claude-code

name	claude-code
description	Router skill for Claude Code (the CLI tool itself). Contains intent hints and a dispatch table that maps a user's question to the surface-specific deep-reference file Claude should Read next. Use when the user asks about Claude Code internals: editing .claude/settings.json or settings.local.json, authoring or debugging hooks (PreToolUse, PostToolUse, Stop, SubagentStop, Notification, UserPromptSubmit, PreCompact, SessionStart, SessionEnd, etc.), writing slash commands or agents, configuring MCP servers in .mcp.json, building plugins or marketplaces, setting ANTHROPIC_* / CLAUDE_* env vars, troubleshooting permission modes, understanding the ~/.claude/ directory layout, or asking "what does <feature> in Claude Code do". Skip: questions about the Anthropic Messages API (use anthropic-api), the Claude Agent SDK (use claude-agent-sdk), the MCP protocol spec (use mcp-spec), or general shell / git / programming topics not specific to Claude Code internals.
user-invocable	true

Claude Code Reference — Router

Field	Value
Claude Code version	v2.1.158 (pinned in CI; updated by pipeline)
Source docs	code.claude.com/docs
GitHub	anthropics/claude-code
npm	@anthropic-ai/claude-code

This skill is auto-updated daily. A matrix pipeline reads the upstream docs and rewrites the per-surface files below. Section structure is stable; content drifts to track upstream.

Dispatch table

Read the surface file(s) matching the user's question. Each surface file holds its complete reference for that one topic and cross-links (via markdown links) to siblings when a fact spans surfaces — so you only need to Read the matching file(s), not all of them.

Surface file	Read when the user asks about…
SKILL-settings.md	settings.json, settings.local.json, project vs user vs local scope, settings keys / defaults / types
SKILL-hooks.md	hook events (PreToolUse, PostToolUse, Stop, SubagentStop, Notification, UserPromptSubmit, PreCompact, SessionStart, SessionEnd), hook input/output JSON shapes, matchers, blocking vs non-blocking
SKILL-slash-commands.md	slash command authoring, frontmatter schema, argument syntax, $ARGUMENTS, command discovery paths
SKILL-mcp.md	.mcp.json schema, MCP transports (stdio / http / sse), server config, MCP tool naming, mcp__ prefix
SKILL-plugins.md	plugin manifest (.claude-plugin/plugin.json), marketplaces (marketplace.json), plugin install / scope, plugin commands / agents / skills / hooks, CLI hint protocol (<claude-code-hint />) for recommending plugins from third-party CLIs
SKILL-cli.md	CLI flags, subcommands, env vars (ANTHROPIC_* / CLAUDE_*), permission modes (default / acceptEdits / plan / bypassPermissions), ~/.claude/ directory layout, IDE integrations, session management (resume, naming, branching, picker, transcript storage), sandbox environments and isolation approaches (sandboxed Bash tool, sandbox runtime, dev containers, VMs)
SKILL-known-issues.md	a user reports a bug, asks about a workaround, mentions an error message, or asks "why does X not work"

When unsure which file to read

1. If the user mentions a specific file path (e.g., .mcp.json, settings.json, plugin.json), read the matching surface file.
2. If the user mentions an error message or unexpected behavior, read SKILL-known-issues.md first.
3. If the question spans multiple surfaces (e.g., "how do hooks and plugins interact"), read both surface files.
4. If still unclear, prefer reading SKILL-cli.md — it's the broadest catch-all surface.

Auto-correction rules

Path-scoped correction rules live in rules/:

Rule file	Triggers on edits to
rules/settings.md	**/.claude/settings.json, **/.claude/settings.local.json
rules/mcp.md	**/.mcp.json
rules/plugins.md	**/.claude-plugin/plugin.json, **/marketplace.json
rules/hooks.md	**/.claude/hooks/**
rules/skills-agents-commands.md	**/.claude/skills/**/SKILL.md, **/.claude/agents/**, **/.claude/commands/**

---

This skill is auto-updated daily by a maintainer-run pipeline. If you spot a bug in this content, file an issue at xiaolai/anthropic-docs — SKILL.md fixes flow back through the next research run, not via PRs.