Plugins & Skills

Two extension mechanisms: Plugins for full CLI extensions, and Skills for lightweight AI prompt templates. For how plugin tools and skills feed into the tool pipeline, see Tool Sources.

Skills

Skills are the simpler extension — markdown files that inject prompt content when invoked as slash commands.

Skill Format

---
name: my-skill
description: What this skill does
context: inline   # or 'fork' to run in a sub-agent
effort: balanced
paths: ["src/**"] # Glob patterns for auto-activation
---

Your skill prompt content here.
This becomes the instruction sent to the model.

How Skills Work

📋

User types /skill-name

📥

Load skill file

Context type?

inline

Inject content as user message

🤖

Model processes in current conversation

fork

Spawn sub-agent

🤖

Sub-agent processes independently

📤

Result returned to main conversation

Skill Sources

1. Bundled skills — shipped with Claude Code in src/skills/bundled/
2. User skills — stored in ~/.claude/skills/
3. Project skills — in .claude/skills/ within a repo

Bundled Skills

Claude Code ships with several built-in skills including:

• /commit — create a git commit
• /review-pr — review a pull request
• /simplify — simplify recently changed code
• And more...

These are PromptCommands — they inject specialized instructions into the conversation.

Plugins

Plugins are full CLI extensions that can add commands, tools, and hooks.

Plugin Manifest

{
  "name": "my-plugin",
  "version": "1.0.0",
  "commands": [
    { "name": "my-command", "description": "...", "handler": "./commands/my-command.js" }
  ],
  "tools": [
    { "name": "MyTool", "handler": "./tools/my-tool.js" }
  ],
  "hooks": [
    { "event": "pre_tool_use", "handler": "./hooks/validate.js" }
  ]
}

Plugin Capabilities

Feature	Description
Commands	Custom slash commands (any type)
Tools	Custom tools available to the model
Hooks	Lifecycle hooks (pre/post query, pre/post tool use)
Settings	Custom configuration options
Permissions	Custom permission rules

Plugin Lifecycle

🚀

CLI Startup

🔍

Discover plugins in ~/.claude/plugins/

📥

Load and validate manifests

Invalid — log warning, skip plugin

Valid — register capabilities

🔌

Register commands, tools, hooks

✅

Plugin active — responding to events

Command Invoked

⌨️

User runs /plugin-cmd

Handler executes

Tool Called

🔧

Model calls plugin tool

Tool handler runs

Hook Fires

🪝

pre/post tool or query

Hook logic runs

Plugin vs Skill

Aspect	Skill	Plugin
Complexity	Low (just markdown)	High (JavaScript/TypeScript)
What it can do	Inject prompts	Commands, tools, hooks, settings
Format	YAML frontmatter + markdown	JSON manifest + JS handlers
Installation	Drop a file in ~/.claude/skills/	Install in ~/.claude/plugins/
Use case	Prompt templates, workflows	Full-featured extensions

Key Source Files

File	Purpose
src/skills/	Skill system (20 files)
src/skills/bundled/	Built-in skills
src/plugins/	Plugin system
src/services/plugins/	Plugin discovery and management
src/tools/SkillTool/	Skill execution tool