自我进化harness跨语言与模型的编码机制研究
原标题：One Recipe, Many Harnesses: What Self-Evolution Encodes Across Languages and Models
AI 摘要
该研究通过固定自我进化配方，在8种编程语言和3个基础模型上分析编码智能体的自我进化机制。研究发现，进化产生的harness主要补偿可恢复的执行缺陷，而非基准过拟合；不同语言共享抽象策略但使用不同的生态系统工具，且共享核心可迁移至通用harness。该研究将自我进化harness重新定位为可解释的补偿层，受语言工程需求和模型行为差距共同塑造。
正文节选
One Recipe, Many Harnesses: What Self-Evolution Encodes Across Languages and Models Abstract Self-evolving harnesses are closed-loop systems in which an agent inspects its own rollouts and edits its prompts, tools, and memory. They reliably improve coding agents in evaluations, but prior work reports aggregate gains rather than analyzing what the evolved artifacts encode. It therefore remains unclear whether they encode benchmark-specific adaptations, language-specific engineering knowledge, or