cc-skills: Pipeline Multi-Agent para Desenvolvimento Test-First
kevin-nous/cc-skills
Pipeline completo: ideia → PRD → issues → código TDD → review → aprendizados
Ver no GitHubConjunto de 6 skills Python que automatiza o fluxo completo de desenvolvimento: da ideia vaga ao PRD, decomposição em issues, implementação test-first, revisão em 4 camadas e documentação de aprendizados. Inclui cc:proofread, um revisor multi-agente que detecta over-claims em textos analíticos antes da publicação. ~13k LOC, 496 testes, pronto para clonar e usar.
Instalação
npx convez-ai kevin-nous/cc-skills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Transformar ideias vagas em PRDs estruturados com critérios de aceitação testáveis
- Decompor requisitos em issues independentes com planejamento em waves
- Implementar código test-first com revisão arquitetural e de segurança automática
- Revisar textos analíticos e relatórios antes de publicar, capturando dados não suportados
Veredito

Ferramenta opcionada e pronta para produção que encadeia metodologias comprovadas (TDD, PRD estruturado, decomposição vertical). Ideal para equipes que querem disciplina de processo sem overhead manual; requer familiaridade com Claude Code e Python para customizar.
README
A tradução para o português deste README ainda está na fila.
cc: skills
I build my own tooling instead of doing the work by hand. This is the agent pipeline I run daily in Claude Code: turn a vague idea into a PRD, decompose it into independently-buildable issues, implement test-first, review across four lenses, and capture what was learned — plus cc:proofread, a multi-agent reviewer that catches over-claims and unsupported numbers in analytical writing before it ships. ~13k lines of Python, 496 tests, runs straight from a clone.
The Workflow
/cc:discover → /cc:decompose → /cc:build → /cc:review → /cc:learn
idea issues TDD code check it document
Each step produces an artifact the next step consumes. Enter at any point.
Skills
| Skill | Purpose | Inputs | Outputs |
|---|---|---|---|
/cc:discover | Explore idea → stress test → PRD | Vague idea | docs/prds/*.md |
/cc:decompose | PRD → independent issues | PRD file | docs/issues/*.md + wave plan |
/cc:build | TDD red-green-refactor per issue | Issue file | Working code + tests |
/cc:review | 4-layer review (functional, security, architecture, adversarial) | Code changes | Findings + fixes |
/cc:proofread | Multi-lens pre-publish review for analytical writing (vault findings, PRDs, Slack drafts). Read-only. | Prose artifact (file or piped text) | Per-lens findings + coverage line + pre-mortem prompt |
/cc:learn | Document what was learned | Any experience | docs/solutions/*.md |
Companion helpers
The skills above depend on two helpers that ship in this repo:
| File | Installs to | Role |
|---|---|---|
commands/grill-me.md | ~/.claude/commands/grill-me.md | Vault-aware grilling skill. Invoked by cc:discover Phase 2. Stress-tests a plan against existing concept pages, queues proposed concept-page edits + decision findings for batch confirmation at the end. |
agents/learnings-researcher.md | ~/.claude/agents/learnings-researcher.md | Vendored subagent. Invoked by cc:discover Step 0 to surface prior vault knowledge before the dialogue starts. |
What it combines
| Source | What we took | What we dropped |
|---|---|---|
| Compound Engineering | Compound learning system, multi-agent review, grill-me interrogation | Separate brainstorm/grill (merged into discover) |
| Matt Pocock TDD | Red-green-refactor discipline, vertical slicing | — |
| Matt Pocock PRD | Structured PRD with testable acceptance criteria | — |
| Matt Pocock Issues | Independent issue decomposition with waves | — |
| Matt Pocock Architecture | Structural confusion detection | Standalone skill (merged into review) |
Install
# Clone
git clone https://github.com/kevin-nous/cc-skills.git ~/cc-skills
# Symlink into global Claude skills
ln -sf ~/cc-skills/cc:discover ~/.claude/skills/cc:discover
ln -sf ~/cc-skills/cc:decompose ~/.claude/skills/cc:decompose
ln -sf ~/cc-skills/cc:build ~/.claude/skills/cc:build
ln -sf ~/cc-skills/cc:review ~/.claude/skills/cc:review
ln -sf ~/cc-skills/cc:proofread ~/.claude/skills/cc:proofread
ln -sf ~/cc-skills/cc:learn ~/.claude/skills/cc:learn
# Companion helpers (cc:discover invokes these)
ln -sf ~/cc-skills/commands/grill-me.md ~/.claude/commands/grill-me.md
ln -sf ~/cc-skills/agents/learnings-researcher.md ~/.claude/agents/learnings-researcher.md
When to use what
- Starting something new? →
/cc:discover - Know what to build, need a plan? →
/cc:decompose - Ready to code? →
/cc:build - Code is done, check it? →
/cc:review - Learned something worth keeping? →
/cc:learn - Quick fix, no ceremony needed? → Just code. Not everything needs the full pipeline.
Tests
The cc:proofread engine ships a full pytest suite (496 tests — schema validation, lens orchestration, snapshot-based output rendering, mechanical-prepass checks). All paths resolve relative to the repo, so it runs straight from a clone:
pip install -r requirements.txt
cd cc:proofread && python3 -m pytest scripts/tests/ -q