CLAIM-2026-10-16-01
SkillForgeBench (arXiv 2610.14918 v1, submitted 10 October 2026) comprises 842 sandboxed tasks across five major professional domains and evaluates eight mainstream agents, using agent skills as unifying execution blueprints.
Why it matters
Carries a figure or a date used in the 2026-10-16 edition; the sources below are archived in this vault.
Sources
- arxiv2026skillforg
- arxivexport2026metadata