microsoft/SkillOpt

Why this repo matters

• •Optimizes frozen LLM agents via text-space training without modifying underlying model weights or requiring expensive fine-tuning.
• •Produces deployable best_skill.md artifacts enabling portable, version-controlled skill reuse across different agent frameworks and environments.
• •Applies neural network training concepts like epochs and validation gates to natural language prompts for systematic agent improvement.