arXiv cs.CL· Xiaoyuan Li, Moxin Li, Keqin Bao, Yubo Ma, Wenjie Wang, Dayiheng Liu, Fuli Feng·· 9 天前AI 评分30
SkillGraph：通过演化技能图实现技能增强的智能体强化学习
SkillGraph: Skill-Augmented Reinforcement Learning for Agents via Evolving Skill Graphs
AI 导读
SKILLGRAPH 将可复用技能表示为有向图节点，用带类型的边编码前置、增强与共现关系，检索时返回有序技能子图以指导多步决策。该图从智能体轨迹和强化学习反馈中持续更新，使技能库与策略协同改进。在 ALFWorld、WebShop 及七个搜索增强 QA 任务上，SKILLGRAPH 相较记忆增强 RL 方法取得 SOTA，在需组合多项技能的复杂任务上提升尤为显著。
正文 · 原文
Abstract:Skill libraries enable large language model agents to reuse experience from past interactions, but most existing libraries store skills as isolated entries and retrieve them only by semantic similarity. This leads to two key challenges for compositional tasks. Firstly, an agent must identify not only relevant skills but also how they depend on and build upon each other. Secondly, it also makes library maintenance difficult, since the system lacks structural cues for deciding when skills should be merged, split, or removed. We propose SKILLGRAPH, a framework that represents reusable skills as nodes in a directed graph, with typed edges encoding prerequisite, enhancement, and co-occurrence relations. Given a new task, SKILLGRAPH retrieves not just individual skills, but an ordered skill subgraph that can guide multi-step decision making. The graph is continuously updated from agent trajectories and reinforcement learning feedback, allowing both the skill library and the agent policy to improve together. Experiments on ALFWorld, WebShop, and seven search-augmented QA tasks show that SKILLGRAPH achieves state-of-the-art performance against memory-augmented RL methods, with especially large gains on complex tasks that require composing multiple skills.
| Comments: | Under Review |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2605.12039 [cs.CL] |
| (or arXiv:2605.12039v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2605.12039 arXiv-issued DOI via DataCite |
Submission history
From: Xiaoyuan Li [view email]
[v1]
Tue, 12 May 2026 12:21:49 UTC (365 KB)
[v2]
Thu, 8 Oct 2026 09:01:47 UTC (397 KB)
来源：arXiv cs.CL · arxiv.org