Claude Code Plugin Builder

Quick Reference (30 seconds)

Plugin Development Essentials - Build Claude Code plugins with correct structure, components, and best practices.

Directory Structure:

• .claude-plugin/plugin.json - Plugin manifest (REQUIRED)
• commands/ - Slash commands at plugin root
• agents/ - Custom agents at plugin root
• skills/ - Agent skills at plugin root
• hooks/ - Event handlers at plugin root
• .mcp.json - MCP server configuration
• .lsp.json - LSP server configuration

Critical Constraint: Component directories (commands, agents, skills, hooks) MUST be at plugin root level, NOT inside .claude-plugin/.

When to Use:

• Creating new Claude Code plugins
• Defining plugin components (commands, agents, skills, hooks)
• Configuring MCP or LSP servers for plugins
• Troubleshooting plugin loading issues
• Migrating standalone configurations to plugin format

---

Implementation Guide

Plugin Directory Structure

Correct Plugin Layout:

my-plugin/
  .claude-plugin/
    plugin.json          # Required: Plugin metadata only
  commands/              # At root level
    my-command.md
  agents/                # At root level
    my-agent.md
  skills/                # At root level
    my-skill/
      SKILL.md
  hooks/                 # At root level
    hooks.json
  .mcp.json              # At root level
  .lsp.json              # At root level
  LICENSE
  CHANGELOG.md
  README.md

Common Mistake to Avoid: Never place component directories inside .claude-plugin folder.

plugin.json Schema

Minimal Configuration:

{
  "name": "my-plugin"
}

Complete Configuration:

{
  "name": "my-plugin",
  "version": "1.0.0",
  "description": "Plugin description explaining purpose",
  "author": {
    "name": "Author Name",
    "email": "author@example.com",
    "url": "https://example.com"
  },
  "homepage": "https://github.com/user/my-plugin",
  "repository": "https://github.com/user/my-plugin",
  "license": "MIT",
  "keywords": ["claude-code", "automation"],
  "commands": ["./commands"],
  "agents": ["./agents"],
  "skills": ["./skills"],
  "hooks": ["./hooks/hooks.json"],
  "mcpServers": ["./.mcp.json"],
  "lspServers": ["./.lsp.json"],
  "outputStyles": ["./output-styles"]
}

Required Fields:

• name: Kebab-case unique identifier (letters, numbers, hyphens only)

Optional Fields:

• version: Semantic versioning (MAJOR.MINOR.PATCH)
• description: Plugin purpose explanation
• author: Object containing name, email, url
• homepage, repository, license, keywords
• Component path references (commands, agents, skills, hooks)
• Server configurations (mcpServers, lspServers)
• Output style references (outputStyles)

Path Configuration Rules

Path Format Requirements:

• All paths must be relative to plugin root
• Paths must start with ./
• Arrays supported for multiple paths
• Default directories are additive (not replaced)

Environment Variables:

• ${CLAUDE_PLUGIN_ROOT} - Plugin installation directory
• ${CLAUDE_PROJECT_DIR} - Current project directory

Example Path Usage:

{
  "commands": ["./commands", "./extra-commands"],
  "hooks": ["./hooks/main.json", "./hooks/validation.json"]
}

Slash Commands

Command File Structure (commands/my-command.md):

---
description: Command description for discovery
---

Command instructions and prompt content.

Arguments: $ARGUMENTS (all), $1, $2 (positional)
File references: @path/to/file.md

Frontmatter Fields:

• description (required): Command purpose for help display

Argument Handling:

• $ARGUMENTS - All arguments as single string
• $1, $2, $3 - Individual positional arguments
• @file.md - File content injection

Command Namespacing: Commands accessed as /plugin-name:command-name

Custom Agents

Agent File Structure (agents/my-agent.md):

---
name: my-agent
description: Agent purpose and capabilities
tools: Read, Write, Edit, Grep, Glob, Bash
model: sonnet
permissionMode: default
skills:
  - skill-name-one
  - skill-name-two
---

Agent system prompt and instructions.

Frontmatter Fields:

• name (required): Agent identifier
• description: Agent purpose
• tools: Comma-separated tool list
• model: sonnet, opus, haiku, inherit
• permissionMode: default, bypassPermissions, plan, passthrough
• skills: Array of skill names to load

Available Tools:

• Read, Write, Edit - File operations
• Grep, Glob - Search operations
• Bash - Command execution
• WebFetch, WebSearch - Web access
• Task - Sub-agent delegation
• TodoWrite - Task management

Agent Skills

Skill Structure (skills/my-skill/SKILL.md):

---
name: my-skill
description: Skill purpose and when to use
allowed-tools: Read, Grep, Glob
---

# Skill Name

## Quick Reference (30 seconds)

Brief overview and key concepts.

## Implementation Guide

Detailed implementation patterns.

## Advanced Patterns

Expert-level knowledge.

Skill Discovery: Model-invoked based on context relevance. Skills load automatically when task context matches skill description.

