Claude Code Docs Skill
mehmetbaykar/claude-code-docs-skill
Documentação offline do Claude Code como skill instalável
Ver no GitHubSkill que espelha localmente a documentação oficial do Claude Code, mantendo 132 páginas Markdown sincronizadas automaticamente a cada 3 horas via GitHub Actions. Permite que agentes locais acessem referências completas de hooks, APIs e guias de desenvolvimento sem dependência de conectividade externa.
Instalação
npx convez-ai mehmetbaykar/claude-code-docs-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
- Desenvolvimento de agentes Claude com acesso offline à documentação
- Integração de referências técnicas em workflows locais
- Consulta rápida de hooks e APIs durante desenvolvimento
Veredito

Solução prática para manter documentação do Claude Code sincronizada localmente. Útil para desenvolvedores que precisam de referências confiáveis em ambientes desconectados, embora seja essencialmente um espelho estático sem lógica de processamento adicional.
README
A tradução para o português deste README ainda está na fila.
Claude Code Docs Skill
Local Agent Skill mirror of the Anthropic Claude Code documentation from https://code.claude.com/docs/en/.
The installable skill lives in skills/claude-code-docs/: SKILL.md is the
entry point, cleaned Markdown copies of every relevant Claude Code page live
under skills/claude-code-docs/references/, and a 3-hour GitHub Action keeps
them in sync with upstream.
Install
npx skills add mehmetbaykar/claude-code-docs-skill
The npx skills CLI discovers the nested skill automatically. Installing the
repo exposes only the skill directory (SKILL.md, provider metadata in
agents/, and references/) to the target agent while repository maintenance
files stay at the repo root.
Usage
Once installed, invoke the skill with a topic from your agent
($claude-code-docs hooks in Codex, /claude-code-docs hooks in Claude Code)
or with no argument to list topics. The full agent-facing usage contract lives
in skills/claude-code-docs/SKILL.md.
What's mirrored
The fetcher reads the Claude Code sitemap at
https://code.claude.com/docs/sitemap.xml, keeps every English documentation
URL whose path starts with /docs/en/, and excludes defensive non-core
subtrees:
/docs/en/tool-use/*/docs/en/examples/*/docs/en/legacy/*/docs/en/api/*/docs/en/reference/*
The mirror contains 132 cleaned Markdown files plus
skills/claude-code-docs/references/INDEX.md and
skills/claude-code-docs/references/docs_manifest.json. Any sitemap entries we
intentionally skip are recorded in docs_manifest.json under skipped.
Update
npx skills update claude-code-docs-skill # update an installed local copy
Upstream refreshes happen automatically every 3 hours; there is nothing to configure on the consumer side.
Refresh locally (maintainers only)
python3 -m venv .venv
.venv/bin/python -m pip install -r scripts/requirements.txt
.venv/bin/python scripts/fetch_claude_code_docs.py
The fetcher reads the sitemap, downloads each page's .md source, cleans MDX
and JSX wrappers into plain Markdown, and rewrites
skills/claude-code-docs/references/INDEX.md and
skills/claude-code-docs/references/docs_manifest.json. Files whose content
hash is unchanged are not rewritten.
Repository layout
.
├── skills/
│ └── claude-code-docs/
│ ├── agents/
│ │ └── openai.yaml # Agent UI metadata + invocation policy
│ ├── SKILL.md # installed skill instructions and routing
│ └── references/ # mirrored docs + INDEX + manifest
├── scripts/
│ ├── fetch_claude_code_docs.py # sitemap -> fetch -> clean -> write
│ └── requirements.txt
└── .github/workflows/
└── update-docs.yml # cron every 3 hours
Troubleshooting
- If docs look stale, check the latest run of Update Claude Code Documentation on this repository and reproduce locally with the steps in "Refresh locally" above.
- If the scheduled fetch fails, the workflow opens or updates an issue automatically.
- If a single page renders poorly, the upstream MDX is preserved under
skills/claude-code-docs/references/_raw/whenever the cleaner falls back, so the source of truth is never lost.
Notes
This repository is an unofficial local mirror packaged as an Agent Skill. It is not affiliated with, endorsed by, or sponsored by Anthropic.
Documentation content belongs to Anthropic and is subject to Anthropic's applicable terms and policies. The MIT license in this repository applies only to the mirroring tool, scripts, skill metadata, and repository-specific code.