Ask the Experts: Evaluating Agent Skills | Nemotron Labs

Tuesday, September 22, 1:00pm - 2:00pm (EDT)

Link:

Time shown in -04:00 America, New York

Agent skills package instructions, instructions, examples, and tool guidance for autonomous agents, but before a skill enters a real workflow, teams need to understand its security risks, overlap with existing skills, and measure usefulness. In this Ask the Experts session, the NVIDIA product and engineering team behind SkillEvaluator and SkillSpector will unpack the three-tier evaluation framework, share benchmark results from 300+ verified skills across 30+ NVIDIA products, and cover how SkillSpector scans for agent-native risks including prompt injection, data exfiltration, trigger abuse, and tool poisoning. SkillSpector is a standalone security scanner for agent skills, combining static analysis with optional LLM semantic analysis. Teams can use it independently or integrate it into broader workflows, including SkillEvaluator’s Tier 1 validation. Expect a look at SkillEvaluator in action.
What you'll learn:

How Skill Lift is calculated and what 300+ verified NVIDIA skill benchmarks reveal about skill effectiveness
How SkillSpector scans for agent-native security risks before a skill enters any workflow
How to use token and execution time data to identify skills that need optimization

Building skills or evaluating your own agent workflows? Drop your questions live, and our skill experts will answer them in real time.

Add to Calendar 2026/09/22 10:00:00 2026/09/22 11:00:00 America/Los_Angeles Ask the Experts: Evaluating Agent Skills | Nemotron Labs Agent skills package instructions, instructions, examples, and tool guidance for autonomous agents, but before a skill enters a real workflow, teams need to understand its security risks, overlap with existing skills, and measure usefulness. In this Ask the Experts session, the NVIDIA product and engineering team behind SkillEvaluator and SkillSpector will unpack the three-tier evaluation framework, share benchmark results from 300+ verified skills across 30+ NVIDIA products, and cover how SkillSpector scans for agent-native risks including prompt injection, data exfiltration, trigger abuse, and tool poisoning. SkillSpector is a standalone security scanner for agent skills, combining static analysis with optional LLM semantic analysis. Teams can use it independently or integrate it into broader workflows, including SkillEvaluator’s Tier 1 validation. Expect a look at SkillEvaluator in action.
What you'll learn:

How Skill Lift is calculated and what 300+ verified NVIDIA skill benchmarks reveal about skill effectiveness
How SkillSpector scans for agent-native security risks before a skill enters any workflow
How to use token and execution time data to identify skills that need optimization

Building skills or evaluating your own agent workflows? Drop your questions live, and our skill experts will answer them in real time. https://www.youtube.com/watch?v=4rI0zATFxYI&themeRefresh=1 Amanda Dalton amandad@nvidia.com false 30

https://www.youtube.com/watch?v=4rI0zATFxYI&themeRefresh=1

Amanda Dalton, amandad@nvidia.com