A workflow-based multi-agent development system using Claude Code. This template provides specialized AI agents orchestrated by customizable workflow templates with automated validation, comprehensive skills, and intelligent learning.
Version: 8.2.0
This template provides a multi-agent system that breaks down software development into specialized roles, orchestrated by flexible workflow templates:
- 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
Integration Agents:
- GitHub Integration Coordinator: Syncs workflow with GitHub issues and PRs
- Atlassian Integration Coordinator: Syncs workflow with Jira and Confluence
Skills System:
- 14+ specialized skills providing domain expertise
- Automatically injected into agent prompts
- Organized by category (analysis, architecture, implementation, testing, documentation)
- 7 Specialized Agents - Clear responsibilities, reusable across workflows
- 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
- Flexible Orchestration - Same agent, different workflows, different behavior
- 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
- Comprehensive Logging - Agent execution logged for analysis
- Metadata Headers - All outputs are self-documenting and traceable
- GitHub Sync - Issues, PRs, and labels
- Jira/Confluence Sync - Tickets and documentation
# Copy .claude directory to your project
cp -r ClaudeMultiAgentTemplate/.claude /path/to/your/project/
# Verify installation
cd /path/to/your/project/.claude
python -m cmat version# List available workflows
python -m cmat queue status
# List agents
python -m cmat agents list
# View learnings (if any)
python -m cmat learnings list# Create enhancement spec
mkdir -p enhancements/my-feature
echo "# My Feature\n\nDescription of what to build..." > enhancements/my-feature/my-feature.md
# Use Claude Code to start a workflow via Task tool with workflow-related agentsSee CUSTOMIZATION.md for adapting to your project.
your-project/
├── .claude/ # Multi-agent system (v8.2.0)
│ ├── cmat/ # Python package
│ │ ├── __init__.py # Version and exports
│ │ ├── __main__.py # CLI entry point
│ │ ├── cmat.py # Main CMAT class
│ │ ├── models/ # Data models (Task, Agent, Learning, etc.)
│ │ ├── services/ # Service classes
│ │ └── utils.py # Utilities
│ ├── agents/ # Agent definitions
│ │ ├── *.md # Agent specifications
│ │ └── agents.json # Agent registry
│ ├── skills/ # Skills system
│ │ ├── skills.json # Skills registry
│ │ └── */SKILL.md # 14+ skills
│ ├── data/ # JSON data files
│ │ ├── task_queue.json # Task queue state
│ │ ├── workflow_templates.json # Workflow definitions
│ │ ├── learnings.json # RAG learnings storage
│ │ ├── models.json # Claude model definitions
│ │ └── tools.json # Tool definitions
│ ├── hooks/ # Automation hooks
│ │ └── on-session-end-cost.sh # Cost tracking hook
│ ├── docs/ # Documentation
│ └── tests/ # Python tests
├── enhancements/ # Feature requests
│ └── feature-name/
│ ├── feature-name.md # Enhancement spec
│ ├── requirements-analyst/
│ │ ├── required_output/
│ │ └── optional_output/
│ ├── architect/
│ │ └── ...
│ └── logs/
└── [your project files]
CMAT (entry point)
├── queue: QueueService # Task state management
├── agents: AgentService # Agent registry and generation
├── skills: SkillsService # Skills loading and prompt building
├── workflow: WorkflowService # Template management and orchestration
├── tasks: TaskService # Prompt building and execution
├── learnings: LearningsService # RAG memory system
└── models: ModelService # Model configuration and cost extraction
Workflow Template
│
├─ Step 0: requirements-analyst
│ ├─ input: "enhancement spec file"
│ ├─ required_output: "analysis.md"
│ └─ on_status:
│ ├─ READY_FOR_DEVELOPMENT → Step 1 (auto_chain)
│ └─ BLOCKED → Stop (halt status)
│
├─ Step 1: architect
│ ├─ input: "step 0 outputs"
│ ├─ required_output: "design.md"
│ └─ on_status:
│ ├─ READY_FOR_IMPLEMENTATION → Step 2 (auto_chain)
│ └─ NEEDS_CLARIFICATION → Stop (halt status)
│
└─ ... (continues through workflow)
Agents report status using YAML completion blocks:
---
agent: implementer
task_id: task_1234567890_12345
status: READY_FOR_TESTING
---pending → active → completed
→ failed
→ cancelled
python -m cmat <command> [options]
# Learnings (RAG memory system)
python -m cmat learnings list # List all learnings
python -m cmat learnings add "<content>" # Add a manual learning
python -m cmat learnings delete <id> # Delete a learning
python -m cmat learnings show <id> # Show learning details
python -m cmat learnings search "<query>" # Search learnings
# Queue management
python -m cmat queue status # Show queue summary
python -m cmat queue list [pending|active|completed|failed|all]
# Agents
python -m cmat agents list # List all agents
python -m cmat agents generate # Regenerate agents.json
# Models
python -m cmat models list # List Claude models
python -m cmat models show <id> # Show model details
