Skip to content
AI & Agents
Skill

/spec-peer-review

Run one requested external review of an approved spec, design doc, RFC, or detailed PRD; produce findings for user-selected incorporation. Excludes implementation reviews and automatic or looping approval gates.

BOOST
From plugin
pedronauck-skills
63787 skills1 command
Install
$ npx -y skills add pedronauck/skills --skill spec-peer-review --agent claude-code

How it fires

How this skill gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.
  • Slash command/spec-peer-review

Context preview

The summary Claude sees to decide when to auto-load this skill.

Run one requested external review of an approved spec, design doc, RFC, or detailed PRD; produce findings for user-selected incorporation. Excludes implementation reviews and automatic or looping approval gates.

SKILL.md

spec-peer-review.SKILL.md
name: spec-peer-review
description: "Run one requested external review of an approved spec, design doc, RFC, or detailed PRD; produce findings for user-selected incorporation. Excludes implementation reviews and automatic or looping approval gates."
disable-model-invocation: true
argument-hint: "[spec-path] [--context p1,p2] [--out dir] [--ide <ide>] [--model <model>] [--reasoning <effort>]"
metadata:
  author: Pedro Nauck
  github: https://github.com/pedronauck
  repository: https://github.com/pedronauck/skills

Spec Peer Review

Review a saved design only when the user requests an independent review. Honor existing approval to review it as-is. Incorporate only selected findings; additional rounds remain opt-in.

Inputs and output

  • Resolve the spec from the request or active context. Ask only when ambiguous.
  • Accept additional context and an output directory. Default to the spec's `qa/` directory for `.compozy/tasks/<slug>/`, otherwise `.peer-reviews/<UTC-timestamp>/`.
  • Use the configured reviewer runtime/model and reasoning. Explicit overrides win; never pin a historical model or silently replace a user-selected model.
  • Keep one findings artifact per round: `<out>/peer-review-findings-roundN.md`. Preserve prior rounds.

Review

1. Read the spec and its linked contract index. Include relevant requirements, concrete examples, test contracts, ADR decisions, and the instructions for affected surfaces. Expand into other context when a dependency or contradiction needs it; do not load every memory file. 2. Use the applicable markers in `references/quality-markers.md`; missing contracts become findings, not an automatic stop before a requested review. 3. Read `references/peer-review-prompt.md` and fill its placeholders, including exact findings path, round, resolved runtime/model, and context paths. The reviewer may write only that findings file. 4. Launch through the available independent reviewer runtime (a configured native reviewer, Compozy session, or herdr TUI). Use that runtime's documented commands; do not assume a retired `compozy exec` API. Preserve existing worker identity for follow-up rounds when available. 5. Inspect the findings, cited evidence, and changed paths. Validate the artifact:

   bash <skill-dir>/scripts/validate-findings.sh --kind techspec --round <N> --path <out>/peer-review-findings-roundN.md

6. Report the verdict, material findings, limitations, and artifact path. Missing/malformed output is an incomplete review, never `READY`. Repair an operational failure within the authorized round; do not start additional substantive rounds automatically. 7. Apply only user-selected findings and record the disposition with changed paths. Reuse authorization already given; ask only for a selection that is still missing. On an explicitly requested next round, send changed files, prior findings/disposition, and the new output path rather than the unchanged corpus again.

This skill does not authorize commits, publishing, or approving a design. A real product/security/compatibility constraint can block readiness; missing ritual sections and stylistic preferences cannot.

Read more
Ships withpedronauck-skills

A curated collection of 87 agent skills for Claude Code and compatible AI coding assistants — 40 original (⭐️), 34 hand-picked (💎), and 13 marketing & business (📣).

Get the whole plugin
Stats
637
Stars
89
Forks
Maintained
Maintenance
Python
Language
1mo ago
Last commit
7mo ago
Created
4h ago
Added

Repo: pedronauck/skills

Other skills on pedronauck-skills.