HiSkill: Empowering LLM Agents with Hierarchical Skill Graphs

2026-07-28

AI Summary

This work addresses the limitation of existing methods that treat skills in a flat manner, neglecting their inherent structural relationships and thereby disconnecting high-level skills from executable actions. To bridge this gap, the authors propose HiSkill, a novel framework that constructs a hierarchical skill graph to explicitly model decomposition, temporal, compatibility, support, and recovery relations among skills. By leveraging subgraph retrieval to guide large language model (LLM) agents in iteratively switching skills, selecting atomic operations, and generating actions, HiSkill integrates structured graph representations, symbolic state tracking, and LLM-based collaborative reasoning. Evaluated across three interactive environments, the approach significantly outperforms current baselines while reducing inference token consumption, effectively linking high-level strategic reasoning with low-level execution.

Abstract

Skills have become an important abstraction for enabling large language model (LLM) agents to reuse past experience in long-horizon interactive tasks. However, existing trajectory-to-skill methods often produce flat collections of high-level textual skills that are stored and retrieved independently, leaving skill relations underutilized and maintaining a gap between high-level skills and executable actions. In this paper, we propose HiSkill, a hierarchical skill graph framework that organizes interaction trajectories into a directed graph with skill nodes, AtomicOp nodes, and typed edges. Specifically, the graph connects reusable high-level skills with executable action templates, while also capturing decomposition, temporal transition, compatibility, support, and recovery relations among them. At inference time, HiSkill retrieves a compact task-relevant subgraph and performs subgraph-guided task execution, where a symbolic task state, an active skill, and the retrieved subgraph guide the LLM agent to switch skills, select AtomicOps, and ground executable actions iteratively. Experiments on three interactive environments show that HiSkill outperforms state-of-the-art baselines while reducing inference token consumption, demonstrating the effectiveness of bridging high-level skills and executable action grounding through a hierarchical skill graph. Our data and code is available at https://github.com/BUPT-GAMMA/HiSkill.

Problem

• hierarchical skill graph
• LLM agents
• skill relations
• action grounding
• long-horizon tasks

Innovation

• Hierarchical Skill Graph
• LLM Agents
• Skill Decomposition
• Action Grounding
• Subgraph-guided Execution