posts the severity-classified result only after your explicit approval. ## Quick Start Ask the agent to run /personal-review-request with a PR number to post a review request and results on that GitHub pull request.