독서리서치레이더투자 프레임워크
로그인 / 회원가입
← 읽기로 돌아가기
뉴스주제 · SkillDAG技能图중대1 분 소요DisconfirmAI 편집부

SkillDAG: Exposing Typed Skill Graphs to LLM Agents Boosts ALFWorld Success by 14.3 Points

New research introduces SkillDAG, which allows LLM agents to directly query and revise skill graphs, overcoming dependency issues in traditional retrieval and significantly improving task success rates in benchmarks.

2026-10-16 12:00 GMT+8

SkillDAG represents a new approach to skill selection for Large Language Model (LLM) agents. Instead of treating skills as independent documents for vector retrieval, it constructs a typed skill graph and exposes it as an agent-callable interface. This method addresses the limitations of enumeration or Top-K embedding retrieval, which often fail to capture dependencies, conflicts, and specializations between skills, leading agents to omit necessary prerequisites during task execution.

Experiments across two benchmarks and three model families show that SkillDAG outperforms full-library prompting, vector retrieval, and Graph-of-Skills. Specifically, it achieved a 14.3-point increase in task success rate in the ALFWorld environment. Furthermore, its ranking remained stable even when the skill pool expanded tenfold, whereas fixed pipeline methods degraded under similar conditions.

These results are self-reported by the author team and have not yet been independently reproduced. The core contribution lies in demonstrating that allowing agents to actively query and revise skill structures is more scalable and effective than having a retriever passively provide opaque bundles of skills, offering a new architectural direction for managing large-scale LLM agent capabilities.

출처

  • arxiv.org↗
로그인 / 회원가입
독서리서치레이더투자 프레임워크
주제

SkillDAG技能图

【背景】随着LLM智能体依赖的技能库规模扩大至数千条,传统基于向量相似度的检索方法因忽略技能间的依赖、冲突等结构关系而失效。新研究提出SkillDAG,将技能库建模为有类型的有向无环图(DAG),并作为可调用接口直接暴露给LLM,而非隐藏在…

3 단계0 건업데이트됨 2026-10-17 13:43

현재까지의 과정

  1. 2026-06-02SkillDAG论文v1版本提交至arXiv,提出将技能图作为智能体可调用接口的核心概念。arxiv.org ↗
  2. 2026-10-15论文更新至v3版本,完善实验数据,确认在ALFWorld上获得+14.3分提升,并在EMNLP 2026 Findings发表。arxiv.org ↗
  3. 2026-10-17官方代码仓库及复现数据集发布,提供完整的基准测试脚本与预计算嵌入缓存。github.com ↗

아직 해결되지 않은 문제

  • 长期自我编辑的可靠性:智能体在线修改图结构是否会在极长周期任务中引入累积误差?
  • 更广泛的基准覆盖:目前主要在ALFWorld和SkillsBench验证,其在真实世界复杂API调用场景中的表现如何?
  • 公平比较细节:不同推理能力底座模型对SkillDAG增益的敏感度差异尚未完全量化。
전체 주제 읽기 →
리서치 계속하기 →