How to Authenticate Agent Skills Before Execution | Nemotron Labs

As agent workflows grow, so does the list of skills they pull in but how do you know what you're actually running? This session takes a hands-on look at agent skill verification, what to look for before a skill enters your workflow, and how NVIDIA Verified Skills give you the provenance, scanning, and signing guarantees you need to run agents you can trust. We'll cover what makes a skill trustworthy from provenance metadata and scan results to cryptographic signing and the Skills Card format and how to build that verification into your workflow before execution. What you'll learn: - How to evaluate agent skills for trust and provenance before adding them to your workflow - How NVIDIA Verified Skills and Skills Cards surface the metadata you need to make confident decisions - How to use the NVIDIA skill card generator to build a verifiable skill from scratch - How NemoClaw and OpenShell enforce runtime sandbox boundaries independent of skill verification Have questions about securing your agent's capability layer? Bring them live we'll answer them in real time.