SkillOpt: Turning AI Agent Skills into Trainable Parameters

Microsoft researchers introduced SkillOpt, a method that treats an agent's skill file as a trainable parameter outside the frozen model. The approach achieved best or comparable results in all 52 evaluation cells across six benchmarks with seven models, including GPT-5.5. Optimized skills are compact, readable, and transferable across models and runtime environments.

Microsoft OpenAI Alibaba/Qwen