GitHub Holds 3.8M Agent Skill Files, But Over Half Are Exact Copies
rohanpaul_ai · x · 2026-08-23
A new paper analyzing GitHub repositories reveals that while there are 3.8 million agent skill files publicly available, over half are exact duplicates of another file, leaving only about 1.9 million unique files.
This highlights the lack of a registry or package manager for the current agent skill format. Skills are typically distributed as folders containing Markdown instructions. Agents read the short description at runtime to decide whether to load them, bypassing compiler checks. This means skills in repos are frozen, and authors cannot push fixes.
The paper also raises the question of whether edited copies of popular skills introduce new commands or network calls not present in the originals—a risk that can now be analyzed at scale.
The full dataset is available as a single SQLite file.
- Paper: GitSkills: A Dataset of Agent Skills on GitHub (arxiv.org/abs/2608.10906)
Related event: Over half of GitHub Agent skill files are duplicates(2 posts)→
More from Research
- Converting GMMs ↔ PEFs for fast KLD approximation — FrnkNlsn · 2026-08-24
- Netflix details its production LLM judge: hundreds of thousands of recommendations scored weekly — omarsar0 · 2026-08-24
- Nature Comment: Provenance, not interpretability, grounds trust in autonomous science — gabepgomes · 2026-08-24
- New Architecture RHEA: Train 1B Model on 8GB VRAM — zemondza · 2026-08-24
- Trained two 16M-param models to do generative CAD with real physics — debreuil · 2026-08-24
- Claude model helps discover complex structure on S^6, solving 60-year-old math problem — Singularitarian · 2026-08-24