Model reports

SkillsBench

Measures whether curated Agent Skills raise agent pass rates, not raw model capability; the meaningful figure is the paired no-Skills versus curated-Skills gap. The release date follows the paper's first submission.