Researchers have developed AUSO (Action-Level Unified Skill Optimization), a method for learning and using skills in AI agents. Unlike existing methods that either keep skills separate from the model or fully internalize them, AUSO unifies skill learning and use through a progressive optimization process. This allows skills to transition from external guidance to decision knowledge whose utilization is adapted to its action-level benefit. Experiments show that AUSO improves agent performance and out-of-distribution generalization over competitive baselines. --- Why it matters: AUSO matters because it addresses the limitations of existing skill learning methods, which can lead to suboptimal performance in complex tasks. By unifying skill learning and use, AUSO enables AI agents to adapt their skills more effectively, leading to better performance and generalization. Source: https://arxiv.org/abs/2608.21292