python -m cmat models set-default <id> # Set default model
# Cost tracking
python -m cmat costs extract <task_id> <transcript_path> [session_id]
python -m cmat costs show <task_id> # Show task cost
python -m cmat costs enhancement <name> # Show enhancement cost
# Version
python -m cmat version # Show CMAT version1. requirements-analyst (Step 0)
Input: enhancements/feature/feature.md
Output: enhancements/feature/requirements-analyst/required_output/analysis_summary.md
Status: READY_FOR_DEVELOPMENT
→ Triggers: architect (Step 1)
2. architect (Step 1)
Input: enhancements/feature/requirements-analyst/required_output/
Output: enhancements/feature/architect/required_output/implementation_plan.md
Status: READY_FOR_IMPLEMENTATION
→ Triggers: implementer (Step 2)
3. implementer (Step 2)
Input: enhancements/feature/architect/required_output/
Output: enhancements/feature/implementer/required_output/implementation_summary.md
Status: READY_FOR_TESTING
→ Triggers: tester (Step 3)
4. tester (Step 3)
Input: enhancements/feature/implementer/required_output/
Output: enhancements/feature/tester/required_output/test_summary.md
Status: TESTING_COMPLETE
→ Triggers: documenter (Step 4)
5. documenter (Step 4)
Input: enhancements/feature/tester/required_output/
Output: enhancements/feature/documenter/required_output/documentation_summary.md
Status: DOCUMENTATION_COMPLETE
→ Workflow complete
Analysis: Requirements Elicitation, User Story Writing, Bug Triage Architecture: API Design, Architecture Patterns, Agent Design Implementation: Error Handling, Code Refactoring, SQL Development Testing: Test Design Patterns, Test Coverage Documentation: Technical Writing, API Documentation UI Design: Desktop UI, Web UI
# View skills assigned to agents
python -m cmat agents list- Create skill directory with
SKILL.md - Register in
skills.json - Assign to agents
- Regenerate:
python -m cmat agents generate
See .claude/docs/SKILLS_GUIDE.md for complete guide.
CMAT includes a RAG-based learnings system that captures insights from agent outputs and user feedback. Learnings are automatically retrieved and injected into agent prompts when relevant.
# Manual learning
python -m cmat learnings add "Always use pytest fixtures for database tests" --tags testing,python
# View learnings
python -m cmat learnings list
# Search for relevant learnings
python -m cmat learnings search "database testing patterns"- Agent outputs: Automatically extracted after task completion
- User feedback: Manually added via CLI
- Code patterns: Detected during analysis (future)
- Python 3.10+ - Core runtime
- Claude Code - Multi-agent orchestration platform
- pyyaml - YAML parsing (optional, included in standard lib on most systems)
Optional:
- Node.js 16+ - For MCP servers (GitHub/Jira integration)
- README.md - This file - Overview and architecture
- INSTALLATION.md - Setup and verification
- QUICKSTART.md - 5-minute quick start
- DEMO.md - Hands-on demo walkthrough
- CUSTOMIZATION.md - Adapting to your project
- CLI_REFERENCE.md - Complete command reference
- QUEUE_SYSTEM.md - Queue operations
- WORKFLOW_GUIDE.md - Workflow patterns
- WORKFLOW_TEMPLATES.md - Template management
- SKILLS_GUIDE.md - Skills system
- AGENT_GUIDE.md - Agent creation
- LEARNINGS_GUIDE.md - RAG memory system
- STATUS_COMPLETION.md - YAML completion blocks
- COST_TRACKING.md - Token usage and costs
- INTEGRATION_GUIDE.md - GitHub/Jira integration
- Install - INSTALLATION.md
- Quick Start - QUICKSTART.md - 5 minute test
- Understand Workflows - .claude/docs/WORKFLOW_GUIDE.md
- Learn Templates - .claude/docs/WORKFLOW_TEMPLATE_GUIDE.md
- Explore Skills - .claude/docs/SKILLS_GUIDE.md
- Customize - CUSTOMIZATION.md
- Workflow Templates: Define agent sequences and orchestration
- Output Directories: Standardized
required_output/andoptional_output/ - Status Transitions: Workflows define what each status means
- YAML Completion Blocks: Structured status reporting from agents
- Task Metadata: Carries workflow context (workflow_name, workflow_step)
- Skills: Domain expertise automatically provided to agents
- Learnings: RAG memory that improves over time
- Validation: Outputs validated before workflow continues
python -m cmat version # Version info
python -m cmat queue status # Queue counts
python -m cmat agents list # Available agents
python -m cmat models list # Claude modelspython -m cmat learnings list # All learnings
python -m cmat learnings add "Learning content"
python -m cmat learnings search "query"python -m cmat costs show <task_id>
python -m cmat costs enhancement <name>- Claude Code: https://claude.ai/code
- Complete Documentation: See
.claude/docs/directory
Ready to start? See QUICKSTART.md for a 5-minute walkthrough, then try DEMO.md for a hands-on example.
Need help? See .claude/docs/WORKFLOW_GUIDE.md for patterns and .claude/docs/CLI_REFERENCE.md for commands.
Want to customize? See CUSTOMIZATION.md for adapting to your project.