Claude Code

Skills

Skills are reusable prompt files that teach Claude how to perform a specific task. They live as SKILL.md files and are invoked via slash commands.

Docs

File Locations

Scope	Path
Personal (all projects)	`~/.claude/skills/<name>/SKILL.md`
Project	`.claude/skills/<name>/SKILL.md`

SKILL.md Structure

.claude/skills/deploy/SKILL.md

---
name: deploy
description: Deploy the app to production
disable-model-invocation: true # only user can invoke, not Claude automatically
allowed-tools: Bash(npm *) # tools Claude can use without asking
---

Instructions for Claude go here. This is plain markdown.

Frontmatter Fields

Field	Description
`name`	Slug used as `/name` slash command
`description`	Shown in `/help`; also used by Claude to decide when to auto-invoke
`disable-model-invocation: true`	Prevents Claude from invoking the skill automatically
`user-invocable: false`	Hides from slash commands; Claude-only
`allowed-tools`	Tools usable without per-call approval (e.g. `Read, Grep, Bash(git *)`)
`argument-hint`	Text shown in autocomplete, e.g. `[filename] [style]`
`context: fork`	Run skill in an isolated subagent
`agent`	Subagent type to use when `context: fork` is set

String Substitutions

Variable	Value
`$ARGUMENTS`	All arguments passed to the slash command
`$0`, `$1`, `$2`	Positional arguments
`${CLAUDE_SKILL_DIR}`	Absolute path to the skill's directory
`${CLAUDE_SESSION_ID}`	Current session ID

Dynamic Context Injection

Use !`command` to run a shell command and inline the output before Claude reads the skill:

## Changes to review

!`git diff HEAD~1`

Check for security issues and performance problems.

Invocation

/skill-name
/skill-name arg1 arg2

Claude can also invoke skills automatically when the task matches the skill's description, unless disable-model-invocation: true is set.

Plugins

Plugins are packaged collections of skills, agents, hooks, and MCP servers — distributed and versioned for sharing across projects and teams.

Docs

Structure

my-plugin/
├── .claude-plugin/
│   └── plugin.json     # required manifest
├── skills/
│   └── my-skill/
│       └── SKILL.md
├── agents/
├── hooks/
│   └── hooks.json
└── .mcp.json

Manifest (plugin.json)

.claude-plugin/plugin.json

{
  "name": "my-plugin",
  "description": "What this plugin does",
  "version": "1.0.0",
  "author": { "name": "Your Name" }
}

Required fields: name, description, version. The name becomes the namespace prefix for all skills.

Invoking Plugin Skills

Plugin skills are namespaced to avoid conflicts:

/my-plugin:skill-name

Installation

# Browse and install via the plugin UI
/plugin

# Load a local plugin during development
claude --plugin-dir ./my-plugin

Skills vs. Plugins

	Skills	Plugins
Slash command	`/skill-name`	`/plugin-name:skill-name`
Stored in	`.claude/skills/`	Directory with `plugin.json`
Versioned	No	Yes (semver)
Can bundle MCP servers	No	Yes
Best for	Single-project extensions	Sharing across teams