Episode 9: Plugin Anatomy - What's Inside a Plugin — Comprehensive Guide to Claude Code Extension Ecosystem

A Plugin is the final form of a Claude Code extension. It's not a single feature, but a "combo meal" that bundles multiple components together for distribution.

The Core Manifest: plugin.json

Every Plugin must have a plugin.json in its root (or in the .claude-plugin/ directory).

{
  "name": "my-awesome-plugin",
  "version": "1.0.0",
  "description": "This is my first plugin",
  "hooks": { ... },       // Define Hooks
  "commands": [ ... ],    // Define Commands
  "keywords": [ ... ]     // Keywords
}

Component Matrix

Different plugins have different focuses. The table below shows the composition of several typical plugins:

Plugin	Skills	Hooks	MCP	Commands	Key Feature
GSD	137	9	0	0	Pure instruction-driven, extreme engineering
caveman	5	2	0	3	Ultra-compressed mode, extreme token saving
claude-mem	7	9	1	0	Background Worker service, long-term memory
claude-hud	0	0	0	2	UI status bar enhancement, compiled project

Typical File Structure

Taking caveman as an example, a typical plugin directory looks like this:

caveman/
├── .claude-plugin/
│   └── plugin.json          # Manifest (defines entry points and Hooks)
├── hooks/
│   ├── activate.js          # SessionStart Hook
│   └── tracker.js           # Message tracking Hook
├── skills/
│   └── caveman/
│       └── SKILL.md         # /caveman slash command
├── commands/
│   └── caveman.toml         # Structured command definition
└── CLAUDE.md                # Plugin-level project rules

Workflow Overview

graph LR
    P[plugin.json] --> S[Skills]
    P --> H[Hooks]
    P --> M[MCP]
    P --> C[Commands]
    
    S -->|Defines| SC[Slash Commands]
    H -->|Responds to| LE[Lifecycle Events]
    M -->|Provides| ET[Extra Tool Capabilities]