Claude Code Bible
sebyx07/claude-code-bible
Padrões de projeto para Claude Code operar em máxima autonomia
Ver no GitHubPadrões estruturais comprovados para maximizar a autonomia do Claude na escrita de código, testes, deploys e gestão de infraestrutura. Abrange arquitetura, skills, MCP, monorepos e documentação comprimida — testado em 50+ projetos com diferentes stacks.
Instalação
npx convez-ai sebyx07/claude-code-bible1 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
- Estruturar projetos para trabalho autônomo com Claude
- Implementar skills, agents e slash commands
- Configurar monorepos multi-agente escaláveis
- Otimizar documentação e memory para economia de tokens
- Automatizar testes, deploys e CI/CD com Claude
- Definir permissões, hooks e quality gates
Veredito

Guia arquitectural altamente técnico e prático para quem quer extrair máxima produtividade do Claude em projetos reais. Abordagem stack-agnóstica com checkpoints de qualidade bem pensados — recomendado para equipes que já dominam CI/CD e workflows de automação.
README
A tradução para o português deste README ainda está na fila.
The Claude Code Bible
Proven patterns for structuring any project so Claude Code operates at maximum autonomy — writing production code, running tests, deploying, managing infrastructure, and learning from your feedback across sessions.
Battle-tested across 50+ projects — the whole spectrum of software. Rails, TypeScript, Python, Rust, Go, React, desktop apps, Android, iOS, Linux drivers, embedded, ML pipelines, CLIs. 18 skills, subagents, slash commands, infrastructure scripts, hooks, MCP integrations, monorepo workspaces. Stack-agnostic.
The two pillars
1. Architecture and coding patterns — SOLID, SRP, small files, service objects, automated testing. What makes code good for humans also makes it trivially navigable for AI. Claude across 50 small focused files is 10× faster than across 5 giant ones.
2. Compressed docs and planning — CLAUDE.md, skills, plans. Every markdown file Claude reads costs tokens on every session. A 50-line skill all signal beats a 500-line one padded with fluff. Write configs like you're paying per word — with AI, you are. See ch. 11.
Table of contents
| # | Chapter | What it covers |
|---|---|---|
| 01 | CLAUDE.md | The project brain Claude reads every session |
| 02 | Skills, Agents, Commands | Skills, subagents, and slash commands |
| 03 | Code Quality | SOLID, SRP, small files, linting, testing — why AI needs clean code |
| 04 | Scripts and Env | Infrastructure scripts, .env files, giving Claude hands |
| 05 | Hooks and Permissions | Quality gates, allow-all tools, trust boundaries |
| 06 | Monorepo Workspace | Multi-repo workspace, multi-agent, scaling across servers |
| 07 | Memory and MCP | Cross-conversation memory, MCP tool integrations, building efficient MCP servers |
| 08 | Ecosystem | Claude Task Master and the broader AI dev ecosystem |
| 09 | Checklist | Step-by-step setup checklist |
| 10 | Planning and Docs | Planning workflow, docs directory, concise documentation |
| 11 | Compressed Config | Writing CLAUDE.md and .claude/* for minimum tokens |
| 12 | Behavioral Rules | Steering how Claude codes — assumptions, simplicity, surgical diffs, success criteria |
The core idea
Claude reads your project every conversation. Good organization → better performance. Small files, clear patterns, documented commands, automated quality — same things that make code good for humans make it trivially navigable for AI.
The workflow this enables
/plan add feature X → Claude explores codebase, creates implementation plan
/implement → Claude executes the plan, writes code + tests
/qa → Claude reviews for regressions, runs full test suite
/deploy → Claude triggers CI, builds, deploys to prod
You approve at checkpoints. Claude does the rest.
Quick wins (do these first)
The four things that give you 80% of the value in under an hour:
- CLAUDE.md with commands — document your test, lint, and deploy commands
- Skip the prompts —
claude --dangerously-skip-permissions. Zero friction. See ch. 5 - Response + coding rules — "No preamble, lead with action, disagree when wrong" changes Claude's personality; the behavioral rules block changes its code
- One lint command —
bin/lintthat auto-fixes everything, wired as a pre-commit hook
New to this? Start with the Checklist.
Stack-agnostic
Language-agnostic. Examples across chapters span Rails, TypeScript, Python, Rust, Go, C, Swift, Kotlin. The structure — CLAUDE.md, skills, scripts, hooks, permissions — works identically across backend services, frontend apps, desktop, mobile, drivers, embedded, ML.
License
MIT