Claude Dev Team Skill
npalacios310/claude-dev-team-skill
Time de software completo automatizado: planejar, construir, validar, securizar, deploy.
Ver no GitHubUma skill portável que simula um time de entrega de software completo dentro do Claude Code. Automatiza todo o ciclo de vida do projeto — da estratégia à produção — com 12 agentes especializados (arquiteto, planejador, engenheiro de design, desenvolvedor backend, QA, segurança, DevOps, etc.) e dois skills de suporte que orquestram workflows end-to-end e fornecem referências de design.
Instalação
npx convez-ai npalacios310/claude-dev-team-skill1 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
- Bootstrap e padronização de novos projetos
- Auditoria e melhoria de repositórios herdados
- Ciclo completo de design e desenvolvimento com loops de crítica
- Revisão de segurança OWASP antes do deploy
- Monitoramento recorrente de inteligência competitiva pós-lançamento
Veredito

Skill sofisticada e bem estruturada que automatiza um pipeline de engenharia real com múltiplos agentes especializados. Requer curva de aprendizado significativa, mas oferece valor genuíno para times que buscam accelerar entrega end-to-end em Claude Code. Stack-agnóstico e pronto para produção.
README
A tradução para o português deste README ainda está na fila.
claude-dev-team-skill
A portable, custom software delivery team for Claude Code — built to start new projects and bring inherited ones up to standard, across any stack.
What's inside
Lifecycle: 0 Solve (optional) → 1 Plan → 2 Build (+ design-critic loop) → 3 QA → 4 Security → 5 Docs → 6 Deploy → 7 Intelligence (optional, schedulable).
Agents (12)
| Agent | Role |
|---|---|
tech-lead-orchestrator | Runs the pipeline and routes work between agents |
solve-strategist | Optional pre-build product strategy + market research — evidence-based PRD that feeds the planner |
project-planner | Precise requirement + dependency-ordered work plan |
design-engineer | Crafted, non-generic, fully responsive UI — picks one Design Kit + Screen Blueprint, builds, then screenshots its work at 360/768/1280 |
backend-engineer | Clean, optimized, modern backend/full-stack code |
design-critic | Adversarial visual reviewer — judges the real screenshots, loops with design-engineer |
qa-breaker | Adversarial QA — tries to break the code, loops with the dev |
security-auditor | OWASP-style review before delivery |
doc-writer | README, codemaps, API docs, ADRs |
devops-deploy | CI/CD, containers, deploy + rollback |
intel-watcher | Optional, recurring post-launch competitive-intelligence digest — built to run on a schedule via /schedule |
project-scout | Read-only explorer/auditor for inherited repos |
Skills (2)
| Skill | Purpose |
|---|---|
project-kickstart | Bootstrap a new project or standardize an existing one |
design-language | The crafted, anti-"AI-look" design reference the design-engineer studies — 6 curated Design Kits (Editorial SaaS, Utility Dashboard, Premium Commerce, Warm Commerce, Dev Tool, Refined Premium) and 10 Screen Blueprints (grow it with your own references) |
Install
As a plugin (recommended — portable across machines/projects)
- Push this folder to a GitHub repo (e.g.
npalacios310/claude-dev-team-skill). - In Claude Code:
/plugin marketplace add https://github.com/npalacios310/claude-dev-team-skill /plugin install claude-dev-team-skill@claude-dev-team-skill
As a user-level copy (this machine only)
Copy-Item "agents\*.md" "$HOME\.claude\agents\" -Force
Copy-Item -Recurse "skills\*" "$HOME\.claude\skills\" -Force
Don't stack both methods. If you install the plugin, remove the user-level copies to avoid duplicates.
Usage
Let the tech-lead coordinate the whole flow, or call an agent directly:
Use the tech-lead-orchestrator to build <feature> end to end.
Use the design-engineer to design a responsive hero section.
Use the solve-strategist to research the market for <product idea> and produce an evidence-based PRD before we plan.
Use the intel-watcher to set up a recurring competitive-intelligence digest for <product> via /schedule.
Notes
- Stack-agnostic: every agent detects the project's stack and adapts.
- The QA↔dev fix loop is driven by the tech-lead from the main session, since a Claude Code subagent cannot dispatch another subagent.
design-engineermust deliver real screenshots (360/768/1280) anddesign-criticreviews them — loop driven by the tech-lead, max 3 rounds, then it escalates to the user.author/ownerin.claude-plugin/*.jsonare set tonpalacios310— change them if you fork.