A graphical user interface for managing multi-agent development workflows using the Claude Multi-Agent Development Template (CMAT) v8.2+.
I created this tool to make the creation and management of CMAT-based projects easier. Rather than using the cmat.sh script directly, users can leverage a comprehensive GUI to create tasks, manage agents with their skills, configure workflow templates, and generate enhancements for their projects.
Once installed and connected to a CMAT project, users can create 'enhancements' - well-structured specification files that include user stories, requirements, constraints, and acceptance criteria. These enhancements can then be processed through customizable multi-agent workflows, with each agent contributing its expertise to deliver comprehensive results.
Default workflows are provided for common tasks, but users can create custom workflows to suit their needs.
- 📋 Task Management - Create, start, cancel, and monitor tasks
- 🎭 Agent Management - Create and edit agents with visual skills assignment (simplified in v5.0)
- 🎯 Skills Management - Browse skills, view agent skills, preview skills prompts
- 📝 Enhancement Generator - AI-assisted creation of enhancement specification files
- 🤖 AI-Powered Generation - Generate task descriptions and enhancement specs using Claude API
- 📋 Workflow Template Management - Create and manage reusable workflow templates with visual editors
- 🚀 Workflow Launcher - Quick workflow launcher with pre-flight validation
- 🔗 Integration Dashboard - Monitor GitHub/Jira/Confluence sync status
- ⚙️ Advanced Settings - Configure Claude model, API key, and token limits
- 🎯 Multi-Project Support - Connect to different CMAT projects
- 🚀 Zero External Dependencies - Uses only Python standard library
- 🧠 Learnings Management - Browse, filter, add, and manage RAG system learnings
- 🤖 Model Configuration - Configure Claude models, pricing, and set default model
- 🔗 Direct Python Integration - Native Python CMAT API integration (no subprocess overhead)
- 📚 RAG System Support - Full support for CMAT's learning extraction and retrieval
- 🔧 Modular Workflow Architecture - Agents are now reusable components, workflows define orchestration
- ⚡ Visual Workflow Template Editor - Configure input patterns, output files, and status transitions through UI
- 🎨 Transition Management - Visually manage what happens when agents output specific status codes
- 📊 Dynamic Workflow Loading - Task creation automatically loads all user-defined workflows
- 🔍 Workflow Context Display - Tasks show their position in workflows with expected inputs/outputs
- ✅ Real-time Workflow Validation - Validates workflows before starting with clear feedback
- Python 3.7 or higher
- Tkinter (included with Python)
- A project using the Claude Multi-Agent Development Template v8.2+ (can be installed directly or connected to an existing project)
- Claude API key (optional, for AI-powered features)
- PyYAML (for CMAT v8.2+ Python integration)
# Clone the repository
git clone https://github.com/yourusername/ClaudeMultiAgentUI.git
cd ClaudeMultiAgentUI
# Run directly
python3 -m src.main
# Or use the launcher script
./run.pypython3 -m src.main
# OR
./run.pyYou can either connect to an existing CMAT project or install to your project.
If you don't have a CMAT project yet, you can install one directly:
The installer:
- Downloads the latest CMAT template from GitHub
- Creates a
.claude/folder with all necessary files - Creates a backup if overwriting an existing installation
- Validates the installation for security and completeness
- Click File > Install CMAT Template...
- Select or enter the directory where you want to install CMAT
- The dialog validates the directory and warns if
.claude/already exists - Click Install to download and install CMAT from GitHub
- Files are downloaded from the CMAT GitHub repo and installed
- When complete, click Connect Now to immediately connect to the new project
If you already have a CMAT project:
- Click File > Connect... (or press
Ctrl+O) - Click Browse...
- Navigate to your CMAT project root directory
- Select the project root (containing
.claude/folder) - The UI will validate the project structure
- Click Connect
The validation checks for:
- ✓ CMAT script (
.claude/scripts/cmat.sh) - ✓ Task queue (
.claude/queues/task_queue.json) - ✓ Workflow templates (
.claude/queues/workflow_templates.json) - ✓ Skills system (
.claude/skills/skills.json) - ✓ Agents (
.claude/agents/agents.json)
For AI-powered features (highly recommended):
- Get a Claude API key from console.anthropic.com
- Click Settings > Claude Settings
- Enter your API key
- Choose model (default: Claude Sonnet 4.5 - recommended)
- Set max tokens (default: 8192 - fine for most tasks)
- Click Save Settings
You can now use:
- Generate with Claude in task creation
- Enhancement Generator for creating specs
- Agent instructions generation when creating agents
See the User Guide for more detailed instructions on system operation.
