NVIDIA Agent Skills
NVIDIA-verified skills for safer AI agent workflows across CUDA-X, NeMo, Omniverse, and physical AI
About NVIDIA Agent Skills
NVIDIA Agent Skills is NVIDIA's public catalog of verified AI-agent skills: portable instruction sets that teach coding agents how to use NVIDIA CUDA-X libraries, AI Blueprints, NeMo, TensorRT-LLM, Omniverse, physical-AI workflows, RAG, data, and inference tools. NVIDIA documents provenance, risk scanning, skill cards, detached signatures, and a public GitHub catalog so teams can inspect what an agent skill does before adding it to Claude Code, Codex, Cursor, or other compatible agents.
Features
Pros & Cons
Pros
- +Official NVIDIA source with public GitHub repo
- +Adds provenance, signatures, risk scanning, and skill cards to agent instructions
- +Covers concrete GPU, simulation, RAG, data, and physical-AI workflows
- +Works as vendor-specific operational guidance instead of generic prompt snippets
- +Useful for teams evaluating agent skills before Claude Code, Codex, or Cursor rollout
Cons
- −A verified catalog is not the same as a safe local install for every skill
- −Some skills may require NVIDIA hardware, platform accounts, or product-specific setup
- −Teams still need to inspect permissions, dependencies, and generated actions before use
- −Evaluation and quality metrics are still rolling out according to NVIDIA docs
Platforms
Tags
Related AI Concepts
Similar Tools
Semantic Kernel
Microsoft's AI orchestration SDK for building agents with .NET, Python, and Java
Free (open-source)Anthropic MCP
Model Context Protocol — universal standard for AI tool integration
Free (open standard)Vercel AI SDK
TypeScript toolkit for building AI-powered web applications
Free (open-source)Instructor
Structured outputs from LLMs using Pydantic
Free (open-source)Need help choosing?
Compare NVIDIA Agent Skills with alternatives side by side
Compare Tools →