Claude Code Reference

Claude Code Extension Reference

Authoritative reference for Claude Code extension types. Consult this before implementing any extension.

Quick Disambiguation

When to Use Each

Use Skills When

• Claude should discover and apply knowledge automatically
• You need multiple reference files, scripts, or templates
• Building team workflows requiring standardized guidance
• Creating capabilities that span multiple steps

NOT for: Simple prompts, one-off commands, external integrations

Use Slash Commands When

• You invoke the same prompt repeatedly
• The prompt fits in a single markdown file
• You want explicit control over when it runs
• Quick templates or reminders

NOT for: Complex multi-file workflows, automatic discovery

Use Hooks When

• You need scripts to run on specific events (tool use, session start/end)
• Automating linting, formatting, or validation after file changes
• Adding context before prompts are processed
• Controlling whether Claude can stop or continue

NOT for: Adding knowledge, user-invoked actions

Use MCP Servers When

• Connecting Claude to external tools, databases, or APIs
• Providing new tool capabilities (not just knowledge)
• Building integrations that require authentication
• Exposing resources from external systems

NOT for: Adding knowledge about how to use tools (use Skills for that)

Key Differences

Skills vs Slash Commands

Skills vs MCP

• Skills tell Claude HOW to use tools (knowledge, patterns, workflows)
• MCP PROVIDES the tools (actual capabilities, external connections)

Example: MCP server connects Claude to your database. A Skill teaches Claude your data model and query patterns.

Hooks vs Skills

• Hooks run scripts on events (automated, no Claude involvement in execution)
• Skills add knowledge to conversations (Claude uses the guidance)

Reference Files

For detailed implementation guidance, consult these reference files:

Skills Quick Reference

Required Structure

skill-name/
  SKILL.md           # Required (under 500 lines)
  reference.md       # Optional detailed docs
  examples.md        # Optional examples
  scripts/           # Optional utility scripts

YAML Frontmatter

---
name: skill-name
description: What it does and when to use it. Include trigger phrases.
allowed-tools: ["Read", "Grep", "Glob"]  # Optional: restricts tools
model: sonnet  # Optional: force specific model
---

Location Hierarchy (higher overrides lower)

1. Enterprise: Managed settings
2. Personal: ~/.claude/skills/
3. Project: .claude/skills/
4. Plugin: Bundled with plugins

Naming Rules

• Lowercase letters, numbers, hyphens only
• Maximum 64 characters
• Use gerund form (writing-skills not skill-writer)

Slash Commands Quick Reference

Required Structure

.claude/commands/command-name.md  # Project command
~/.claude/commands/command-name.md  # Personal command

Frontmatter Options

---
allowed-tools: Bash(git:*), Read, Write
argument-hint: [arg1] [arg2]
description: Brief description
model: claude-3-5-haiku-20241022
disable-model-invocation: false
---

Features

• $ARGUMENTS - All arguments passed
• $1, $2, etc. - Positional arguments
• !git status`` - Execute bash before command
• @file.ts - Reference files

Namespacing

• .claude/commands/frontend/test.md creates /test (shows "project:frontend")
• Project commands override personal commands with same name

Hooks Quick Reference

Configuration Location

// ~/.claude/settings.json (user)
// .claude/settings.json (project)
{
  "hooks": {
    "EventName": [
      {
        "matcher": "ToolPattern",
        "hooks": [
          {
            "type": "command",
            "command": "your-script.sh"
          }
        ]
      }
    ]
  }
}

Hook Events

Matcher Patterns

• Exact: Write matches only Write tool
• Regex: Edit|Write matches either
• * or empty: matches all tools

Exit Codes

• 0: Success (stdout shown in verbose mode)
• 2: Blocking error (stderr fed to Claude)
• Other: Non-blocking error

MCP Quick Reference

Adding Servers

# Remote HTTP (recommended)
claude mcp add --transport http name https://url

# Remote SSE (deprecated)
claude mcp add --transport sse name https://url

# Local stdio
claude mcp add --transport stdio name -- npx -y package

Scopes

• local (default): Your project only
• project: Shared via .mcp.json
• user: All your projects

Management Commands

claude mcp list              # List servers
claude mcp get <name>        # Get details
claude mcp remove <name>     # Remove server
/mcp                         # In Claude Code: status and auth

MCP Tool Naming

Tools follow pattern: mcp__<server>__<tool>

Example: mcp__github__search_repositories

Common Misconceptions

Misconception: Skills are like slash commands

Reality: Skills are automatically discovered based on description matching. Slash commands require explicit /command invocation.