Hooks Configuration

hooks.json Structure:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Write",
        "hooks": [
          {
            "type": "command",
            "command": "bash ./hooks/validate-write.sh"
          }
        ]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "prompt",
            "prompt": "Verify operation completed successfully"
          }
        ]
      }
    ]
  }
}

Available Hook Events:

• PreToolUse - Before tool execution
• PostToolUse - After tool execution
• PermissionRequest - Permission dialog display
• UserPromptSubmit - User message submission
• SessionStart - Session initialization
• SessionEnd - Session termination
• Stop - Execution interruption
• SubagentStop - Sub-agent completion
• Notification - Notification trigger
• PreCompact - Before context compaction

Hook Types:

• command: Execute bash command
• prompt: Send LLM prompt

Matcher Patterns:

• Exact tool name: "Write", "Bash"
• Wildcard: "*" matches all tools
• Tool-specific filtering based on tool name

MCP Server Configuration

.mcp.json Structure:

{
  "mcpServers": {
    "my-server": {
      "command": "npx",
      "args": ["-y", "@my-org/mcp-server"],
      "env": {
        "API_KEY": "${API_KEY}"
      }
    }
  }
}

Transport Types:

• stdio: Standard input/output communication
• http: HTTP-based transport
• sse: Server-sent events transport

Fields:

• command: Executable command
• args: Command arguments array
• env: Environment variables
• type: Transport type (default: stdio)
• url: Server URL (for http/sse)

LSP Server Configuration

.lsp.json Structure:

{
  "lspServers": {
    "python": {
      "command": "pylsp",
      "args": [],
      "extensionToLanguage": {
        ".py": "python",
        ".pyi": "python"
      },
      "env": {
        "PYTHONPATH": "${CLAUDE_PROJECT_DIR}"
      }
    }
  }
}

Required Fields:

• command: LSP server executable
• extensionToLanguage: File extension to language mapping

Optional Fields:

• args: Command arguments
• env: Environment variables

---

Advanced Patterns

Development Workflow

Local Development:

# Test single plugin
claude --plugin-dir ./my-plugin

# Test multiple plugins
claude --plugin-dir ./plugin-one --plugin-dir ./plugin-two

Testing Components:

• Commands: /plugin-name:command-name invocation
• Agents: /agents to list, then invoke by name
• Skills: Ask questions relevant to skill domain
• Hooks: Trigger events and check debug logs

Debugging:

# Enable debug mode
claude --debug

# Validate plugin structure
claude plugin validate

# View plugin errors
/plugin errors

Security Best Practices

Path Security:

• Always use ${CLAUDE_PLUGIN_ROOT} for plugin-relative paths
• Never hardcode absolute paths
• Validate all inputs in hook scripts
• Prevent path traversal attacks

Permission Guidelines:

• Apply least privilege for tool access
• Limit agent permissions to required operations
• Validate hook command inputs
• Sanitize environment variables

Distribution and Installation

Plugin Installation Scopes:

• user: ~/.claude/settings.json (personal)
• project: .claude/settings.json (team)
• local: .claude/settings.local.json (developer)

CLI Commands:

# Plugin management
claude plugin install <plugin-name>
claude plugin uninstall <plugin-name>
claude plugin list
claude plugin enable <plugin-name>
claude plugin disable <plugin-name>
claude plugin update <plugin-name>

# Marketplace
claude plugin marketplace add <url>
claude plugin marketplace list

marketplace.json Structure:

{
  "name": "my-plugin",
  "versions": {
    "1.0.0": {
      "url": "https://github.com/user/plugin/releases/v1.0.0",
      "checksum": "sha256:..."
    }
  }
}

---

Troubleshooting

Common Issues:

Plugin Not Loading:

• Verify .claude-plugin/plugin.json exists
• Check plugin.json syntax validity
• Confirm name field uses kebab-case
• Ensure component directories at root level

Commands Not Found:

• Check command file has .md extension
• Verify YAML frontmatter with description
• Confirm commands path in plugin.json
• Test with /plugin-name:command-name

Hooks Not Triggering:

• Verify hooks.json syntax
• Check matcher pattern matches tool name
• Confirm hook command executable
• Enable debug mode for detailed logs

MCP Server Failures:

• Verify command exists in PATH
• Check environment variables set correctly
• Confirm transport type matches server
• Test server independently first

---

Works Well With

• moai-foundation-claude - Claude Code configuration and patterns
• moai-foundation-core - Core development workflows
• moai-workflow-project - Project initialization
• moai-domain-backend - Backend plugin development
• moai-domain-frontend - Frontend plugin development

---

Reference Files

Extended Documentation:

• Templates Reference - Complete plugin templates
• Migration Guide - Converting standalone configs
• Examples - Working plugin examples
• Validation - Plugin validation rules

---

Status: Production Ready Last Updated: 2025-12-25 Maintained by: MoAI-ADK Team