Claude Code¶

Claude Code is Anthropic's official CLI for Claude with plugin and sub-agent support.

Plugin Structure¶

Claude Code plugins use a specific structure with .claude-plugin/plugin.json as the manifest:

my-plugin/
├── .claude-plugin/
│   └── plugin.json          # Required: Plugin manifest
├── CLAUDE.md                # Optional: Context for the model
├── commands/                # Optional: Slash commands (Markdown)
│   └── build.md
├── skills/                  # Optional: Reusable skills (Markdown)
│   └── review/
│       └── SKILL.md
├── agents/                  # Optional: Sub-agent definitions (Markdown)
│   └── scanner.md
└── hooks/                   # Optional: Lifecycle hooks
    └── hooks.json

Plugin Manifest¶

The .claude-plugin/plugin.json file defines your plugin:

{
  "name": "my-plugin",
  "version": "1.0.0",
  "description": "A helpful plugin for developers",
  "author": {
    "name": "Your Name",
    "url": "https://github.com/yourname"
  },
  "homepage": "https://github.com/yourname/my-plugin",
  "repository": "https://github.com/yourname/my-plugin",
  "license": "MIT",
  "keywords": ["automation", "development"],
  "commands": "./commands/",
  "skills": "./skills/",
  "agents": "./agents/",
  "hooks": "./hooks/hooks.json"
}

Manifest Fields¶

Field	Description
`name`	Unique plugin name
`version`	Semantic version
`description`	Plugin description
`author`	Author information
`commands`	Path to commands directory
`skills`	Path to skills directory
`agents`	Path to agents directory
`hooks`	Path to hooks configuration

Commands¶

Commands are Markdown files with YAML frontmatter:

---
name: build
description: Build the project
allowed_tools:
  - Bash
  - Read
---

Build the project using the appropriate build system.
Detect the project type and run the correct command.

Command with Arguments¶

---
name: release
description: Execute release workflow for $ARGUMENTS
---

# Release

Execute the complete release workflow for the specified version.

## Usage

/release v1.2.3

Skills¶

Skills are reusable capabilities in skills/skill-name/SKILL.md:

---
name: code-review
description: Review code for best practices
---

# Code Review Skill

Review the provided code for:

- Security vulnerabilities
- Performance issues
- Code style and maintainability
- Error handling

Agents¶

Agents are sub-agent definitions that can be spawned via the Task tool:

---
name: security-scanner
description: Scan code for vulnerabilities. Use when reviewing security.
model: sonnet
tools: Read, Grep, Glob, Bash
skills: code-review
---

# Security Scanner Agent

You are a security expert specializing in code review.

## Your Responsibilities

1. Check for hardcoded secrets
2. Review authentication implementations
3. Identify injection vulnerabilities
4. Assess data validation

Agent Fields¶

Field	Description
`name`	Agent identifier
`description`	When to use this agent
`model`	haiku, sonnet, or opus
`tools`	Comma-separated tool list
`skills`	Comma-separated skill dependencies

Sub-Agent System¶

Claude Code's Task tool enables spawning specialized sub-agents:

Task(subagent_type="Plan", prompt="Design the implementation")
Task(subagent_type="Explore", prompt="Find all API endpoints")

Built-in Sub-Agent Types¶

Type	Description
`Bash`	Command execution specialist
`general-purpose`	Multi-step task handling
`Explore`	Fast codebase exploration
`Plan`	Implementation planning

Custom Sub-Agents¶

Define custom agents in agents/ directory and reference by name:

Task(subagent_type="security-scanner", prompt="Audit the auth module")

Parallel Execution¶

Multiple sub-agents can run in parallel:

> Run security-scanner and qa-agent in parallel to review this code

Installation Methods¶

From Local Path¶

claude plugin add /path/to/my-plugin

From GitHub¶

claude plugin add github:owner/repo/path/to/plugin

Plugin Management¶

# List installed plugins
claude plugin list

# Remove a plugin
claude plugin remove plugin-name

# Update a plugin
claude plugin update plugin-name

Context File (CLAUDE.md)¶

The CLAUDE.md file provides persistent context:

# My Plugin

This plugin automates release workflows.

## Available Commands

- `/release <version>` - Execute release
- `/check` - Run validation

## Dependencies

- `git` - Version control
- `gh` - GitHub CLI

MCP Server Configuration¶

Global Config¶

~/.claude.json:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@anthropic-ai/mcp-server-filesystem"]
    }
  }
}

Per-Project Config¶

.claude/settings.json:

{
  "mcpServers": {
    "custom-server": {
      "command": "./server.js",
      "args": ["--config", "config.json"]
    }
  }
}

Hooks¶

Plugins can intercept lifecycle events:

{
  "pre_prompt_submit": [
    {
      "command": "scripts/validate.sh",
      "description": "Validate before submission"
    }
  ],
  "post_tool_execution": [
    {
      "command": "scripts/log-tool.sh",
      "description": "Log tool usage"
    }
  ]
}

Available Tools¶

Tool	Description
`Read`	Read file contents
`Write`	Write/create files
`Edit`	Edit files in place
`Bash`	Execute shell commands
`Glob`	Find files by pattern
`Grep`	Search file contents
`Task`	Spawn sub-agents
`WebFetch`	Fetch web content
`WebSearch`	Search the web
`TodoWrite`	Manage task lists

Models¶

Model	Description	Use Case
`haiku`	Fast, lightweight	Quick tasks, simple queries
`sonnet`	Balanced performance	Most development tasks
`opus`	Most capable	Complex reasoning, architecture

Converting from Canonical¶

aiassistkit provides adapters for format conversion:

import (
    "github.com/agentplexus/assistantkit/commands/core"
    "github.com/agentplexus/assistantkit/commands/claude"
)

canonical := &core.Command{
    Name:        "build",
    Description: "Build the project",
    Prompt:      "Build the project using the appropriate build system.",
}

// Convert to Claude Markdown format
adapter := &claude.Adapter{}
mdBytes, err := adapter.Marshal(canonical)

Comparison with Other Platforms¶

Feature	Claude Code	Gemini CLI	Kiro CLI
Manifest	JSON	JSON	JSON
Commands	Markdown	TOML	N/A
Skills	Markdown	N/A	N/A
Agents	Markdown	N/A	JSON
Sub-agents	Yes (Task tool)	No	Yes
Context	CLAUDE.md	GEMINI.md	Steering files
MCP Support	Yes	Yes	Yes
Marketplace	PR submission	Auto-indexed	Manual

Plugin Marketplace¶

Claude plugins can be submitted to the official marketplace via PR to anthropics/claude-plugins-official.

See Claude Marketplace for detailed submission instructions.