---
title: "A First Look at Coding Agents' Compliance with AI Contribution Rules in Open-Source Communities"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.26819v1"
permalink: "https://feed.astralor.com/items/itm_23b70acc7ae1ee7c7157ee17"
published: "2026-07-29T12:14:17.000Z"
score: 81
channel: "ai"
tags: ["AI Coding Agent","Claude","GPT-4","RepoComplianceBench","Compliance","Open-Source Governance"]
---

# A First Look at Coding Agents' Compliance with AI Contribution Rules in Open-Source Communities

## AI commentary

> This paper pioneers the evaluation of AI Agent rule compliance, directly addressing open-source communities' anxiety over AI-generated PRs. If you are an open-source maintainer, the data is sobering—don't expect agents to follow rules voluntarily; embed rules into tooling instead. For AI developers, it highlights a new alignment dimension: model capability alone is insufficient; agents must actually read and obey rules.

## At a glance

Open-source communities overwhelmed by AI-generated contributions have established rules ranging from total bans to mandatory disclosure, verification gates, and human sign-offs. This study builds RepoComplianceBench, covering 106 issues from 49 repositories, to evaluate whether four frontier coding agents retrieve and comply with these rules. Findings show agents almost never proactively retrieve rules; while prompts improve disclosure and verification, they never refuse to contribute in AI-banned repositories. Enforcement of bans and human escalation remain open challenges.

## Key points

- Constructed RepoComplianceBench with 106 issues from 49 repositories to specifically evaluate coding agents' compliance with AI contribution rules.
- Tested four frontier models and found agents almost never proactively retrieve contribution rules from repositories.
- Reminder prompts, rule quotes, and verifier feedback improved agent performance on disclosure and verification tasks.
- Agents never refused to contribute in AI-banned repositories under any tested condition, and human escalation remained unenforced.

## Full text

Open source communities have been flooded with AI-generated contributions. In defense, they have written contribution rules to regulate coding agents' behavior, spanning from a total ban, mandatory disclosure, to verification gates and human sign-offs. Yet, whether coding agents read and follow those rules, and behave in open source repositories, remains unknown. To estimate real-world rule compliance of coding agents, we curate 106 issues from 49 repositories containing AI contribution rules into RepoComplianceBench. We judge the trajectory of each run against the repository's rules, measuring whether the agent refuses to contribute, discloses its assistance truthfully, clears the required verification gates, or escalates critical steps to a human. We also test if extra prompts, rule disclosure, or feedback from the compliance verifier help with the situation. Our experiments on four frontier models show that today's agents almost never proactively retrieve the contribution rules. Agents pick up disclosure and verification with reminder prompts, rule quotes, and verifier feedback; however, they never refuse to contribute in AI-banned repositories under any condition we tested. The status reveals that verification and disclosure issues are solvable with existing mechanisms, yet enforcing bans and human escalations remains an open problem.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_23b70acc7ae1ee7c7157ee17