Claude MPM Agent Templates

A curated collection of specialized AI agent templates for Claude MPM - the multi-agent orchestration framework for Claude Code.

Key Features:

• 🏗️ BASE-AGENT.md Inheritance System - Convention-based template inheritance eliminates duplication
• 📦 Hierarchical Organization - Agents organized by functional relationships (engineer, qa, ops, etc.)
• 🤖 Auto-Deployment - Intelligent agent selection based on project type detection
• 🔍 Agent Discovery - Cache scanning with semantic matching for recommendations
• ⚡ 57% Less Duplication - Shared instructions managed in one place

Overview

This repository contains production-ready agent templates in Markdown format with YAML frontmatter. Each agent is designed for specific engineering, operations, or quality assurance tasks within the Claude MPM framework.

The repository uses a hierarchical organization with BASE-AGENT.md inheritance to eliminate duplication and maintain consistency across related agents. When building an agent, content from BASE-AGENT.md files at each directory level is automatically combined, allowing shared instructions to be defined once and inherited by all agents in that category.

Repository Structure

claude-mpm-agents/
├── agents/                          # Deployable agent templates
│   ├── BASE-AGENT.md               # Universal instructions (ALL agents inherit)
│   │
│   ├── claude-mpm/                 # Claude MPM framework agents
│   │   ├── BASE-AGENT.md          # MPM framework awareness
│   │   └── mpm-agent-manager.md   # Agent discovery & deployment
│   │
│   ├── universal/                  # Cross-cutting concerns
│   │   ├── memory-manager.md
│   │   ├── research.md
│   │   └── ...
│   │
│   ├── engineer/                   # Implementation specialists
│   │   ├── BASE-AGENT.md          # Shared engineering principles
│   │   ├── core/
│   │   │   └── engineer.md
│   │   ├── frontend/
│   │   │   ├── react-engineer.md
│   │   │   ├── nextjs-engineer.md
│   │   │   └── ...
│   │   ├── backend/
│   │   │   ├── python-engineer.md
│   │   │   ├── golang-engineer.md
│   │   │   └── ...
│   │   └── ...
│   │
│   ├── qa/                         # Quality assurance
│   │   ├── BASE-AGENT.md          # Shared QA standards
│   │   ├── qa.md
│   │   ├── api-qa.md
│   │   └── web-qa.md
│   │
│   └── ops/                        # Operations & deployment
│       ├── BASE-AGENT.md          # Shared ops protocols
│       ├── core/
│       │   └── ops.md
│       ├── platform/
│       │   ├── vercel-ops.md
│       │   └── ...
│       └── ...
│
├── templates/                       # Reference materials (NOT agents)
│   ├── circuit-breakers.md        # PM violation patterns
│   ├── pm-examples.md             # PM behavior examples
│   ├── validation-templates.md    # Verification templates
│   └── ...
│
├── build-agent.py                   # Build script for flattening agents
└── AUTO-DEPLOY-INDEX.md            # Project type detection rules

BASE-AGENT.md Inheritance System

This repository uses a convention-based inheritance system to reduce duplication and maintain consistency.

How It Works

When building an agent, content is combined in this order:

1. Agent-specific content ({agent-name}.md) - Unique instructions for this agent
2. Directory BASE-AGENT.md (if exists) - Shared instructions for agents in this directory
3. Parent BASE-AGENT.md files (recursive) - Instructions from parent directories
4. Root BASE-AGENT.md (always) - Universal instructions for ALL agents

Example: For agents/engineer/backend/python-engineer.md:

python-engineer.md
  + agents/engineer/backend/BASE-AGENT.md (if exists)
  + agents/engineer/BASE-AGENT.md
  + agents/BASE-AGENT.md

BASE-AGENT.md Levels

Root Level (agents/BASE-AGENT.md)

Universal instructions for ALL agents:

• Git workflow standards
• Memory routing protocols
• Output format standards
• Handoff protocols
• Quality standards

Category Level (agents/engineer/BASE-AGENT.md, agents/qa/BASE-AGENT.md, etc.)

Shared instructions for all agents in a category:

• Engineer: SOLID principles, code reduction, type safety, testing requirements
• QA: Testing strategies, coverage standards, memory-efficient testing
• Ops: Deployment verification, security scanning, monitoring requirements

Subcategory Level (e.g., agents/engineer/frontend/BASE-AGENT.md)

Specific instructions for subcategories (optional, as needed)

Building Flattened Agents

Agents in this repository are modular (source files) and must be flattened before use:

# Build a single agent
./build-agent.py agents/engineer/backend/python-engineer.md

# Build all agents
./build-agent.py --all

# Build to specific output directory
./build-agent.py --all --output-dir dist/agents

# Validate all agent definitions
./build-agent.py --validate

Built agents are placed in dist/agents/ with the full inheritance chain combined.

Why This Approach?

Benefits:

• ✅ DRY (Don't Repeat Yourself): Common instructions defined once
• ✅ Consistency: All agents in a category follow same standards
• ✅ Maintainability: Update shared instructions in one place
• ✅ Modularity: Agent-specific content stays focused
• ✅ Flexibility: Add BASE-AGENT.md at any level as needed

Example: All engineering agents inherit:

• Root: Git workflows, output standards (200 lines)
• Engineer: SOLID principles, code reduction (300 lines)
• Agent: Python-specific patterns (150 lines)

Total: 650 lines, but only 150 lines are Python-specific!