Create standalone tasks with optional source files. Tasks auto-complete but don't auto-chain to other agents.
View comprehensive task details including workflow context, skills used, and full prompts sent to Claude.
v4.x: Agents defined their own orchestration (inputs, outputs, next agents) v5.0: Workflows define orchestration, agents are reusable components
- Agent Reusability - Same agent can work in multiple workflows with different configurations
- Flexible Workflows - Change workflow orchestration without modifying agents
- Visual Workflow Editing - Configure input patterns, outputs, and transitions through UI
- Dynamic Workflow Loading - Create custom workflows and they appear everywhere automatically
-
Workflow Template Editor - Full visual editor for creating/editing workflows
- Configure input patterns with placeholders
- Set required output filenames
- Manage status transitions visually
- Validate workflows before saving
-
Workflow Starter Dialog - Quick launcher for starting workflows
- Select workflow template
- Choose or create enhancement spec
- Pre-flight validation checks
- One-click start
-
Simplified Agent Management - Agents are now just capabilities
- No workflow orchestration in agent definition
- Focus on tools and skills
- Workflow behavior configured separately
-
Enhanced Task Details - Shows workflow context
- Current step in workflow
- Expected inputs/outputs
- Expected status codes
- Workflow name and position
The application uses sensible defaults in src/config.py:
- Auto-refresh interval - Default: 3 seconds
- Max log lines - Default: 1000 lines
- Window size - Default: 1200x750
- Status colors - Blue (pending), Orange (active), Green (completed), Red (failed)
- Priority colors - Red (critical), Orange (high), Black (normal), Gray (low)
Configure via Settings > Claude Settings:
- API Key - Your Claude API key from console.anthropic.com
- Model - Choose from Opus 4, Sonnet 4.5, Sonnet 4, or Haiku 4
- Max Tokens - Output token limit (defaults to model maximum)
- Timeout - Request timeout in seconds (default: 90)
Settings are persisted to ~/.claude_queue_ui/settings.json
ClaudeMultiAgentUI/
├── src/
│ ├── __init__.py
│ ├── main.py # Main application
│ ├── config.py # Configuration + ClaudeConfig
│ ├── settings.py # Settings persistence
│ ├── utils/ # Utility modules
│ │ ├── __init__.py
│ │ ├── claude_api_client.py # Centralized API client
│ │ ├── cmat_interface.py # CMAT interface (v5.0)
│ │ ├── cmat_installer.py # CMAT template installer
│ │ ├── workflow_migration_utils.py # v4.x → v5.0 migration tools
│ │ ├── text_utils.py # Slug conversion, validation
│ │ ├── path_utils.py # Path utilities
│ │ └── time_utils.py # Time formatting
│ ├── models/ # Data models
│ │ ├── __init__.py
│ │ ├── task.py
│ │ ├── agent.py
│ │ ├── workflow_template.py # v5.0 workflow models
│ │ ├── queue_state.py
│ │ └── ...
│ └── dialogs/ # Dialog windows
│ ├── __init__.py
│ ├── base_dialog.py # ABC base class
│ ├── mixins/ # Dialog mixins
│ │ ├── __init__.py
│ │ └── claude_generator_mixin.py
│ ├── connect.py
│ ├── task_create.py # v5.0 - dynamic workflows
│ ├── task_details.py # v5.0 - workflow context
│ ├── agent_details.py # v5.0 - simplified
│ ├── enhancement_create.py
│ ├── workflow_template_manager.py
│ ├── workflow_template_editor.py # v5.0 - full visual editor
│ ├── workflow_step_editor.py # v5.0 - NEW
│ ├── workflow_transition_manager.py # v5.0 - NEW
│ ├── workflow_starter.py # v5.0 - NEW
│ ├── workflow_viewer.py # v5.0 - uses templates
│ ├── claude_settings.py
│ ├── working.py
│ └── ...
├── tests/ # Unit tests
├── assets/ # Images, icons
├── README.md
├── QUICKSTART.md
├── USER_GUIDE.md
├── requirements.txt
├── setup.py
└── LICENSE
The UI supports managing multiple CMAT projects:
- Connect to Project A
- Work with Project A (tasks, agents, enhancements)
- File > Connect... to switch to Project B
- Work with Project B
- Switch back anytime
The last connection is saved and auto-reconnects on launch.
