Plugin Development

Plugin Structure

my-plugin/
├── .claude-plugin/
│   └── plugin.json          # Required: metadata
├── commands/                # Slash commands
│   └── hello.md
├── agents/                  # Subagents
│   └── reviewer.md
├── skills/                  # Agent skills
│   └── code-review/
│       └── SKILL.md
├── hooks/                   # Event handlers
│   └── hooks.json
└── .mcp.json               # MCP server config

Creating Commands

File: commands/my-command.md

---
description: Short description for help
---

# Command Name

Instructions for Claude to execute this command.
Use $ARGUMENTS for user input.
Use $1, $2 for positional parameters.

Command naming: /plugin-name:command-name

Creating Skills

Directory: skills/my-skill/SKILL.md

---
name: my-skill
description: Description with triggers. Triggers: keyword1, keyword2
---

# My Skill

Instructions and workflows...

Creating Hooks

File: hooks/hooks.json

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Write|Edit",
        "hooks": [
          {
            "type": "command",
            "command": "${CLAUDE_PLUGIN_ROOT}/scripts/script.sh"
          }
        ]
      }
    ]
  }
}

Supported Events

• PreToolUse - Before tool execution
• PostToolUse - After successful execution
• PostToolUseFailure - After failed execution
• SessionStart - Session begins
• SessionEnd - Session ends
• UserPromptSubmit - User sends prompt

Creating Agents

File: agents/my-agent.md

---
description: Agent specialization
capabilities: ["code-review", "testing"]
---

# Agent Name

Detailed role description...

Environment Variables

• ${CLAUDE_PLUGIN_ROOT} - Plugin root directory
• $TOOL_NAME - Current tool name
• $TOOL_INPUT - Tool input
• $TOOL_OUTPUT - Tool output