SkillDAG: Self-Evolving Typed Skill Graphs for LLM Skill Selection at Scale
The paper introduces SkillDAG, a novel approach for selecting skills in large language models (LLMs) by modeling inter-skill relationships as a typed directed graph. This method allows for dynamic retrieval and evolution of skills during execution, improving performance metrics significantly. SkillDAG outperforms existing methods, demonstrating enhanced candidate ranking and recall capabilities.
- ▪SkillDAG models inter-skill relationships as a typed directed graph.
- ▪The approach allows for dynamic retrieval and evolution of skills during execution.
- ▪SkillDAG achieved a 67.1% success rate and a 27.3% reward, surpassing previous benchmarks.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2606.03056 (cs) [Submitted on 2 Jun 2026] Title:SkillDAG: Self-Evolving Typed Skill Graphs for LLM Skill Selection at Scale Authors:Tong Bai, Zhenglin Wan, Pengfei Zhou, Xingrui Yu, Wangbo Zhao, Yang You, Ivor W. Tsang View a PDF of the paper titled SkillDAG: Self-Evolving Typed Skill Graphs for LLM Skill Selection at Scale, by Tong Bai and 6 other authors View PDF HTML (experimental) Abstract:As LLM agents adopt large skill libraries, selecting the right subset becomes a structural problem rather than a similarity-matching one: skills depend on, conflict with, specialize, or duplicate one another, a structure invisible to both full enumeration and embedding similarity.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
