Pack to Claude Skill Pipeline

A comprehensive system for capturing workflows and automatically generating Claude Skills from recorded user interactions.

Overview

This pipeline transforms real-world workflows into Claude Skills through four progressive phases:

1. Phase 1 (MVP): Recording & Basic Skill Generation
2. Phase 2: Improving Skill Quality (confidence, descriptions, robustness)
3. Phase 3: Collaboration & Deployment (team features)
4. Phase 4: Ecosystem Integration and Marketplace

Features

• Workflow Recording: Capture screen video and user interaction events
• Vision-Language Analysis: Extract meaningful steps from video frames
• Automated Skill Generation: Create Claude Skills with proper structure and metadata
• Confidence Scoring: Rate the reliability of each generated step
• Team Collaboration: Version control, sharing, and deployment features
• Ecosystem Integration: Connect with Claude's skill marketplace and MCP plugins

Quick Start

# Install dependencies
pip install -r requirements.txt

# Record a workflow
python -m pack2skill record --output my_workflow.json

# Generate a skill
python -m pack2skill generate my_workflow.json --output ./skills/

# Install the skill
python -m pack2skill install ./skills/my-skill/

Project Structure

pack2skill/
├── core/                 # Core functionality
│   ├── recorder/        # Workflow recording (screen + events)
│   ├── analyzer/        # Video frame analysis and captioning
│   ├── generator/       # Skill generation logic
│   └── utils/           # Shared utilities
├── quality/             # Phase 2: Quality improvements
│   ├── confidence.py    # Confidence scoring
│   ├── description.py   # Description optimization
│   └── robustness.py    # Edge case handling
├── team/                # Phase 3: Team features
│   ├── versioning.py    # Version control integration
│   ├── deployment.py    # Skill deployment
│   └── testing.py       # Testing framework
├── ecosystem/           # Phase 4: Ecosystem integration
│   ├── marketplace.py   # Marketplace features
│   └── integrations.py  # MCP and plugin support
├── cli/                 # Command-line interface
└── tests/               # Test suite

Requirements

• Python 3.9+
• FFmpeg (for screen recording)
• CUDA-capable GPU (recommended for vision models)

Documentation

See the docs/ directory for detailed documentation:

• Installation Guide
• Recording Guide
• Skill Generation
• Team Setup
• API Reference

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

License

MIT License - see LICENSE for details.

Acknowledgments

Built for the Claude Skills ecosystem by Anthropic.