SkillCAT: Contrastive Assessment and Topology-Aware Skill Self-Evolution for LLM Agents

SkillCAT is a new framework for skill self-evolution in LLM agents that decouples the skill extraction and evaluation process into three distinct stages: Contrastive Causal Extraction (CCE), Assessment-Augmented Evolution (AAE), and Topology-Aware Task Execution (TTE). It leverages multiple trajectories for each task to identify effective skill patches and compiles them into a routable sub-skill topology, optimizing inference by loading only relevant capabilities. Evaluated on benchmarks like SpreadsheetBench and DocVQA, SkillCAT improves performance by up to 40.40% over existing methods, offering a training-free approach to enhance agent capabilities.