---
title: "编码代理对开源社区 AI 贡献规则的遵守情况初探"
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编码代理","Claude","GPT-4","RepoComplianceBench","合规性","开源治理"]
---

# 编码代理对开源社区 AI 贡献规则的遵守情况初探

## AI 点评

> 这篇论文是‘AI Agent 规则合规’领域的先行者，直接回应了开源社区对AI刷PR的焦虑。如果你是开源项目维护者，这里的数据值得警惕——不要指望Agent自觉遵守规则，必须将规则植入工具链。对于AI开发者，这提示了Agent安全对齐的新维度：不能只看模型能力，还要看它是否‘读规则’。

## 速览

开源社区被 AI 生成贡献淹没，纷纷制定贡献规则（禁止、强制披露、验证闸门、人工签批等）。本研究构建 RepoComplianceBench 基准，涵盖 49 个代码仓库的 106 个 Issue，评估四种前沿模型的编码代理是否检索、遵守这些规则。实验发现：代理几乎从不主动检索规则；在提示下可改进披露与验证，但始终拒绝在禁止 AI 的仓库中停止贡献。表明禁止与人工升级仍是开放问题。

## 关键点

- 构建包含 49 个仓库、106 个 Issue 的 RepoComplianceBench 测试集，专门评估编码代理对 AI 贡献规则的遵守情况。
- 测试了四种前沿模型，发现代理几乎从不主动检索仓库中的贡献规则。
- 通过提醒提示、规则引用和验证器反馈，代理在披露和验证任务上表现提升。
- 但在明确禁止 AI 贡献的仓库中，代理在任何条件下都不会拒绝贡献，人工升级也未能自动触发。

## 全文

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.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_23b70acc7ae1ee7c7157ee17