Skill Self-Play: Pushing the Frontier of LLM Capability with Co-Evolving Skills

ID: fe769637
原文链接: https://arxiv.org/abs/2607.22529
PDF: https://arxiv.org/pdf/2607.22529v1
作者: Siyuan Huang, Pengyu Cheng, Haotian Liu, Tao Chen, Yihao Liu, Jingwei Ni, Shijie Zhou, Ziyi Yang, Gangwei Jiang, Mengyu Zhou, Yu Cheng, Xiaoxi Jiang, Guanjun Jiang
日期: 2026-07-24
更新: 2026-07-24
分类: models
来源类型: article
标签: self-play, skill-evolution, reinforcement-learning, tool-use, self-improvement
质量评分: 5/5
抓取时间: 2026-07-29T06:19:39Z

中文导读

Skill Self-Play (Skill-SP) 是一个共同进化框架，解决了LLM自我进化中任务多样性与验证可靠性之间的根本矛盾它将Agent技能作为中间层：每个技能在特定场景下提供可验证执行，而动态路由保持开放式任务多样性系统包含提议者求解者和动态技能控制器在工具使用和推理基准上，对强模型持续提升上限，对初始对齐不良的模型实现显著逆转

为什么值得关注

Skill Self-Play (Skill-SP) 是一个共同进化框架，解决了LLM自我进化中任务多样性与验证可靠性之间的根本矛盾它将Agent技能作为中间层：每个技能在特定场景下提供可验证执行...

本文以 arXiv 摘要与条目既有摘要为据：程序性技能/自进化/形式化研究/路由等主题直接对应 Agent 工程与评估实践中的可靠性、可验证性与部署权衡。

关键信息

论文标题：Skill Self-Play: Pushing the Frontier of LLM Capability with Co-Evolving Skills
作者：Siyuan Huang, Pengyu Cheng, Haotian Liu, Tao Chen, Yihao Liu, Jingwei Ni, Shijie Zhou, Ziyi Yang, Gangwei Jiang, Mengyu Zhou, Yu Cheng, Xiaoxi Jiang, Guanjun Jiang
arXiv：https://arxiv.org/abs/2607.22529
发布时间：2026-07-24
更新时间：2026-07-24
arXiv 分类：cs.CL
关联标签：self-play, skill-evolution, reinforcement-learning, tool-use, self-improvement

English Abstract

LLM training is shifting from manual design and annotation to interaction-driven self-evolution. However, existing self-evolutionary methods face a fundamental dilemma between task diversity and verification reliability: environment-bound methods obtain precise feedback but confine learning to narrow domains, while open-ended self-generation broadens the task space but lacks reliable verification, allowing misleading rewards to pollute the training loop. We identify agent skills as a powerful middle ground to reconcile this tension: each skill ensures deep, verifiable execution in a specific scenario, while dynamic routing across skills maintains open-ended task variety. Leveraging this insight, we introduce Skill Self-Play (Skill-SP), a co-evolutionary framework comprising a proposer, a solver, and a dynamic skill controller. Orchestrated via a reinforcement learning loop, these components co-evolve in a continuous self-play loop: the proposer generates challenging tasks conditioned on dynamically sampled skills; the solver explores candidate solutions to push its capability boundaries; and the skill controller collects execution feedback to update and expand the skill library. This interactive co-evolution effectively bridges the gap between structured verification and open-ended exploration. Empirical evaluations on tool-use and reasoning benchmarks demonstrate that Skill-SP, serving as a robust evolution engine, consistently pushes the performance ceiling of competent backbones while catalyzing striking turnarounds for initially misaligned models. Our code is available at https://github.com/Qwen-Applications/skill-self-play.

English Summary

Skill Self-Play (Skill-SP) is a co-evolutionary framework that reconciles the tension between task diversity and verification reliability in LLM self-evolution. It uses agent skills as a middle ground: each skill provides verifiable execution in a specific scenario while dynamic routing maintains open-ended task variety. The system comprises a proposer (generates challenging tasks from sampled skills), a solver (explores solutions to push capability boundaries), and a dynamic skill controller (updates the skill library from execution feedback). Evaluations on tool-use and reasoning benchmarks show consistent performance ceiling improvements for strong backbones and striking turnarounds for misaligned models.

Obsidian Notes

内容由 opencli arxiv paper 拉取 arXiv 元数据与摘要生成。
中文导读与价值判断均锚定在条目已有摘要、论文摘要、作者、日期与分类信息上；未补充论文摘要之外的实验细节。