What problem does it solve? Reviewing a large or complex pull request requires hours of up-front investigation before you can even start forming opinions. This Skill does that investigation for you by dispatching parallel subagents across distinct review concerns, then synthesizing their findings into a concise review guide — so you walk into the PR already oriented, without anyone pre-writing your verdicts. ## Core Features & Use Cases - Primer agent: A single agent first writes a factual primer covering the PR's purpose, file-by-file map, key concepts, test inventory, and existing review comments. - Four parallel investigation agents: Style & CLAUDE.md adherence, correctness (with git history analysis), tests & coverage gaps, and design & code quality informed by a mined reference of your real past review comments. - Guide, not verdicts: The output is a structured review guide — hotspots, open questions, cross-cutting observations, and a "verified fine" section — never a ready-to-post review, and nothing is ever posted to GitHub. - Use Case: You are assigned a 15-file PR touching LLM prompt logic. Invoke the Skill, get a 2-minute orientation, 3-7 hotspots with path:line citations, and a dossier for drill-down, then write your own 5-10 inline comments in a fraction of the usual time. ## Quick Start Ask the assistant to run the multi-agent PR review on PR number 123 and produce a review guide for you to read before reviewing it yourself.
