Skill Constellations: Tracing the Supply Chain of Agent Skills on GitHub
Study of 2.2 million GitHub agent-skill copies finds stars miss the repositories that spread risk.
Researchers build the first dated copy network of agent skills from the git history of every SKILL.md in GitSkills, covering 2,193,119 adoptions on GitHub. Skills are instructions and scripts that agents such as Claude Code and Codex run with the user's permissions, and copies rarely pick up fixes made at the source. A model of which repositories others copy from shows that reviewing its top 100 prevents 14.9% of later high-risk skill adoptions, versus 0.5% for the 100 most starred. The authors recommend distributing versioned references instead of copies.
- Git history yields 2,193,119 dated skill adoptions on GitHub.
- A few source repositories account for almost all copies.
- GitHub stars do not identify those source repositories.
- Copies rarely inherit later security fixes from the source.
- Top 100 audit targets stop 14.9% of risky later adoptions.
- Products
- Claude CodeCodexGitSkills
- Organizations
- GitHub
From the original article
1 min read at huggingface.co
Agent skills are SKILL. md instructions and scripts that AI coding agents such as Claude Code and Codex run with the permissions of their user. Developers share skills by copying them between repositories, which makes them a software supply chain without a registry, versions or provenance.
By Fahd Seddik · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.