A workflow-based multi-agent development system with a graphical UI. CMAT provides specialized AI agents orchestrated by customizable workflow templates with automated validation, comprehensive skills, and intelligent learning.
Version: 10.6.1
CMAT is a multi-agent system that breaks down software development into specialized roles, orchestrated by flexible workflow templates:
Core Agents:
- Requirements Analyst: Analyzes user needs and creates implementation plans
- Architect: Designs system architecture and technical specifications
- Implementer: Writes production-quality code
- Tester: Creates and runs comprehensive test suites
- Documenter: Maintains project documentation
- Code Reviewer: Reviews code for quality and security
Specialist Agents:
- Security Auditor: Security analysis, threat modeling, vulnerability scanning
- Performance Engineer: Profiling, optimization, load testing
- Data Engineer: Data modeling, ETL design, data quality
- DevOps Engineer: CI/CD, infrastructure-as-code, container orchestration
- Accessibility Specialist: WCAG compliance, screen reader testing
- API Integration Specialist: API integration, webhooks, rate limiting
- Refactoring Specialist: Code improvement, design patterns, tech debt
Integration Agents:
- GitHub Integration Coordinator: Syncs workflow with GitHub issues and PRs
- Atlassian Integration Coordinator: Syncs workflow with Jira and Confluence
Skills System: 38 specialized skills providing domain expertise, automatically injected into agent prompts.
- 17 Specialized Agents - Clear responsibilities, reusable across workflows
- Graphical UI - Full-featured tkinter interface for managing projects
- Dark Mode Support - 15 ttkbootstrap themes with live preview (View > Toggle Dark Mode or Ctrl+D)
- Workflow Templates - Define agent sequences, inputs, outputs, and transitions
- Output Validation - Automatic validation of required outputs
- Automated Workflows - Template-driven intelligent task chaining
- Task Queue System - Organize and track work
- Skills System - Domain expertise in reusable modules
- Multiple Theme Support - Light and dark themes with live preview
| Light Theme | Dark Theme |
|---|---|
 |
 |
- Learnings System - RAG-based memory that improves over time
- YAML Completion Blocks - Structured status reporting from agents
- Cost Tracking - Automatic token usage and cost tracking per task
- Model Management - Configure and track Claude model pricing
- GitHub Sync - Issues, PRs, and labels
- Jira/Confluence Sync - Tickets and documentation
- Python 3.10+ - Core runtime
- Claude Code - Multi-agent orchestration platform
- tkinter - GUI framework (included with Python)
- pyyaml - YAML parsing for agents and workflows
- pillow - Image processing for UI icons
- ttkbootstrap - Modern UI themes with dark mode support
Optional:
- Node.js 16+ - For MCP servers (GitHub/Jira integration)
# Clone the repository
git clone https://github.com/anthropics/claude-multi-agent-template.git
cd claude-multi-agent-template
# Run the setup script (recommended)
./setup_dev.sh # Linux/macOS
.\setup_dev.ps1 # Windows PowerShell
# Or manually:
# Create and activate virtual environment
python3 -m venv .venv
source .venv/bin/activate # Linux/macOS
.\.venv\Scripts\Activate.ps1 # Windows PowerShell
# Install in editable mode
pip install -e . # Core only (just to run CMAT)
pip install -e ".[dev]" # With dev tools (pytest, black, mypy, ruff)Note: Always activate the virtual environment before running CMAT:
source .venv/bin/activate # Linux/macOS
.\.venv\Scripts\Activate.ps1 # Windows PowerShell
cmatIDE Setup: If using PyCharm, VS Code, or another IDE, configure the Python interpreter to use .venv/bin/python in the project directory.
cmatFor a hands-on tutorial, it is recommended you follow DEMO.md before starting on your own use.
