Microsoft’s SkillOpt sharpens AI agents with a single Markdown file
A small text file can make AI agents significantly sharper. Researchers from Microsoft and three Chinese universities have unveiled SkillOpt, a technique that turns a single Markdown document into a performance booster for AI models. The approach focuses on refining instruction documents that guide AI agents through tasks, applying training principles to the way those instructions are written.
How SkillOpt works
SkillOpt treats instruction documents like miniature training datasets. By structuring and formatting these documents according to learned patterns, the method increases the accuracy of AI agents on procedural tasks. The resulting Markdown file can be used across different models and agent environments without modification, making it a portable efficiency tool rather than a model-specific tweak.
What this means for developers
For teams relying on AI agents, SkillOpt offers a low-friction way to improve results. There is no need to retrain models or rewrite code; simply swap in the optimized Markdown file and observe the gains. Early results show a measurable jump in task completion rates for GPT-5.5, and the same file can be deployed with Codex, Claude Code, and other agent frameworks.
The method hints at a broader trend: fine-tuning the interface between humans and AI, rather than the AI itself. If a plain text file can deliver measurable improvements, it underscores how much performance can hinge on the clarity and structure of instructions.
Source: The Decoder. AI-assisted editorial synthesis — TechnoExpress.