Microsoft's SkillOpt Boosts LLM Agent Accuracy by Up to 24.8 Points Across 52 Benchmarks Without Touching Model Weights

Microsoft's SkillOpt framework dramatically boosts LLM agent accuracy by up to 24.8 points across 52 benchmarks without modifying model weights, using reusable natural-language skills and now available on PyPI for developers to deploy immediately.

• Microsoft's SkillOpt is a text-space optimizer that trains reusable natural-language skills for frozen LLM agents using trajectory-driven edits, validation-gated updates, and deployable best_skill.md artifacts without modifying model weights.
• Across six benchmarks, seven target models, and three execution harnesses, SkillOpt achieves best or tied-best performance on all 52 evaluated cells, lifting GPT-5.5 average accuracy by up to +24.8 points inside the Codex agentic loop.
• Now available on PyPI via pip install skillopt, the framework supports multiple backends including OpenAI, Claude, Qwen, and MiniMax, and introduces a new SkillOpt-Sleep module that enables nightly offline self-evolution for local coding agents.