首次编辑后的需求：挖掘真实编码智能体会话中的后期需求出现与返工
原标题：Requirements After the First Edit: Mining Late Requirement Emergence and Rework in Real-World Coding-Agent Sessions
AI 摘要
该研究基于SWE-chat数据集中的3553个真实编码智能体会话，分析了后期需求出现对代码失效的影响。研究发现，需求到达后的代码失效量约为非需求编辑的两倍，且这种负担在会话中未见下降。受控实验表明，延迟披露会将实现转移到揭示后，而仅提前警告无需求内容则无效果。该研究首次将后期需求出现与行级代码失效直接关联，揭示了编码智能体交互中可测量的代码返工来源。
正文节选
Requirements After the First Edit: Mining Late Requirement Emergence and Rework in Real-World Coding-Agent Sessions Abstract. Coding agents often implement changes before users have fully articulated their requirements, echoing a pattern long documented in requirements engineering: stakeholders often cannot state a constraint until part of the system already exists to react to. In traditional software projects, this volatility is known to be associated with schedule and cost overrun, but that ev