What problem does it solve? Pull request reviews often miss real bugs or drown reviewers in unvalidated noise. This Skill runs a structured, multi-phase review pipeline that maximizes valid findings while enforcing independent validation, so no finding is reported without file:line evidence and reviewer confirmation. ## Core Features & Use Cases - Parallel explorer lanes with risk-family coverage: Dispatches base and micro explorer lanes across six dimensions and eleven repository-agnostic risk families, scaled to diff size and risk tier. - Independent reviewer and critic validation: Explorer candidates are validated by separate reviewer subagents and challenged by adversarial critics, with the orchestrator forbidden from judging candidates itself. - Cross-harness runtime profiles: Runs on structured lane controllers (Profile A), native parallel subagents (Profile B), or single-context sequential passes (Profile C), with ledger persistence and completion gates. - Use Case: A maintainer receives a large PR touching auth and concurrency code. The Skill checks out the exact PR head, ingests existing review comments and CI signals, dispatches parallel explorers, validates candidates through reviewers and critics, and produces a grouped report with a handoff artifact for follow-up fixes. ## Quick Start Ask the agent to run a deep swarm PR review on a specific pull request URL and wait for the validated findings report.
