Microsoft SkillOpt upgrades AI agent skills automatically

Microsoft has introduced SkillOpt, an open-source framework designed to automatically upgrade AI agent skills without altering the underlying model weights. Traditional skill optimization involves manual, trial-and-error updates to text-based instructions, a slow and error-prone process. SkillOpt treats agent skill documents as trainable objects, using deep-learning-style optimization to systematically explore and implement performance-enhancing modifications. This approach ensures changes are mathematically sound, preventing performance regressions and improving accuracy significantly across various benchmarks, including models like GPT-5.5. The resulting compact, transferable skill artifacts allow AI agents to adapt more effectively to new domains and complex enterprise workflows.