What problem does it solve? When working on NVIDIA-related tasks (Jetson setup, CUDA, cuOpt routing, RAPIDS, Omniverse, NeMo training, Dynamo inference), users often don't know that a dedicated NVIDIA agent skill already exists for their workflow. This Skill acts as a capability detector and catalog router that checks the live NVIDIA skills catalog and recommends the right skill before falling back to general guidance. ## Core Features & Use Cases - Catalog-First Discovery: Queries the live NVIDIA skills catalog via npx skills add nvidia/skills --list (with GitHub and build.nvidia.com fallbacks) before naming any skill, avoiding fabricated recommendations. - Taxonomy Routing: Maps ambiguous requests to stable NVIDIA catalog lanes (Agentic AI, Physical AI, Robotics, Vision AI, Inference AI, Decision Optimization, and more) using the bundled taxonomy reference. - False-Positive Guardrails: Stays quiet on generic tasks like Express routes, React optimization, Kubernetes deploys, or video editing that merely share vocabulary with NVIDIA domains. - Use Case: A user asks how to install drivers on a Jetson board. The Skill checks the live catalog, recommends the jetson-quick-start skill with its install command, and asks for approval before installing anything. ## Quick Start Ask the agent to find an NVIDIA skill for your task, for example by saying: is there an NVIDIA skill that can help me accelerate my pandas ETL on GPUs?
