首页 > AI前沿 > Evolve Where It Matters: Coverage-Guided Self-Evolution of Agent Skills

Evolve Where It Matters: Coverage-Guided Self-Evolution of Agent Skills

arXiv自然语言 2026-11-20 13:00 31 阅读 查看原文
用户需要我扮演一个专业的新闻排版工程师,将一段纯文本新闻正文排版成语义化的HTML片段。 规则如下: 1. 只输出HTML片段本身,不要Markdown围栏(如```html),不要解释。 2. 每个自然段落用`

`包裹,长文用`

`/`

`划分小节。 3. 列表内容用`
  • `,引用用`
    `,代码用`
    `,重要句子可用``。
    4. 保留原文所有信息,不得增删改内容。
    5. 不要输出``/``/``等文档级标签。
    
    现在来看正文内容:
    "Agent skills package procedural knowledge as natural-language instructions, and a skill helps only when its instructions are correct and properly followed by the agent. Self-evolving agents revise their own skills from the runs they make, but existing methods judge each run as a whole, by its outcome or by where it went wrong, and cannot tell whether a failure comes from a wrong instruction or from one the agent did not follow. Instruction-level coverage was introduced to test skills and labels each instruction of a run as follow, not-follow, or not covered, which separates these causes. We leverage it for skill self-evolution with SkillPlumber, a framework in which an agent evolves its own skills where coverage shows they fall short. A persistent root cause analysis agent turns each label into a revision question and checks each revision against the next run, while a maintainer distills investigations across tasks into a shared strategy library. Across three benchmarks and three executors, SkillPlumber solves 64.7% to 70.5% of the tasks, 25.6 to 30.4 percentage points higher than the initial skill and 8.7 to 14.0 percentage points higher than the strongest baseline, and removing coverage reduces recovered failures by 20.8% on SkillsBench with GPT-5.5."
    
    分析正文结构:
    这其实是一段连续的学术/技术新闻摘要。没有明显的小节标题,也没有列表、引用或代码。
    为了符合“长文用`

    `/`

    `划分小节”以及排版要求,我需要合理地划分小节,添加`

    `或`

    `,但不能改变原文信息。 或者,如果原文就是一段,我可以加一个小标题,比如`

    SkillPlumber框架与指令级覆盖

    `,然后把原文分为几个自然段,用`

    `包裹,并对重要句子使用``。 原文可以拆分为几个逻辑段落: 段落1:介绍Agent skills和现有方法的局限性。 "Agent skills package procedural knowledge as natural-language instructions, and a skill helps only when its instructions are correct and properly followed by the agent. Self-evolving agents revise their own skills from the runs they make, but existing methods judge each run as a whole, by its outcome or by where it went wrong, and cannot tell whether a failure comes from a wrong instruction or from one the agent did not follow." 段落2:引入Instruction-level coverage和SkillPlumber。 "Instruction-level coverage was introduced to test skills and labels each instruction of a run as follow, not-follow, or not covered, which separates these causes. We leverage it for skill self-evolution with SkillPlumber, a framework in which an agent evolves its own skills where coverage shows they fall short." 段落3:介绍SkillPlumber的工作机制。 "A persistent root cause analysis agent turns each label into a revision question and checks each revision against the next run, while a maintainer distills investigations across tasks into a shared strategy library." 段落4:介绍实验结果。 "Across three benchmarks and three executors, SkillPlumber solves 64.7% to 70.5% of the tasks, 25.6 to 30.4 percentage points higher than the initial skill and 8.7 to 14.0 percentage points higher than the strongest baseline, and removing coverage reduces recovered failures by 20.8% on SkillsBench with GPT-5.5." 是否需要加`

    `?规则说“长文用`

    `/`

    `划分小节”。这段文本不算特别长,但为了体现排版能力,可以加一个总标题`

    SkillPlumber:基于指令级覆盖的技能自进化框架

    `。但注意规则4:“保留原文所有信息,不得增删改内容”。如果加了标题,算不算“增”内容? 严格来说,添加标题属于增加了原文没有的文本。为了严格遵守“不得增删改内容”,我最好不要自己编造小节标题,或者如果必须划分,只能使用原文中的词汇,但原文没有现成的标题。 其实,如果原文没有标题,我不加`

    `/`

    `也是合理的,因为“长文用...”,如果不算长文或者原文没分节,可以不用。或者我可以把第一句作为`

    `?不,那会改变句子结构。 为了安全起见,不添加原文不存在的小节标题,直接使用`

    `分段,并对关键词加``。 等等,如果原文就是一整段,我拆分成多个`

    `,算不算“改”内容?拆分段落通常被视为排版,不改变文字内容。 我将其分为3到4个自然段。 段落1:

    Agent skills package procedural knowledge as natural-language instructions, and a skill helps only when its instructions are correct and properly followed by the agent. Self-evolving agents revise their own skills from the runs they make, but existing methods judge each run as a whole, by its outcome or by where it went wrong, and cannot tell whether a failure comes from a wrong instruction or from one the agent did not follow.

    段落2:

    Instruction-level coverage was introduced to test skills and labels each instruction of a run as follow, not-follow, or not covered, which separates these causes. We leverage it for skill self-evolution with SkillPlumber, a framework in which an agent evolves its own skills where coverage shows they fall short.

    段落3:

    A persistent root cause analysis agent turns each label into a revision question and checks each revision against the next run, while a maintainer distills investigations across tasks into a shared strategy library.

    段落4:

    Across three benchmarks and three executors, SkillPlumber solves 64.7% to 70.5% of the tasks, 25.6 to 30.4 percentage points higher than the initial skill and 8.7 to 14.0 percentage points higher than the strongest baseline, and removing coverage reduces recovered failures by 20.8% on SkillsBench with GPT-5.5.

    检查规则: 1. 只输出HTML片段本身,不要Markdown围栏、不要解释。 -> 满足。 2. 每个自然段落用`

    `包裹,长文用`

    `/`

    `划分小节。 -> 这里没有添加小节,因为原文没有明显小节,且加小节标题会违反“不得增删改内容”。如果一定要加,可以把原文的某些句子提取出来,但不妥。直接分段落即可。 3. 列表内容用`
    • `,引用用`
      `,代码用`
      `,重要句子可用``。 -> 没有列表、引用、代码,使用了``。
      4. 保留原文所有信息,不得增删改内容。 -> 满足,文字完全一致,只是加了HTML标签和段落拆分。
      5. 不要输出``/``/``等文档级标签。 -> 满足。
      
      再审视一下段落拆分是否会被视为“改内容”。通常,将连续文本按语义拆分为段落是排版的一部分,不改变字符内容。
      原文本:
      "Agent skills package procedural knowledge as natural-language instructions, and a skill helps only when its instructions are correct and properly followed by the agent. Self-evolving agents revise their own skills from the runs they make, but existing methods judge each run as a whole, by its outcome or by where it went wrong, and cannot tell whether a failure comes from a wrong instruction or from one the agent did not follow. Instruction-level coverage was introduced to test skills and labels each instruction of a run as follow, not-follow, or not covered, which separates these causes. We leverage it for skill self-evolution with