Microsoft's SkillOpt boosts GPT-5.5 by using nothing but a trained Markdown file

Microsoft and researchers from three Chinese universities have developed SkillOpt, a method that uses formatted Markdown instructions to improve the performance of large language models on procedural tasks. By applying training principles to these documents, the technique increased GPT-5.5's scoring by 23 points. This approach demonstrates that AI agents can be significantly optimized through structured documentation rather than just additional model weights, allowing these improvements to be easily transferred across different systems.