Microsoft @microsoft SkillOpt
Train reusable natural-language agent skills like neural nets — epochs, learning rates, and validation gates, without touching model weights.
- Type
- CLI
- Updated
- 2026-09-16
Install
via Positronick CLI
Official
SkillOpt is a Microsoft Research text-space optimizer that treats a compact skill document as the trainable state of a frozen LLM agent. A separate optimizer model turns scored rollouts into bounded add/delete/replace edits; a candidate is accepted only when it improves a held-out validation score. The deployed artifact is a compact best_skill.md that runs against the unchanged target model. The PyPI package ships skillopt-train, skillopt-eval, and skillopt-sleep (a nightly offline self-evolution companion). Source checkout adds benchmark configs, data materializers, and integration shells for Claude Code, Codex, Copilot, Devin, and OpenClaw. Paper: arXiv:2605.23904.