Agent Categories

Claude MPM Framework Agents

Framework-level agents with MPM-specific awareness:

• mpm-agent-manager - Agent discovery, cache scanning, intelligent recommendations, and deployment orchestration

Universal Agents

Cross-cutting concerns that apply to multiple domains:

• memory-manager - Project-specific memory management
• product-owner - Product strategy and prioritization
• project-organizer - Project organization and workflow
• research - Codebase investigation and analysis
• code-analyzer - Code review and pattern identification
• content-agent - Content optimization and SEO

Engineering Agents

Core

• engineer - General-purpose software engineering

Frontend

• web-ui - Generic web UI development
• react-engineer - React with hooks and modern patterns
• nextjs-engineer - Next.js App Router and Server Components
• svelte-engineer - Svelte 5 with Runes and SvelteKit

Backend

• python-engineer - Python 3.12+ with type safety and async
• golang-engineer - Go with concurrency patterns
• phoenix-engineer - Elixir/Phoenix with LiveView, Ecto, and OTP foundations
• java-engineer - Java/Spring Boot with hexagonal architecture
• ruby-engineer - Ruby/Rails with service objects
• rust-engineer - Rust with ownership and safety
• php-engineer - PHP/Laravel with strict types
• javascript-engineer - Node.js/Express backend

Mobile

• dart-engineer - Dart/Flutter with clean architecture
• tauri-engineer - Tauri desktop applications

Data

• data-engineer - Data pipelines and ETL
• typescript-engineer - TypeScript 5.6+ with branded types

Specialized

• refactoring-engineer - Safe code refactoring
• agentic-coder-optimizer - Build system optimization
• imagemagick - Image optimization specialist
• prompt-engineer - Prompt optimization for Claude 4.5

Quality Assurance Agents

• qa - General quality assurance
• api-qa - API testing and validation
• web-qa - Web application testing with Playwright

Operations Agents

Core

• ops - General DevOps and infrastructure

Platform

• vercel-ops - Vercel deployment
• gcp-ops - Google Cloud Platform
• clerk-ops - Clerk authentication
• local-ops - Local development with PM2/Docker

Tooling

• version-control - Git operations and release coordination

Security

• security - Security analysis and vulnerability assessment

Documentation

• documentation - Technical documentation and API docs
• ticketing - Ticket management and tracking

Template Format

Each agent template uses Markdown with YAML frontmatter:

---
name: agent-name
description: Agent purpose and capabilities
agent_id: unique-identifier
agent_type: engineer|qa|ops|universal|documentation
model: sonnet|opus|haiku
version: 2.0.0
tags:
  - technology
  - domain
category: engineering|qa|ops|research
---

# Agent Name

Agent-specific instructions...

<!-- BASE-AGENT.md content automatically appended during build -->

Usage

With Claude MPM

These templates are designed to be used with Claude MPM:

1. Build agents (flattened with inheritance):

   ./build-agent.py --all --output-dir ~/.claude-mpm/agents

2. Deploy agents via Claude MPM CLI:

   claude-mpm agents deploy <agent-name>

3. Use in Claude Code via delegation:

   PM delegates to engineer agent for implementation tasks
   

Customizing Templates

For Project-Specific Needs

1. Fork this repository
2. Modify BASE-AGENT.md files for project standards
3. Add custom agents as needed
4. Build and deploy to your Claude MPM instance

For Temporary Changes

1. Edit agent files in agents/ directory
2. Rebuild with ./build-agent.py --all
3. Deploy updated agents

Template Schema

All templates follow the Claude MPM Agent Template Schema v1.3.0:

• Metadata: Name, version, description, tags, category
• Configuration: Model selection, resource limits, timeouts
• Capabilities: Memory limits, CPU allocation, network access
• Dependencies: Python packages, system requirements
• Skills: Reusable skill references
• Knowledge: Domain expertise, best practices, constraints
• Interactions: Input/output formats, handoff patterns
• Memory Routing: Memory categorization and keywords

Development

Building Agents

# Install Python 3.9+ (no additional dependencies required)

# Build single agent
./build-agent.py agents/engineer/backend/python-engineer.md

# Build all agents
./build-agent.py --all

# Validate all agents
./build-agent.py --validate

# Custom output directory
./build-agent.py --all --output-dir ~/my-agents

Adding New Agents

1. Determine category: universal, engineer, qa, ops, etc.
2. Create agent file: agents/{category}/{subcategory}/{agent-name}.md
3. Add YAML frontmatter: name, description, agent_id, etc.
4. Write agent-specific content: Focus on unique instructions
5. Build and validate: ./build-agent.py --validate

Modifying BASE-AGENT.md Files

When updating shared instructions:

1. Edit appropriate BASE-AGENT.md:

   • Root level: Universal changes
   • Category level: Category-wide changes
   • Subcategory level: Specific subcategory changes

2. Rebuild all affected agents:

   ./build-agent.py --all

3. Validate changes:

   ./build-agent.py --validate

Version History

See CHANGELOG.md for detailed version history.

Contributing

See CONTRIBUTING.md for development guidelines.

License

Part of the Claude MPM project. See main repository for license details.

Related

• Claude MPM: https://github.com/bobmatnyc/claude-mpm
• Documentation: https://docs.claude.com/en/docs/claude-code
• Community: Claude MPM GitHub Discussions

---

Repository Structure: Agent templates with BASE-AGENT.md inheritance for consistency and maintainability.