NVIDIA Developer

Ask the Experts: Evaluating Agent Skills | Nemotron Labs

Published 2026-09-22 · Duration 55:19

Summary

This session details the evaluation of autonomous AI agent skills using NVIDIA's open-source framework, SkillEvaluator. The system provides a comprehensive, multi-tier pipeline to ensure that skills are safe, reliable, and genuinely useful before deployment. The evaluation process covers static security scanning (detecting risks like prompt injection and data exfiltration), deduplication checks, and live agent testing in a sandbox environment. This framework is critical for establishing quality gates in AI-driven workflows.

Download summary

Key takeaways

  1. Three-Tier Evaluation Framework 2:00

    Skill evaluation utilizes a three-tier framework: 1) Static/Security checks (using SkillSpector), 2) Deduplication checks against an internal registry, and 3) Live Agent Testing, which assesses if the skill improves task performance in a sandbox environment.

  2. Security Scanning for Agent Risks 4:00

    SkillSpector is a standalone security scanner that checks for agent-native risks, including prompt injection, data exfiltration, trigger abuse, and tool poisoning. It combines static analysis with optional LLM semantic analysis.

  3. Measuring Skill Utility (Skill Lift) 7:30

    The evaluation measures 'Skill Lift' across five dimensions: efficiency, correctness, effectiveness, discoverability, and security. This determines if the skill genuinely helps the agent perform the task better compared to a baseline.

  4. Live Agent Testing Methodology 9:00

    Live agent evaluation runs the agent in a secure sandbox, performing an A/B test by comparing the agent's performance when the skill is available versus when it is restricted. This confirms if the skill is functionally necessary for the task.

Technical details

  • SkillEvaluator CLI Usage 200s

    The primary tool is the SkillEvaluator CLI. To start, users must install the CLI and configure environment variables for the LLM provider (e.g., `build.nvidia.com` or `open-anthropic`). The core validation command is `skill_evaluator validate <skill_name>`.

  • Evaluation Data Preparation 1000s

    Agent evaluation requires a dataset. Users can provide a custom dataset via an `eval.json` file, or utilize the CLI's auto-pilot mode, which analyzes the skill to create a suitable dataset.

  • Multi-Agent/Plugin Architecture 1650s

    The system is evolving beyond single skills to 'plugins' that run across multiple agents and harnesses. This allows for assessing cross-references and webhooks, which is a more complex task than evaluating a single skill.

  • Secure Runtime Environment 2100s

    All agent runs, especially those involving evaluation, must occur in a secure, sandboxed mode to prevent external connections, data leakage, or unauthorized actions.

Mentioned resources

  • SkillEvaluator (Open Source Tool/CLI)
  • SkillSpector (Security Scanner)
  • NVIDIA/skills repo (Skill Repository)
  • github.com/nvidia/skills (Benchmark Repository)

Channel & topics

Watch on YouTube · Back to latest

This independent, AI-assisted summary is provided for commentary and informational purposes. It may contain errors or omit important context. Please watch the original video for the creator's complete presentation. Video, thumbnail, and related copyrights belong to their respective owners.