Misconception: Hooks add knowledge to Claude

Reality: Hooks run scripts on events. They can inject context (via stdout for UserPromptSubmit/SessionStart) but don't teach Claude new concepts.

Misconception: MCP servers teach Claude how to use tools

Reality: MCP servers provide tools. Skills teach Claude how and when to use those tools effectively.

Misconception: I need MCP for external APIs

Reality: Claude has built-in WebFetch and WebSearch. Use MCP when you need authenticated access, complex tool logic, or custom data sources.

Misconception: Skills execute code

Reality: Skills provide knowledge. They can bundle scripts that Claude executes using Bash, but the skill itself doesn't run code.

Troubleshooting

Skill Not Triggering

1. Check description includes trigger phrases users would say
2. Verify SKILL.md path: ~/.claude/skills/name/SKILL.md or .claude/skills/name/SKILL.md
3. Restart Claude Code after changes
4. Run claude --debug to see loading errors

Hook Not Running

1. Check JSON syntax in settings file
2. Verify matcher pattern matches tool name (case-sensitive)
3. Check script is executable: chmod +x script.sh
4. Use full paths with $CLAUDE_PROJECT_DIR

MCP Server Not Connecting

1. Check server status: /mcp
2. Verify URL or command is correct
3. For OAuth servers, authenticate via /mcp
4. Check timeout: MCP_TIMEOUT=10000 claude

Slash Command Not Appearing

1. Verify file location: .claude/commands/name.md
2. Check YAML frontmatter syntax
3. Ensure description frontmatter is set (required for SlashCommand tool)
4. Restart Claude Code

Extension Decision Flowchart

Need to extend Claude Code?
            |
            v
    Is it knowledge/guidance?
        |           |
       YES         NO
        |           |
        v           v
   Single file?   External tool/data?
     |    |        |           |
    YES   NO      YES         NO
     |    |        |           |
     v    v        v           v
 Command  Skill   MCP      Event-driven?
                               |    |
                              YES   NO
                               |    |
                               v    v
                             Hook  Reconsider need

Plugin Marketplaces

Distribute plugins (skills, hooks, commands, agents, MCP servers) via marketplaces:

Creating a Marketplace

// .claude-plugin/marketplace.json
{
  "name": "my-marketplace",
  "owner": { "name": "Your Name" },
  "plugins": [
    {
      "name": "my-skill",
      "source": "./plugins/my-skill",
      "description": "What it does"
    }
  ]
}

Plugin Sources

Installing from Marketplaces

# Add a marketplace
/plugin marketplace add owner/repo

# Install a plugin
/plugin install plugin-name@marketplace-name

# Update marketplace
/plugin marketplace update

Team Configuration

// .claude/settings.json
{
  "extraKnownMarketplaces": {
    "company-tools": {
      "source": {"source": "github", "repo": "your-org/plugins"}
    }
  }
}

See: https://code.claude.com/docs/en/plugin-marketplaces

Deep-Dive Skills

For detailed implementation guidance, install these companion skills:

From npm package:

# If installed via: bun add claude-code-sdk
ln -s node_modules/claude-code-sdk/skills/creating-hooks ~/.claude/skills/
ln -s node_modules/claude-code-sdk/skills/creating-mcp-servers ~/.claude/skills/
ln -s node_modules/claude-code-sdk/skills/transcript-intelligence ~/.claude/skills/
ln -s node_modules/claude-code-sdk/skills/writing-skills ~/.claude/skills/

From GitHub:

git clone https://github.com/hgeldenhuys/claude-code-sdk /tmp/sdk
cp -r /tmp/sdk/skills/* ~/.claude/skills/

Version Information

This reference is based on Claude Code documentation as of January 2025. For the latest information:

• Skills: https://code.claude.com/docs/en/skills
• Slash Commands: https://code.claude.com/docs/en/slash-commands
• Hooks: https://code.claude.com/docs/en/hooks
• MCP: https://code.claude.com/docs/en/mcp
• Plugin Marketplaces: https://code.claude.com/docs/en/plugin-marketplaces