claude-docs

name	claude-docs
description	Ground Claude Code configuration and feature work — and any work on Anthropic's products — in the latest OFFICIAL docs instead of training knowledge, which drifts because these ship constantly. Three doc hosts. Claude Code (code.claude.com) — skills, hooks, subagents, slash/custom commands, MCP servers, settings.json, permissions, plugins, statusline, output styles, memory/CLAUDE.md, the CLI, the Agent SDK. The Anthropic API / developer platform (platform.claude.com) — Messages API, models, prompt caching, tool use, batch, files, citations, prompt engineering, Managed Agents, Agent Skills. The Help Center (support.claude.com) — account, billing, SSO/JIT/SCIM, Connectors, Claude in Chrome, how-to guides, release notes. Use when building, editing, auditing, reviewing, or debugging any of the above — including authoring or scaffolding a skill, subagent, hook, slash command, or plugin — or answering "how does Claude Code / the Claude API do X" or "what's the latest way to configure X". Always prefer this over guessing.
when_to_use	Use to GROUND work in current official docs — look up the authoritative page, verify a field or behavior, cite it. For actually building, debugging, or migrating a Claude API / Anthropic SDK app (prompt caching, model version bumps, tool-use wiring) prefer the claude-api skill; this skill supplies the docs, not the implementation.

Claude docs (live)

Fetch current official Anthropic documentation on demand. Never answer a Claude Code, Claude API, or Anthropic product question from memory when the live docs can confirm it — these products ship often and training data drifts.

The three hosts (route by domain of the question)

Pick the authoritative host first; don't blend them.

Host	Authoritative for	Index (llms.txt)