arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.13428cs.SEcs.AI
收藏

EvoClaw: 评估AI代理在持续软件演化中的表现

SWE-Milestone: Evaluating AI Agents on Continuous Software Evolution

USC(美国斯克利普斯大学);UCR(加州大学河滨分校);UCSD(加州大学圣地亚哥分校);Army Research Office(陆军研究办公室);Stanford(斯坦福大学);Princeton(普林斯顿大学);Haven;OpenHands
Gangda Deng, Zhaoling Chen, Zhongming Yu, Haoyang Fan, Yuhong Liu, Yuxin Yang, Dhruv Parikh, Rajgopal Kannan, Le Cong, Mengdi Wang, Qian Zhang, Viktor Prasanna,… 展开作者

Gangda Deng, Zhaoling Chen, Zhongming Yu, Haoyang Fan, Yuhong Liu, Yuxin Yang, Dhruv Parikh, Rajgopal Kannan, Le Cong, Mengdi Wang, Qian Zhang, Viktor Prasanna, Xiangru Tang, Xingyao Wang

AI总结:

针对现有基准测试忽视软件演化中时间依赖和技术债务的问题,提出EvoClaw基准,通过从提交日志重建可验证里程碑DAG,评估AI代理在持续开发中维持系统完整性和限制错误累积的能力。

摘要:

随着AI代理越来越多地被部署为长期运行的系统,自主构建并持续演化定制软件以在动态环境中进行交互变得至关重要。然而,现有基准测试在孤立的、一次性的编码任务上评估代理,忽视了真实世界软件演化中固有的时间依赖和技术债务。为弥补这一差距,我们引入了DeepCommit,一个从嘈杂的提交日志中重建可验证里程碑DAG的代理管道,其中里程碑被定义为功能内聚的开发目标。这些可执行序列使得EvoClaw成为可能,这是一个新颖的基准测试,要求代理维持系统完整性并限制错误累积,这些是当前基准测试中大部分缺失的长期软件演化的维度。我们对4个代理框架下的12个前沿模型的评估揭示了一个关键弱点:整体性能得分从孤立任务上的>80%显著下降到持续设置中的最多38%,暴露了代理在长期维护和错误传播方面的深刻困境。

英文摘要:

Real-world software must continuously evolve to meet ever-changing and open-ended requirements. AI agents, increasingly deployed as long-running systems, are now entrusted to drive this evolution. Yet, existing benchmarks evaluate agents on isolated, one-off coding tasks, neglecting the temporal dependencies and technical debt inherent in real-world software evolution. To bridge this gap, we introduce DeepCommit, an agentic pipeline that reconstructs verifiable Milestone DAGs from noisy commit logs, where milestones are defined as functionally cohesive development goals. These executable sequences enable SWE-Milestone, a benchmark that evaluates agents on streams of milestone-level tasks, requiring them to sustain system integrity and limit error accumulation, dimensions of long-term software evolution largely missing from current benchmarks. Our evaluation of 12 frontier models across 4 agent frameworks reveals a critical vulnerability: overall performance scores drop significantly from >80% on isolated tasks to 38.03% in continuous settings, exposing agents' profound struggle with long-term maintenance and error propagation.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