VCE-Skill: Enhancing Skill Self-Evolution with Version-Change Experience

2026-08-17

AI Summary

This study addresses the limitation of existing agent skill evolution approaches that rely solely on current trajectories while neglecting public version history knowledge. To overcome this, we propose VCE-Skill, a novel framework that pioneers the distillation of public skill version changes into structured prior experience. Through an adaptive fusion mechanism, these priors are integrated with task trajectories to drive efficient self-evolution. Experimental results demonstrate that VCE-Skill improves average skill self-evolution scores by 3.20–4.98 points and significantly enhances cross-model transferability. These findings effectively validate the critical role of integrating external evolutionary knowledge with task-specific evidence in advancing agent capability development.

Abstract

Agents increasingly rely on reusable skills to encode task knowledge, tool-use procedures, and validation rules. Existing skill self-evolution methods primarily revise skills using execution trajectories collected from current tasks, leaving the evolution knowledge accumulated in public skill version histories largely untapped. Our pilot study reveals a clear complementarity between the two sources: public skill changes provide reusable evolution priors, whereas trajectories provide evidence grounded in the current task. Motivated by this, we propose VCE-Skill, which distills noisy and implementation-specific public skill changes into reusable, structured version-change experience and adaptively fuses it with trajectory-derived proposals from the base evolver, thereby exploiting external experience while retaining task-specific evidence. Extensive experiments demonstrate that VCE-Skill improves skill self-evolution, increasing mean scores by 3.20--4.98 points; transfer experiments further show that the resulting skills achieve stronger cross-model transfer performance. Our work highlights public skill version changes as a previously underexplored yet effective source of prior knowledge and advances trajectory-driven skill self-evolution.

Problem

• Skill Self-Evolution
• Version-Change Experience
• Agent Skills
• Execution Trajectories

Innovation

• Skill Self-Evolution
• Version-Change Experience
• Adaptive Fusion
• Cross-Model Transfer
• Evolution Priors