AI Skills Development Pipeline

A sophisticated web-based system that generates advanced AI agent skills through iterative refinement. The system runs a 3-stage pipeline combining multiple AI models to research, analyze, and create structured skill templates.

Features

• Multi-Stage Pipeline: Research → Analysis → Template Generation
• Iterative Refinement: Automatically improve templates based on analysis
• Interactive Web UI: Clean, responsive interface for managing the pipeline
• Template Editor: Live preview and editing capabilities
• Export Options: Download individual templates or complete pipeline runs
• Version History: Track iterations and improvements over time

Quick Start

Prerequisites

• UV - Python package installer
• OpenRouter API key (sign up at openrouter.ai)

Installation

1. Install UV (if not already installed)

   # On macOS/Linux
   curl -LsSf https://astral.sh/uv/install.sh | sh
   
   # On Windows
   powershell -c "irm https://astral.sh/uv/install.ps1 | iex"
   
   # Or with pip
   pip install uv
   

2. Clone the repository

   git clone <repository-url>
   cd skills_FFT
   

3. Install dependencies with UV

   # Install and create virtual environment
   uv sync
   
   # Or explicitly
   uv venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate
   uv pip install -e .
   

4. Configure environment

   cp .env.example .env
   

   Edit .env and add your OpenRouter API key:

   OPENROUTER_API_KEY=your_api_key_here
   

5. Run the application

   uv run app.py
   

6. Open in browser Navigate to http://localhost:5001

Usage

Generating a Skill Template

1. Enter your query in the main interface (e.g., "Research quantum computing optimization techniques")
2. Click "Start Pipeline" to begin the 3-stage process
3. Wait for the pipeline to complete
4. Review the generated template
5. Optionally iterate to refine further (if new instructions are generated)

Working with Templates

• Preview: Click "Preview & Edit" to view and modify the template
• Export: Download templates as ZIP files
• History: View all previous iterations and improvements
• Management: List, preview, and delete generated templates

Architecture

Pipeline Stages

1. Domain Research (Perplexity Llama 3.1 Sonar)

   • Comprehensive research on the topic
   • Identifies key concepts, methodologies, and applications
   • Provides reference sources and further reading

2. Deep Analysis (DeepSeek Coder V2)

   • Critical analysis of research findings
   • Identifies issues, shortcomings, and gaps
   • Suggests enhancements and improvements
   • Generates new research instructions if needed

3. Template Generation (DeepSeek Coder V2)

   • Creates structured skill templates
   • Includes YAML metadata with proper formatting
   • Provides clear instructions and examples

Data Structure

Generated templates are stored in skills_output/ with the following structure:

skills_output/
└── YYYY-MM-DD_HH-MM-SS/
    ├── skill.md              # Final template
    ├── metadata.json         # Pipeline information
    ├── iteration_1/          # First iteration
    │   ├── research.md
    │   ├── analysis.md
    │   ├── template.md
    │   └── instruction.txt
    └── iteration_2/          # Second iteration (if applicable)

API Reference

Pipeline Endpoints

• POST /api/pipeline/start

  • Start a new pipeline run
  • Body: {"query": "your query here"}

• POST /api/pipeline/iterate

  • Run an iteration of the pipeline
  • Body: {"session_id": "session-id"}

Template Endpoints

• GET /api/templates

  • List all generated templates

• GET /api/template/<timestamp>

  • Get specific template content

• POST /api/template/<timestamp>/save

  • Save edited template
  • Body: {"content": "template content"}

• GET /api/template/<timestamp>/export

  • Download template as ZIP
  • Query: ?type=template or ?type=full

• DELETE /api/template/<timestamp>

  • Delete a template and all files

Configuration

Environment Variables

• OPENROUTER_API_KEY (required): Your OpenRouter API key
• FLASK_ENV: Environment (default: development)
• FLASK_DEBUG: Debug mode (default: True)
• SECRET_KEY: Flask secret key

Model Configuration

Models can be configured in config.py:

RESEARCH_MODEL = "perplexity/llama-3.1-sonar-large-128k-online"
ANALYSIS_MODEL = "deepseek/deepseek-coder-v2"
TEMPLATE_MODEL = "deepseek/deepseek-coder-v2"
MAX_ITERATIONS = 3

Testing

Run the test suite with UV:

# Run all tests
uv run pytest tests/

# Run with coverage
uv run pytest --cov=. tests/

# Run specific test file
uv run pytest tests/test_pipeline.py

# Install test dependencies first (if needed)
uv sync --group test

Development Tools

UV provides integrated development tool support:

# Format code
uv run black .
uv run isort .

# Lint code
uv run flake8 .

# Type checking
uv run mypy .

# Run pre-commit hooks
uv run pre-commit run --all-files

Development

Project Structure

skills_FFT/
├── app.py                 # Main Flask application
├── config.py              # Configuration management
├── pipeline.py            # Pipeline logic
├── requirements.txt       # Python dependencies
├── templates/             # Jinja2 HTML templates
├── static/                # CSS, JavaScript, images
├── utils/                 # Utility modules
├── tests/                 # Test files
└── skills_output/         # Generated templates (git-ignored)

Adding New Models

To add a new model to the pipeline:

1. Update config.py with the model name
2. Modify the appropriate stage function in pipeline.py
3. Update prompts as needed for the new model

Customizing Prompts

Prompts are defined in pipeline.py in each stage function:

• run_research(): Research prompt
• run_analysis(): Analysis prompt
• run_template_generation(): Template generation prompt

Troubleshooting

Common Issues

1. API Key Error

   • Ensure OPENROUTER_API_KEY is set correctly
   • Verify the key has access to the required models

2. Pipeline Timeout

   • Increase timeout in pipeline.py if needed
   • Check network connectivity

3. Template Not Found

   • Verify the timestamp is correct
   • Check skills_output/ directory

Debug Mode

Enable debug mode by setting:

FLASK_DEBUG=True

This will provide detailed error messages and auto-reload on code changes.

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests for new functionality
5. Run the test suite
6. Submit a pull request

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• OpenRouter for providing access to multiple AI models
• Flask for the web framework
• The open-source community for various tools and libraries

Support

For support, please:

1. Check the troubleshooting section
2. Search existing issues
3. Create a new issue with details about your problem

---

Note: This tool requires an active OpenRouter API subscription and may incur costs based on usage.