Improving skill optimization methods for AI agents

Researchers introduced a method allowing AI agents to improve skill prompts by ranking candidates, achieving 40% to 70% lower token costs compared to conventional methods. NVIDIA also detailed an approach to compile public agent skills into reinforcement learning environments.