What problem does it solve? Deep pull request review often produces either shallow rubber-stamp approvals or floods of unvalidated findings. This Skill orchestrates a structured, graph-guided review pipeline that maximizes valid findings while filtering out false positives through independent reviewer validation and critic challenge. ## Core Features & Use Cases - Parallel Explorer Lanes: Six fixed check-type lanes (correctness, security, dependencies, docs, tests, performance) scan the full diff for candidate findings with high recall. - Independent Validation Pipeline: Candidates are parsed, chunked, and routed to reviewer subagents and critics so the orchestrator never self-confirms findings, enforcing an anti-self-review rule. - Deterministic Signal Ingestion: Existing PR comments, bot reviews, CI failures, coverage deltas, and mergeability state are ingested as obligations and candidates before exploration begins. - Use Case: When reviewing a large security-sensitive PR, run this Skill to build a context pack, dispatch parallel explorers, validate every candidate with file:line evidence, and produce a grouped report with a feedback handoff artifact. ## Quick Start Ask the AI to run a swarm PR review on a specific pull request URL or the current branch diff and wait for the validated findings report.
