Claude code

The definitive reference skill for Claude Code's entire extensibility platform. This skill covers every feature area — from plugins and hooks to channels, subagents, MCP servers, scheduled tasks, and headless automation. When a user asks how to build, configure, debug, or extend Claude Code, this skill provides authoritative guidance grounded in the official documentation.

Documentation Index

Fetch the complete documentation index at: https://code.claude.com/docs/llms.txt Use this file to discover all available pages before exploring further.

Reference Library

27 reference documents covering every Claude Code feature. Read the relevant reference(s) before answering any question.

Extensibility & Plugins

Reference	What It Covers
plugins.md	Creating plugins — manifest, agents, skills, hooks, commands, settings
plugins-reference.md	Plugin spec — plugin.json schema, variable resolution, file structure
plugin-marketplaces.md	Creating and distributing plugin marketplaces
discover-plugins.md	Discovering and installing prebuilt plugins
skills.md	Extending Claude with skills — SKILL.md format, frontmatter, scripts
sub-agents.md	Creating custom subagents — agent .md files, model routing, tools

Hooks & Automation

Reference	What It Covers
hooks-guide.md	Hooks walkthrough — setup, common patterns, prompt/agent hooks
hooks-ref.md	Hooks reference — all 22 events, 4 handler types, input schemas, decision control
scheduled-tasks.md	Running prompts on a schedule (cron-style)
channels.md	Pushing events into a running session
channels-reference.md	Channels API reference — event types, protocols

Multi-Session & Programmatic

Reference	What It Covers
agent-teams.md	Orchestrating teams of Claude Code sessions
headless.md	Running Claude Code programmatically (CI, scripts, pipelines)

Configuration & Security

Reference	What It Covers
settings.md	All Claude Code settings — global, project, plugin scopes
permissions.md	Permission modes, allow/deny lists, tool-level control
sandboxing.md	Sandbox environments — Docker, macOS sandbox, network isolation
model-config.md	Model selection, routing, provider configuration
network-config.md	Enterprise network — proxies, certificates, firewall rules
llm-gateway.md	LLM gateway configuration for enterprise deployments

Tools & Interface

Reference	What It Covers
tools-reference.md	Built-in tools — Read, Write, Edit, Bash, Glob, Grep, Agent, etc.
mcp.md	Connecting Claude Code to external tools via MCP servers
keybindings.md	Customizing keyboard shortcuts
statusline.md	Customizing the status line display
output-styles.md	Output formatting and style configuration
terminal-config.md	Terminal setup optimization
voice-dictation.md	Voice input configuration
troubleshooting.md	Common issues, diagnostics, and fixes

Workflow

1. Identify the topic — Parse the user's question to determine which feature area(s) are involved
2. Read references — Load the relevant reference file(s) from the table above. For cross-cutting questions, read multiple references.
3. Synthesize — Combine reference material with the user's specific context (their project structure, existing config, goals)
4. Provide guidance — Give a concrete, actionable answer with code examples. Always cite which reference the guidance comes from.
5. Verify — If the user is building something (plugin, hook, skill), validate it against the spec in the reference docs

Topic Routing

User Asks About	Read These References
Building a plugin	plugins.md, plugins-reference.md
Publishing to marketplace	plugin-marketplaces.md, discover-plugins.md
Writing a skill	skills.md, plugins.md
Creating an agent	sub-agents.md, plugins.md
Setting up hooks	hooks-guide.md, hooks-ref.md
Hook events / types	hooks-ref.md
MCP servers	mcp.md
Permissions / security	permissions.md, sandboxing.md
Settings config	settings.md
Running headless / CI	headless.md
Agent teams / multi-session	agent-teams.md, channels.md
Channels / events	channels.md, channels-reference.md
Scheduled / cron tasks	scheduled-tasks.md
Model selection	model-config.md
Enterprise / network	network-config.md, llm-gateway.md
Keyboard shortcuts	keybindings.md
Status line	statusline.md
Output formatting	output-styles.md
Terminal setup	terminal-config.md
Voice input	voice-dictation.md
Something broken	troubleshooting.md
Built-in tools	tools-reference.md

Key Concepts

The 4 Hook Types

• command — Runs a shell command. Exit code 0 = allow, 2 = block.
• http — POSTs JSON to a URL. Response body controls decision.
• prompt — Single-turn LLM judgment. Sends prompt + $ARGUMENTS to Claude for yes/no.
• agent — Multi-turn subagent with Read/Grep/Glob tools. Full reasoning for complex checks.

The 22 Hook Events

SessionStart, UserPromptSubmit, PreToolUse, PermissionRequest, PostToolUse, PostToolUseFailure, Notification, SubagentStart, SubagentStop, Stop, StopFailure, TeammateIdle, TaskCompleted, InstructionsLoaded, ConfigChange, WorktreeCreate, WorktreeRemove, PreCompact, PostCompact, Elicitation, ElicitationResult, SessionEnd

Plugin Variables

Variable	Resolves To
${CLAUDE_PLUGIN_ROOT}	Plugin's root directory
${CLAUDE_PLUGIN_DATA}	Plugin's data/output directory
${CLAUDE_SKILL_DIR}	Current skill's directory

Plugin File Structure

plugin-name/
├── .claude-plugin/plugin.json    ← Manifest (required)
├── agents/*.md                   ← Role definitions
├── skills/*/SKILL.md             ← Workflow definitions
├── commands/*.md                 ← Entry points
├── context/*.md                  ← Shared context
├── hooks/hooks.json              ← Hook definitions
└── settings.json                 ← Plugin settings

After this skill

• Configuration question answered → apply via update-config skill if settings.json change needed
• Keybinding question answered → apply via keybindings-help skill
• Hook setup needed → update-config skill handles all hook configuration