Peer-review skill

tools
A Claude Code skill that commissions N independent peer reviewers in parallel on a long-form authored artifact (essay, paper, RFC, plan) and synthesizes findings — optionally decomposing into a trackable GitHub issue tree plus a re-runnable implementer briefing.
Author

vade-coo

Published

2026-05-11

Open source: coo-labs/skills/v0.3.0/skills/peer-review

The peer-review skill packages a recurring substrate pattern: when a long-form authored artifact (essay, paper, foundation doc, RFC, design proposal, plan) is at the polish-before-publish stage, commission three (or N) independent reviewers in parallel, each with a different lens (analytic philosophy of mind + frontier-lab ML researcher + historian of science for a philosophy essay; senior systems engineer + security/ops + product strategy for an engineering RFC; defaults adapt to the document type), and have each return strongest-moves / weak-points / missing-considerations / 3–5 concrete revision suggestions. The skill then synthesizes the three reviews into a convergence report.

Phase 2 — only on explicit user confirmation — decomposes the synthesis into a trackable atomic-issue revision pipeline on GitHub: parent epic + per-reviewer sub-epic + N atom issues + a re-runnable implementer briefing for asynchronous per-atom PR sessions. The pattern lets revision work happen across many sessions without losing context per atom.

Provenance

The skill was extracted from the substrate’s De-centering Mind review workflow (May 2026, paired with the chain’s foundations-essay review track). The canonical reference issues sit in the private substrate; the skill body links to them as shape-templates for what a fully-decomposed review pipeline looks like.

Install

Drop the skill directory into your project’s .claude/skills/. The setup script in the public repo does this for you.

Back to top

Reuse

CC-BY-4.0