An empirical study analyzing 29,896 software artifacts to understand which open-source repositories are suitable for transformation into AI agent skills.

The default extraction backend in this repository is Alibaba Bailian (qwen-plus).

• 35.8% of analyzed projects show high skillability potential
• 9,033 GitHub repositories identified as promising conversion candidates
• Skillability is independent of popularity (Spearman r_s = 0.003)
• High-potential projects concentrate in Data Retrieval, Multimedia, and System Infrastructure

Skillability measures how suitable a software project is for transformation into an agent-facing skill across six dimensions:

1. Task Clarity - How focused and well-bounded the purpose is
2. Interface Clarity - How explicit the invocation interface is
3. Composability - How naturally it fits into workflows
4. Automation Value - How much manual effort it removes
5. Deployment Friction - Difficulty to deploy (reverse-coded)
6. Operational Risk - Risk of automated execution (reverse-coded)

Formula: SS = 0.25×TC + 0.20×IC + 0.20×C + 0.25×AV + 0.05×DF' + 0.05×OR'

pip install -r requirements.txt
cp .env.example .env
# edit .env and set BAILIAN_API_KEY

# 1. Crawl GitHub repositories
python crawl_github_repos.py

# 2. Sample data
python data_sampler.py

# 3. Run skillability scoring
python main.py

# 4. Generate visualizations
python generate_paper_figures.py

├── paper/                      # Paper manuscript and figures
│   ├── paper_v7.md            # Latest paper version
│   └── assets/                # Figures used in the Markdown paper
├── latex/                      # Optional LaTeX/PDF paper version
│   ├── main.tex               # Main source
│   └── main.pdf               # Built paper PDF
├── docs/                       # Interactive visualization website
│   ├── index.html             # Main page
│   ├── repos.json             # Top 250 repositories
│   └── assets/                # Figure images
├── output/                     # Generated analysis outputs (gitignored)
├── validation/                 # Validation scripts
├── main.py                     # Main analysis pipeline
├── scorer.py                   # Skillability scoring logic
├── visualizer.py               # Visualization generation
├── crawl_github_repos.py       # GitHub data collection
├── scrape_clawhub_skills.py    # Clawhub marketplace scraper
└── requirements.txt            # Python dependencies

View the interactive visualization at docs/index.html featuring:

• Six-dimension skillability framework
• Distribution analysis across 29,896 artifacts
• Category rankings and insights
• Paginated table of 250 top-ranked repositories

Deploy to GitHub Pages by publishing the docs/ directory.

• Total artifacts: 29,896 (2,200 Clawhub + 27,696 GitHub)
• High-skillability rate: 35.8% overall (75.7% Clawhub, 32.6% GitHub)
• Mean skillability: 2.95 overall (3.75 Clawhub, 2.88 GitHub)

1. Data Retrieval & Search (3.42, 48.3% high-skillability)
2. Multimedia Content (3.38, 46.1%)
3. System Infrastructure (3.31, 44.2%)

@article{skillability2026,
  title={From Software Repositories to Agent Skills: An Exploratory Empirical Study of Skillability in Open-Source Ecosystems},
  author={Anonymous Authors},
  year={2026}
}

MIT

Generated outputs under output/ and output_large/ are intentionally not tracked.