In the UI:
- File > Install... (or Ctrl+I)
- Select your project directory
- CMAT copies templates to
your-project/.claude/and connects to it
In the UI:
- File > New Enhancement
- Fill in the enhancement details
- Enhancement is created in
your-project/enhancements/
In the UI:
- Select a workflow template
- Select the enhancement
- Click Start Workflow
- Agents execute automatically
claude-multi-agent-template/
├── src/
│ ├── core/ # Core CMAT services
│ │ ├── __init__.py
│ │ ├── cmat.py # Main orchestration class
│ │ ├── models/ # Data models
│ │ ├── services/ # Service classes
│ │ └── utils.py
│ │
│ └── ui/ # Graphical UI
│ ├── __init__.py
│ ├── main.py # Entry point
│ ├── dialogs/ # UI dialogs
│ ├── models/ # UI models
│ └── utils/ # UI utilities
│
├── templates/ # Copied to target projects
│ ├── .claude/
│ │ ├── agents/ # Agent definitions
│ │ ├── skills/ # Skills system
│ │ ├── data/ # JSON data files
│ │ └── hooks/ # Automation hooks
│ └── enhancement-templates/ # Enhancement templates
│
├── tests/ # Test suite
├── docs/ # Documentation
├── demo/ # Demo calculator project
│
└── pyproject.toml # Package configuration
your-project/
├── .claude/
│ ├── agents/ # Agent definitions
│ ├── skills/ # Skills
│ ├── data/ # Queue, workflows, learnings
│ ├── hooks/ # Automation hooks
│ └── logs/ # Execution logs
├── enhancements/ # Feature requests
│ └── feature-name/
│ ├── feature-name.md # Enhancement spec
│ ├── requirements-analyst/
│ ├── architect/
│ ├── implementer/
│ ├── tester/
│ └── documenter/
└── [your project files]
CMAT / CMATInterface
├── queue: QueueService # Task state management
├── agents: AgentService # Agent registry
├── skills: SkillsService # Skills loading
├── workflow: WorkflowService # Orchestration
├── tasks: TaskService # Execution
├── learnings: LearningsService # RAG memory
├── models: ModelService # Model config & costs
└── tools: ToolsService # Tool definitions
Workflow Template
│
├─ Step 0: requirements-analyst
│ ├─ input: "enhancement spec"
│ ├─ required_output: "analysis.md"
│ └─ on_status:
│ ├─ READY_FOR_DEVELOPMENT → Step 1 (auto)
│ └─ BLOCKED → Stop
│
├─ Step 1: architect
│ └─ on_status:
│ ├─ READY_FOR_IMPLEMENTATION → Step 2 (auto)
│ └─ NEEDS_CLARIFICATION → Stop
│
└─ ... (continues through workflow)
Agents report status using YAML completion blocks:
---
agent: implementer
task_id: task_1234567890_12345
status: READY_FOR_TESTING
---Completion statuses (workflow continues):
READY_FOR_DEVELOPMENT- Requirements completeREADY_FOR_IMPLEMENTATION- Design completeREADY_FOR_TESTING- Implementation completeTESTING_COMPLETE- Tests passedDOCUMENTATION_COMPLETE- Docs updated
Halt statuses (workflow pauses):
BLOCKED: <reason>- Cannot proceedNEEDS_CLARIFICATION: <question>- Requires inputTESTS_FAILED: <details>- Tests need attention
- WORKFLOW_GUIDE.md - Workflow patterns
- SKILLS_GUIDE.md - Skills system
- QUEUE_SYSTEM_GUIDE.md - Task queue operations
- CUSTOMIZATION_GUIDE.md - Adapting to your project
- LEARNINGS_GUIDE.md - RAG memory system
- COST_TRACKING.md - Token usage and costs
- INTEGRATION_GUIDE.md - GitHub/Jira integration
# Install with dev dependencies (includes pytest, black, mypy, ruff)
pip install -e ".[dev]"
# Run all tests
pytest
# Run with coverage
pytest --cov=core --cov=ui# Create virtual environment
python -m venv .venv
source .venv/bin/activate # Linux/macOS
.\.venv\Scripts\Activate.ps1 # Windows PowerShell
# Install in development mode
pip install -e ".[dev]"- Try the Demo - DEMO.md - Hands-on walkthrough with the calculator project
- Customize - docs/CUSTOMIZATION_GUIDE.md - Adapt for your project
- Learn Workflows - docs/WORKFLOW_GUIDE.md - Workflow patterns
- Explore Skills - docs/SKILLS_GUIDE.md - Domain expertise
- Claude Code: https://claude.ai/code
- Complete Documentation: See
docs/directory
