Microsoft's SkillOpt boosts GPT-5.5 by using nothing but a trained Markdown file
23 points. That's how much a single trained Markdown file just boosted GPT-5.5 on procedural tasks — and it transfers across Claude, Codex, and every other model.
Why it matters
SkillOpt demonstrates a novel optimization approach that decouples instruction design from model weights, offering a portable, low-cost way to improve agent performance across competing models. This shifts competitive advantage from model capability to instruction engineering.
The key facts
5 to knowMicrosoft + 3 Chinese universities developed SkillOpt
23-point performance boost on procedural tasks using trained Markdown file
Method transfers across models: GPT-5.5, Claude Code, Codex
Uses traditional model training principles applied to instruction documents
Optimization via Markdown file only — no model retraining required
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: Microsoft and three Chinese universities have developed SkillOpt, a method that optimizes instruction documents for AI agents using principles from traditional model training. A simple Markdown file is enough to boost GPT-5.5 by about 23 points on procedural tasks, and the same file transfers…