Toward Skill-Native LLMs: Skill Entropy for Benchmarking and Training Long-Horizon Reasoning Paper • 2608.05139 • Published Aug 5 • 27