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

Microsoft, in collaboration with three Chinese universities, has introduced SkillOpt, a novel method that enhances the performance of GPT-5.5 using a trained Markdown file, resulting in a 23-point improvement on procedural tasks. This approach leverages traditional model training principles and allows for the optimization of instruction documents that can be effectively transferred across various models and agent environments, including Codex and Claude Code. This development is significant for practitioners as it offers a lightweight, adaptable solution for improving AI agent performance without extensive retraining.