What problem does it solve? Single-pass manuscript reviews tend to be overly charitable, missing weak physical interpretation, thin novelty, or unsupported significance claims. This Skill replaces that with a staged, evidence-aware referee panel that fails closed on unsupported scientific claims. ## Core Features & Use Cases - Six-stage review panel: Runs a full-manuscript reader, literature reviewer, mathematical-soundness reviewer, physical-soundness reviewer, significance reviewer, and a final adjudicating referee, each in a fresh context with compact JSON stage artifacts. - Strict preflight and evidence contracts: Validates project state, manuscript presence, artifact manifests, bibliography audits, and reproducibility manifests before reviewing, and blocks on missing review-grade evidence. - Revision round handling: Detects prior referee reports and author responses to run R2/R3 rounds with suffixed artifacts, then routes the outcome to respond-to-referees, manuscript edits, or arxiv submission. - Use Case: After drafting a physics paper in a GPD project, run the review to get a claim index, per-stage concern reports, a validated referee decision, and a LaTeX/Markdown referee report with an accept/revise/reject recommendation. ## Quick Start Ask the agent to peer review the manuscript in the paper directory of the current GPD project.
