plugin-authoring（插件创作）

Plugin Authoring — Official Best Practices

What Plugins Do

Plugins bundle skills, agents, hooks, and MCP servers into shareable packages. Skills get namespaced: /plugin-name:skill-name.

Directory Structure

my-plugin/
├── .claude-plugin/
│   └── plugin.json          # Manifest (ONLY file inside .claude-plugin/)
├── skills/                  # Agent Skills with SKILL.md
│   └── code-review/
│       └── SKILL.md
├── commands/                # Skills as simple Markdown files
│   └── review.md
├── agents/                  # Custom agent definitions
│   └── reviewer.md
├── hooks/
│   └── hooks.json           # Hook configurations
├── .mcp.json                # MCP server configurations
├── .lsp.json                # LSP server configurations
├── settings.json            # Default settings (only "agent" key supported)
└── README.md                # Documentation

Critical: skills/, agents/, hooks/ go at plugin root, NOT inside .claude-plugin/.

plugin.json Manifest

See quick-ref/manifest-schema.md for the complete schema.

{
  "name": "my-plugin",
  "description": "A code quality plugin with review skills and formatting hooks",
  "version": "1.0.0",
  "author": {
    "name": "Your Name"
  },
  "homepage": "https://github.com/user/my-plugin",
  "repository": "https://github.com/user/my-plugin",
  "license": "MIT"
}

• name: unique identifier, becomes skill namespace prefix
• version: semantic versioning (MAJOR.MINOR.PATCH)
• Only name and description are required

Namespacing

Plugin skills are prefixed: /my-plugin:hello, /my-plugin:review. This prevents conflicts between plugins with same skill names.

Plugin Components

Skills in plugins

Same as standalone skills but in plugin-root/skills/. Each needs SKILL.md with frontmatter.

Agents in plugins

Same Markdown format as .claude/agents/. Placed in plugin-root/agents/.

Hooks in plugins

Create hooks/hooks.json — same format as hooks in settings.json:

{
  "hooks": {
    "PostToolUse": [{
      "matcher": "Write|Edit",
      "hooks": [{ "type": "command", "command": "jq -r '.tool_input.file_path' | xargs prettier --write" }]
    }]
  }
}

MCP servers in plugins

Configure in .mcp.json at plugin root. Use ${CLAUDE_PLUGIN_ROOT} for relative paths:

{
  "my-server": {
    "command": "${CLAUDE_PLUGIN_ROOT}/servers/my-server",
    "args": ["--config", "${CLAUDE_PLUGIN_ROOT}/config.json"],
    "env": { "API_KEY": "${API_KEY}" }
  }
}

Default settings

settings.json at plugin root — currently only agent key supported:

{ "agent": "security-reviewer" }

This activates the named agent as main thread when plugin is enabled.

Development & Testing

# Test locally (no install needed)
claude --plugin-dir ./my-plugin

# Multiple plugins
claude --plugin-dir ./plugin-one --plugin-dir ./plugin-two

# Verify components
/help                    # Skills listed under namespace
/agents                  # Agents visible
/mcp                     # MCP servers running

Changes require restart to take effect.

Distribution

1. Version control — push to Git repository
2. Plugin marketplace — submit via claude.ai/settings/plugins/submit
3. Team sharing — configure team marketplace in .claude/plugins.json

When Plugin vs Standalone

Tip: Start standalone in .claude/, convert to plugin when ready to share.

Anti-Patterns

Checklist

• .claude-plugin/plugin.json with name, description, version
• Components at plugin root (not inside.claude-plugin/)
• Skills have proper SKILL.md with frontmatter
• MCP configs use ${CLAUDE_PLUGIN_ROOT} for paths
• Tested with --plugin-dir
• README.md with installation instructions
• Semantic versioning

Reference

• Full manifest schema
• Directory structure details