convez.ai
SkillAutomaçãoAvançado
kevin-nous

cc-skills: Pipeline Multi-Agent para Desenvolvimento Test-First

kevin-nous/cc-skills

Pipeline completo: ideia → PRD → issues → código TDD → review → aprendizados

Ver no GitHub
InicialPluginScore · de 0 a 100

Conjunto 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-skills

1 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

Estrelas
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

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

SkillPurposeInputsOutputs
/cc:discoverExplore idea → stress test → PRDVague ideadocs/prds/*.md
/cc:decomposePRD → independent issuesPRD filedocs/issues/*.md + wave plan
/cc:buildTDD red-green-refactor per issueIssue fileWorking code + tests
/cc:review4-layer review (functional, security, architecture, adversarial)Code changesFindings + fixes
/cc:proofreadMulti-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:learnDocument what was learnedAny experiencedocs/solutions/*.md

Companion helpers

The skills above depend on two helpers that ship in this repo:

FileInstalls toRole
commands/grill-me.md~/.claude/commands/grill-me.mdVault-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.mdVendored subagent. Invoked by cc:discover Step 0 to surface prior vault knowledge before the dialogue starts.

What it combines

SourceWhat we tookWhat we dropped
Compound EngineeringCompound learning system, multi-agent review, grill-me interrogationSeparate brainstorm/grill (merged into discover)
Matt Pocock TDDRed-green-refactor discipline, vertical slicing—
Matt Pocock PRDStructured PRD with testable acceptance criteria—
Matt Pocock IssuesIndependent issue decomposition with waves—
Matt Pocock ArchitectureStructural confusion detectionStandalone 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
Ver o original no GitHub

Bom para

Relacionados