SkillVine: Agent Skill Evolution via Branching Exploration

2026-09-26

AI Summary

This study addresses the limitations of linear skill updates in large language model (LLM) agents, which are prone to local optima and constrained exploration paths during skill evolution. To overcome these challenges, this work formulates skill evolution as a graph search problem and proposes a trunk-branch collaborative search mechanism. Specifically, an intelligent parent node selector is designed to balance exploration and exploitation, while an adaptive granularity update rule is introduced to optimize version evolution within the skill library. Experiments conducted across five benchmarks and two LLMs demonstrate that the proposed method effectively discovers superior skill versions, achieving state-of-the-art performance on nine out of ten task configurations. These results indicate a significant enhancement in the autonomous evolution capabilities of agent skills.

Abstract

Agent skills encapsulate reusable procedural knowledge that enables LLM agents to perform tasks, and they can be improved automatically using trajectories from interactions with the environment. This is the classic problem of skill evolution. Existing approaches predominately follow a linear evolution paradigm, in which updates are sequentially applied to the latest skill-library version. As a result, they inevitably fall into local optima, leaving many promising evolution paths unexplored. We propose SkillVine, an automatic skill-evolution framework that formulates skill evolution as a graph search problem and employs a branching exploration strategy. Equipped with a trunk-branch collaborative searching mechanism, an intelligent parent-node selector, and an adaptive-granularity update rule, SkillVine achieves a balance between exploration and exploitation. We evaluate SkillVine on 5 benchmarks with two LLMs. Results show that SkillVine discovers better skill-library versions along branches than along the linear trunk and achieves the best test performance in nine of ten benchmark-model combinations.

Problem

• Skill Evolution
• LLM Agents
• Local Optima
• Linear Evolution Paradigm
• Exploration

Innovation

• Skill Evolution
• Branching Exploration
• Graph Search
• Trunk-Branch Collaborative Searching
• Adaptive-Granularity Update