GitSkills Mines 3.8 Million SKILL.md Files Across 282,200 GitHub Repos — Nine Months After Anthropic Published the Spec
arXiv 2608.10906·high signal
Researchers collected 3,797,117 SKILL.md files from 282,200 public GitHub repositories in July 2026, yielding 1,877,981 distinct file contents, and packaged the corpus with front matter, folder contents, repo metadata and partial commit history into a single self-contained SQLite file. Anthropic introduced the agent-skill format as an open specification in October 2025, so this measures nine months of organic adoption. For builders it is the first hard number on how widely the skill format actually propagated, and a ready-made dataset for studying skill reuse, authorship, maintenance and security.