Search papers, labs, and topics across Lattice.
This paper introduces VCE-Skill, a novel approach that enhances skill self-evolution by integrating public skill version histories with current task execution trajectories. By distilling public skill changes into structured version-change experiences and adaptively fusing them with task-specific proposals, VCE-Skill effectively leverages external knowledge while maintaining relevance to current tasks. Experimental results indicate that this method significantly boosts skill self-evolution performance, with mean score improvements ranging from 3.20 to 4.98 points and enhanced cross-model transfer capabilities.
Public skill version histories can dramatically enhance task-specific skill evolution, leading to significant performance gains in AI agents.
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.
