SkillCommit: Evolving Agent Skills through Behaviorally Validated Scope Expansion

Yu He, Weikai Yang

Large language model (LLM) agents can continually improve without parameter updates by converting historical experience into reusable procedural knowledge. However, existing method…