Claude Skill Playbook
Littlemowmow/claude-skill-playbook
Orquestre Claude como team lead com swarms de agents e controle de token budget
Ver no GitHubKit completo de automação e otimização para Claude, com instruções globais (CLAUDE.md), playbook de delegação eficiente, sources catalog e installer. Permite usar Claude como um team lead orquestrando swarms de sub-agents mais baratos, mantendo controle de token budget e planos revisáveis antes de execução.
Instalação
npx convez-ai Littlemowmow/claude-skill-playbook1 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
- Automação de fluxos de desenvolvimento com múltiplos agents paralelos
- Otimização de uso de tokens em sessões longas com Claude
- Delegação estruturada de tarefas entre models de diferentes tiers (Opus/Haiku)
Veredito

Skill maduro e bem documentado que transforma a forma de usar Claude, passando de chatbot linear para orquestração de agents. Requer leitura completa do PLAYBOOK.md para ganho real, mas a estrutura de code é sólida e o conceito de efficient frontier + agent swarms é genuinamente inovador.
README
A tradução para o português deste README ainda está na fila.
Claude Skill Playbook — Mr. Faraaz Edition
A complete, downloadable handover of the Claude Code skill stack, the GitHub sources behind it, and the operating playbook for driving Claude the way it is meant to be driven: a high-tier model orchestrating swarms of cheaper sub-agents, every session running inside a token budget, every plan reviewed before a line of code is written.
This repo is the whole package. Clone it, run the installer, drop the CLAUDE.md
into ~/.claude/, and your Claude behaves like a senior operator on day one.
What's in here
| File | What it is |
|---|---|
CLAUDE.md | The global instructions file. Drop into ~/.claude/CLAUDE.md. Contains the canary protocol (Mr. Faraaz), the efficiency + agent-swarm doctrine, and the full activation playbook. This is the brain. |
SOURCES.md | Every GitHub repo / marketplace you need to clone or add, grouped by family, with install method per source. |
PLAYBOOK.md | The deep dive: how to delegate with efficient-frontier, how agent swarms work, how the design skills and superpowers fire, how visual-plan closes every plan. |
install.sh | One script. Clones every public skill repo into ~/.claude/skills/, wires the symlinks, adds the plugin marketplaces, and registers the Playwright MCP server. |
mcp/.mcp.json | Project-scope MCP template (Playwright). Copy to a project root to give Claude a live browser in that repo. |
.gitignore | Keeps secrets and local cruft out. |
Quick start
# 1. clone this repo
git clone https://github.com/Littlemowmow/claude-skill-playbook.git
cd claude-skill-playbook
# 2. install every skill source into ~/.claude/skills
bash install.sh
# 3. install the brain
cp CLAUDE.md ~/.claude/CLAUDE.md
# then open it and change "Mr. Faraaz" to your own name if you want a
# different canary — but read the Canary Protocol section first.
# 4. restart Claude Code. It should greet you as Mr. Faraaz and confirm
# the skills are loaded.
Then read PLAYBOOK.md start to finish. That's the part most people skip,
and it's the part that actually makes the difference.
The one idea, if you read nothing else
Don't use Claude like a chatbot that does one thing at a time. Use it like a team lead:
- The expensive model (Opus / Fable) makes the judgment calls — architecture, priorities, ambiguity, final review.
- It dispatches swarms of cheaper sub-agents to do the token-heavy grunt work in parallel — searching, reading, mechanical edits, test runs.
- Every session opens by loading the efficiency + budget skills, so the expensive tokens go to thinking, not grinding.
- Every plan gets a visual, reviewable form before execution.
That's the whole game. PLAYBOOK.md shows you exactly how.