NVIDIA open-sources SkillEvaluator to vet AI agent skills before deployment

anthara_ai · x · 2026-11-19

NVIDIA released SkillEvaluator, an open-source (Apache 2.0) framework for evaluating AI agent skills and plugins before integrating them into real workflows. Already at 2.7k GitHub stars, it runs quality gates, detects semantic overlap, generates synthetic evaluation datasets, and performs live agent evaluation comparing runs with and without a skill to measure its actual effect on task results. Ships with an arXiv paper (2608.20614), docs, and a developer livestream.

Original post →

More from coding & agent

coding & agent channel →