20260924_skillsbench_neurips

🎉 “SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks” has been accepted to the NeurIPS 2026 Evaluations and Datasets Track (acceptance rate 25.8%)! SkillsBench pairs 87 tasks across 8 domains with curated Agent Skills and finds that Skills raise the average agent pass rate from 33.9% to 50.5%. Thanks to the 70+ co-authors led by Xiangyi Li. arXiv:2602.12670