claude-docs

name	claude-docs
description	Claude Code 공식 문서 레퍼런스. 스킬 생성/개선, hooks, subagents, agent teams, MCP, plugins, permissions, memory, CLI 설정 참조. Use when: creating skills, improving SKILL.md, writing hooks, configuring subagents, building agent teams, setting up MCP, reviewing permissions. Keywords: "스킬 만들", "skill 만들", "SKILL.md", "hook 설정", "hook 만들", "subagent", "sub-agent", "agent team", "MCP 설정", "plugin", "permission", "allowed-tools", "frontmatter"
argument-hint	[topic: skills|hooks|subagents|teams|mcp|plugins|permissions|memory|settings|cli|best-practices|create|workflows|overview|monitoring]
allowed-tools	Bash(bash:*), Read, Grep, Glob

Claude Code Documentation Reference

Dynamic Documentation Loading

!bash "$CLAUDE_PROJECT_DIR/.claude/skills/claude-docs/scripts/load-docs.sh" $ARGUMENTS

Document Index

39 official docs available in claude-docs/. Full index at .claude/skills/claude-docs/references/doc-index.md.

Key Documents by Topic

Topic	Primary Doc	Lines	Secondary
Skills	skills.md	678	best-practices.md (599)
Hooks	hooks-guide.md	633	hooks.md (1553)
Sub-agents	sub-agents.md	811
Agent Teams	agent-teams.md	387
MCP	mcp.md	1198
Plugins	plugins.md	410	plugins-reference.md (743)
Permissions	permissions.md	258	andboxing.md (261)
Memory	memory.md	299
Settings	settings.md	936
CLI	cli-reference.md	161	headless.md (171)

SKILL.md Frontmatter Quick Reference

---
name: skill-name                    # Required. Matches directory name.
description: |                      # Required. Multi-line. Triggers auto-invocation.
  What this skill does.
  Use when: [trigger conditions]
  Keywords: "keyword1", "keyword2"  # Auto-trigger keywords
argument-hint: "[args description]" # Shown in /skill-name help
allowed-tools: Tool1, Tool2        # Tools the skill can use
  # Bash(command:pattern)           # Restrict bash to specific commands
  # Bash(bash:*)                    # Allow all bash
  # Read, Write, Edit, Glob, Grep  # File tools
  # WebFetch, WebSearch             # Web tools
  # Task(agent-name)                # Specific subagent
context: fork                       # Optional. Forks context (no conversation history)
---

Dynamic Context Syntax

# Static content here...

## Dynamic Section
!`bash "command" $ARGUMENTS`

# More static content...
!`bash "another-command"`

• $ARGUMENTS: User's arguments after /skill-name
• Commands run at skill load time, output replaces the line
• Use $CLAUDE_PROJECT_DIR for project-relative paths

Skill Creation Checklist

When creating or improving a skill:

1. Choose pattern: Simple (static text) | Script (dynamic !cmd``) | Reference (doc loading) | Protocol (multi-step)
2. Write frontmatter:
   • name matches directory name
   • description includes trigger keywords for auto-invocation
   • allowed-tools is minimal (principle of least privilege)
   • argument-hint if skill accepts arguments
3. Write body: Instructions, templates, examples
4. Test: /skill-name runs without errors, arguments work

This Project's Skill Patterns

Simple Pattern (static instructions)

weather/, calendar/, maps/ - API calls + response format templates

Script Pattern (dynamic data)

finance/ - !bash "script.sh"`` loads live data at skill invocation

Reference Pattern (doc loading)

agui/, a2ui/, a2a-protocol/ - Large reference docs with examples

Protocol Pattern (multi-step workflows)

youtube-shorts/, x-to-shorts/ - Sequential steps with intermediate outputs

Loading Additional Documents

If the loaded topic doesn't cover what you need:

# Read any doc directly
Read claude-docs/FILENAME.md

# Search across all docs
Grep "pattern" claude-docs/