First Dataset of Agentic AI Coding Tool Configurations: 15,591 Artifacts Across 4,738 Repos
arXiv·medium signal
Researchers scraped 40,585 GitHub repos and found 4,738 using agentic AI coding tools, extracting 15,591 configuration artifacts and 148,519 AI-co-authored commits. They identify eight configuration mechanisms: Context Files, Skills, Subagents, Commands, Rules, Settings, Hooks, and MCP configs. The dataset (CC BY 4.0 on Zenodo) is the first empirical look at how developers actually configure Claude Code, Codex, and similar tools in production.