GitSkills Dataset: 3.79M Agent Skill Files from GitHub
JeremyCMorgan · x · 2026-08-22
GitSkills is a new research dataset comprising 3.79 million SKILL.md files collected from 282,200 public GitHub repositories. Following the open specification introduced by Anthropic in October 2025, these files store instructions for language model agents. The deduplicated dataset is packaged as a single SQLite database enriched with metadata, enabling empirical research on the adoption, reuse, structure, authorship, maintenance, and security of agent skills.
More from coding & agent
- OpenKnowledge releases open-source plugin for Google's OKF LLM wiki standard — solyarisoftware · 2026-08-22
- JackRabbitOS Released: Open Source Voice Platform for Rabbit R1 — jesselyu · 2026-08-22
- AI Agent Payment Risks: How to Restrict Spending to Approved Merchants? — Hopeful-Horse7580 · 2026-08-22
- Study: 25% of Public MCP Configs Leak API Keys in Plaintext — checkpointdev · 2026-08-22
- One-click script automates vLLM setup on dual 4090s — doodlestein · 2026-08-22
- Salesforce study: Memory compounds mistakes in self-improving agents — rohanpaul_ai · 2026-08-22