GuideSkill: Evolving Executable LLM Agent Skills for Guideline-Grounded Clinical Reasoning
GuideSkill: Evolving Executable LLM Agent Skills for Guideline-Grounded Clinical Reasoning
GuideSkill：面向临床指南推理的可执行大模型智能体技能演进
Abstract: Clinical practice guidelines (CPGs) encode diagnostic criteria, but LLM systems typically retrieve guideline text or absorb it through training rather than execute its rules. We introduce GuideSkill, an external reasoning layer that compiles disease-specific criteria into executable functions returning ordinal diagnostic-support scores.
摘要： 临床实践指南（CPGs）中编码了诊断标准，但大语言模型（LLM）系统通常只是检索指南文本或通过训练吸收其内容，而非真正执行其中的规则。我们引入了 GuideSkill，这是一个外部推理层，它将特定疾病的诊断标准编译为可执行函数，并返回序数诊断支持评分。
GuideSkill-Zero is initialized from guidelines, while GuideSkill-Evo uses case—diagnosis pairs to refine covered skills and add missing diagnoses. At inference, an LLM proposes a differential diagnosis, grounds the features required by each matched skill, and fuses its ranking with the executed skill scores.
GuideSkill-Zero 基于指南进行初始化，而 GuideSkill-Evo 则利用病例-诊断对来优化已涵盖的技能并补充缺失的诊断。在推理阶段，大模型会提出鉴别诊断，提取每个匹配技能所需的特征，并将模型自身的排序与执行后的技能评分进行融合。
Across four benchmarks and four backbones, GuideSkill-Zero improves macro-average accuracy over guideline RAG by 13.45% on average. GuideSkill-Evo achieves the highest macro-average for every backbone, improves over direct inference by 18.49% relatively, and increases gold-label skill coverage from 56.5% to 99.5%.
在四个基准测试和四个骨干模型上，GuideSkill-Zero 的宏平均准确率平均比基于指南的检索增强生成（RAG）提高了 13.45%。GuideSkill-Evo 在所有骨干模型上均达到了最高的宏平均准确率，相对直接推理提升了 18.49%，并将金标准技能覆盖率从 56.5% 提高到了 99.5%。
On Qwen3.5-9B, it also exceeds the strongest parameter-update baseline by 11.16% without updating the backbone. Expert evaluation further indicates that GuideSkill produces clinically sound and broadly acceptable skills, suggesting that its initialized and evolved rules are reliable and practically meaningful.
在 Qwen3.5-9B 模型上，它在不更新骨干模型参数的情况下，比最强的参数更新基线高出 11.16%。专家评估进一步表明，GuideSkill 生成的技能在临床上是合理的且具有广泛的可接受性，这表明其初始化和演进后的规则是可靠且具有实际意义的。
These results support executable skills as a model-agnostic mechanism for combining guideline-derived procedures with case-derived diagnostic patterns.
这些结果证明，可执行技能作为一种模型无关的机制，能够有效地将源自指南的程序与源自病例的诊断模式相结合。