microsoft/SkillOpt

This page introduces an external open-source repository. It is not an HDATF product.

What it does

SkillOpt trains reusable natural-language skills for LLM agents without changing model weights. It edits a skill based on agent trajectories, accepts updates only when they pass validation, and outputs the best version as a best_skill.md file.

How it helps ATF

Worth comparing for Harness, which checks results with evidence and hands work to executors. Accepting a skill change only after it passes validation could be a reference for improving the instructions given to executors.

License

MIT Permissive. Commercial use and changes are allowed if the copyright notice is kept.

More in this category

• shadcn/improve
  An agent skill that audits a codebase and writes self-contained implementation plans as markdown files for cheaper models to execute. The skill does not write code itself, and an execute mode dispatches a cheaper executor and reviews its work.
• Leonxlnx/taste-skill
  Taste Skill is a set of portable Agent Skills that push AI coding agents toward better layout, typography, motion and spacing in the interfaces they build. It also includes image-generation skills for web, mobile and brand reference boards.
• DietrichGebert/ponytail
  A skill for AI coding agents such as Claude Code that pushes them to write as little code as possible instead of over-building. The readme reports less code and lower cost than the same agent without the skill.
• hardikpandya/stop-slop
  Stop Slop is a skill file that teaches Claude or another LLM to find and remove patterns typical of AI writing, such as banned phrases and structural clichés. It scores prose on directness, rhythm, trust, authenticity and density.
• xbtlin/ai-berkshire
  Investment research skills for Claude Code and Codex that structure the methods of four value investors. Multiple agents research in parallel and challenge each other, and reports must end in a clear pass, fail or gray-zone verdict.

Only repositories in the ranked Trendshift lists are included, and the lists are used only to find candidates. We do not copy their ranks. Descriptions, licenses and star counts come from each GitHub repository. The notes are our own reading. We have not tested these projects, and a place on a trending list does not prove quality.