arXivDaily arXiv每日学术速递 周一至周五更新
​通知:联邦节假日调休:哥伦布日 / 原住民日(Columbus Day / Indigenous Peoples' Day),今日(13日)无更新。
arXiv 2606.01314cs.AI

SkillSmith: 技能与工具的协同进化用于自我改进的智能体系统

SkillSmith: Co-Evolving Skills and Tools for Self-Improving Agent Systems

Shanghai Jiao Tong University(上海交通大学);Eastern Institute of Technology(东部技术研究所);University of Science and Technology of China(中国科学技术大学);Southeast University(东南大学);Ningbo Institute of Digital Twin(宁波数字孪生研究所)

Yangbo Wei, Zhen Huang, Shaoqiang Lu, Junhong Qian, Qifan Wang, Chen Wu, Lei He

AI总结:

提出SkillSmith框架,通过统一提案空间和Lotka-Volterra生态效用模型实现技能与工具的协同进化,在多个基准测试中显著提升性能。

摘要:

最近的自进化智能体表明,技能可以通过执行被发现、精炼和积累。然而,现有的技能进化框架通常假设固定的工具层,并独立评估每个技能,限制了它们修复工具级故障或推理技能间交互的能力。我们提出SkillSmith,一个协同感知的技能-工具协同进化框架。SkillSmith引入了一个统一的提案空间,其中反思产生原子束,共同修改技能和工具,允许在技能进化识别出可重用的能力缺口时,对工具进行包装、编辑、组合、拆分或淘汰。为了指导这种联合搜索,SkillSmith维护了一个受Lotka-Volterra动力学启发的生态效用模型,其中从执行轨迹估计的交互矩阵捕获技能间的成对互补和冲突,并为检索、变异优先级和淘汰提供压力信号。此外,SkillSmith记录反模式,包括失败特征、因果归因和补救措施,以加速诊断并否决重复已知错误的提案。在包括WildClawBench在内的三个基准测试和五个Qwen3.5模型规模上的实验表明,SkillSmith始终优于强基线,并且随着任务复杂性和多技能共激活的增加,增益会放大。

英文摘要:

LLM agents increasingly store capabilities in external skill and tool libraries. Libraries can grow, but execution must fit a finite context window and improvement consumes a finite iteration budget. We formulate external-policy evolution under these two budgets and introduce SkillSmith. A skill ecosystem allocates context through replicator dynamics, using randomized masking and Bayesian estimation to learn skill effects and interactions. Four typed operations repair tools and move reusable logic across the skill-tool boundary, freeing context for the ecosystem. Conflict-driven learning distills failed proposals into version-scoped nogoods that guide repairs and veto repeated failing proposals. We prove utility ascent for fixed-model continuous dynamics, conditions under which lower context costs expand attainable utility, and the need for synchronized interface migration under persistent exact-match dependencies. Across five Qwen3.5 scales, SkillSmith improves over EvoSkill by up to 10.2 and 9.8 percentage points on OfficeQA and SealQA, respectively, and sustains improvement beyond SkillClaw on WildClawBench. Skills compete for context; tools are how the system makes room.

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。

↑