Workflows define how agents work together to process enhancements:
- Workflows > Manage Templates... → "Create New Template"
- Enter template name and description
- Add steps:
- Select agent for each step
- Configure input pattern (where step reads from)
- Set output filename (what step creates)
- Manage transitions (what happens on each status)
- Save template
Two methods to start workflows:
Method 1: Workflow Starter (Recommended)
- Workflows > Start Workflow... (
Ctrl+Shift+W) - Select workflow template from dropdown
- Choose existing enhancement or create new one
- Pre-flight checks validate everything
- Click Start Workflow
Method 2: Quick Start in Task Create
- Tasks > Create Task... (
Ctrl+N) - Select workflow from dropdown (loads dynamically)
- First agent and settings pre-filled
- Enter details and create
Workflows > View Active Workflows... (Ctrl+W)
- See real-time progress of all workflows
- Progress bars show completion percentage
- Agent status indicators (✓ completed, → active, ○ pending)
- Uses actual workflow templates (not hardcoded)
- Works with custom workflows of any length
Cause: Directory structure doesn't match CMAT v5.0 requirements.
Solutions:
- Verify you're selecting project root, not
.claude/directory - Check that
.claude/scripts/cmat.shexists - Check that
.claude/queues/workflow_templates.jsonexists - Run
ls -la .claude/to verify structure - If v4.x or earlier, workflows will load but may need updating
Cause: Path to cmat.sh is invalid.
Solutions:
- Use File > Connect... to browse for project
- Don't manually edit connection path
- Verify
cmat.shhas execute permissions:chmod +x .claude/scripts/cmat.sh
Solutions:
- Click Refresh or press
F5 - Check connection status in header
- Verify queue file exists:
.claude/queues/task_queue.json - Enable auto-refresh: Settings > Auto Refresh Task List
Symptoms: Start task option grayed out.
Solutions:
- Task must be in "Pending" status
- Active tasks cannot be restarted
- Cancel task first, then recreate if needed
Cause: Agent execution failed without updating status.
Solutions:
- Check task log for errors
- Cancel task
- Fix issue in source file
- Create new task
Cause: Workflow template not saved properly.
Solutions:
- Workflows > Manage Templates...
- Verify template exists in list
- Edit template and save
- Refresh task creation dialog
Cause: Status transitions not configured.
Solutions:
- Edit workflow template
- Edit the step that's not chaining
- Click "Manage Transitions"
- Add transition for the status code agent outputs
- Save changes
Solutions:
- Edit workflow template
- Click "Validate Workflow"
- Review error messages
- Fix missing input/output configurations
- Ensure transition targets exist
Cause: Claude API key not configured.
Solutions:
- Menu: Settings > Claude Settings
- Enter API key from console.anthropic.com
- Save settings
- Try generation again
Cause: Complex request taking too long.
Solutions:
- Settings > Claude Settings
- Increase timeout (try 120-180 seconds)
- Switch to faster model (Sonnet 4.5 or Haiku 4)
- Simplify input (shorter description, fewer reference files)
Solutions:
- Use Claude Opus 4 for complex tasks
- Provide more detailed description
- Add relevant reference files
- Regenerate with different wording
Cause: Agent has no skills assigned.
Solutions:
- Agents > Manage Agents...
- Edit agent
- Go to Skills tab
- Assign skills
- Save agent
Cause: Missing skills.json file.
Solutions:
- Verify file exists:
.claude/skills/skills.json - If missing, reinstall CMAT template
- Check file permissions
Cause: Long-running API call.
This is normal:
- Complex enhancements take 30-60 seconds
- Watch working animation to confirm it's processing
- Don't close window, wait for completion
Solutions:
- Disable auto-refresh if not needed
- Close other resource-intensive applications
- Reduce number of active tasks
- Clear finished tasks: Tasks > Clear Finished...
Cause: Network slow or GitHub unavailable.
Solutions:
- Check internet connection
- Try again later
- Use VPN if GitHub blocked
- Download template manually from GitHub
Cause: No write access to target directory.
Solutions:
- Choose directory you own
- Check directory permissions
- Don't install to system directories
- Try different location
Contributions are welcome! Please:
- Fork the repository
- Create a feature branch
- Make your changes
- Add tests if applicable
- Submit a pull request
MIT License - see LICENSE file for details
- Claude Multi-Agent Template - The multi-agent system this UI manages
Built with ❤️ for the Claude Multi-Agent Development Template




