GitHub Agent-Configuration Prevalence Study: Dataset and Reproduction Scripts
Year: 2026
Type: other
Abstract: Empirical dataset, analysis pipeline, figures, and article PDF from a prevalence study of AI coding-agent configuration files across 10,008 public GitHub repositories (9,357 distinct organisations). Measures the governance gap in agent-configuration management: exact-duplicate propagation (10.1% fork-adjusted, SHA-256), version-control immaturity (58% single-commit; 0.4 vs 0.6 commits/month age-normalised), and permission declaration rates (<1% of agent configs vs 33% of GitHub Actions workflows). Includes the full SQLite analysis store (≈86 MB), the TypeScript analysis pipeline source, and nine vector-PDF figures. Companion to the preprint "A Deterministic Control Plane for LLM Coding Agents" (arXiv cs.SE, 2026).
Author: Madatha, Padmaraj
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7165490204
