BlockBeats News, March 3rd, according to 1M AI News, a research team from the Singapore Management University, Heidelberg University, Bamberg University, and King’s College London published a paper on arXiv, quantitatively assessing for the first time the impact of repository-level configuration file AGENTS.md on AI programming Agent efficiency. AGENTS.md is an instruction file stored in the root directory of a code repository, used to explain the project’s architecture, build commands, coding standards, and operational constraints to AI Agents, similar to Anthropic Claude Code’s CLAUDE.md and GitHub Copilot’s copilot-instructions.md, currently adopted by over 60,000 GitHub repositories.
The research team conducted paired experiments on 124 merged PRs (each with code changes of no more than 100 lines) in 10 repositories using OpenAI Codex (gpt-5.2-codex), running under two conditions: with and without AGENTS.md. The results showed that with AGENTS.md, the median runtime decreased from 98.57 seconds to 70.34 seconds (a decrease of 28.64%), the median output tokens decreased from 2,925 to 2,440 (a decrease of 16.58%), with no significant difference in task completion behavior (Wilcoxon signed-rank test, p
The researchers noted that AGENTS.md transforms Agent guidance from “brief hints” to “version-controlled, reviewable, collaboratively maintained configuration artifacts,” and they recommended that development teams adopt it as a standard practice in repositories. As a limitation, the study only tested a single Agent with OpenAI Codex, the sample was limited to small-scale PRs, and there was no comprehensive code correctness evaluation.
