skills: a catalog of verified instruction sets that teach AI coding agents to use NVIDIA software and hardware optimally

What it solves

AI coding agents often struggle to use complex, specialized software libraries and hardware platforms optimally. This project provides a catalog of "skills"—portable instruction sets—that teach agents how to correctly implement workflows for NVIDIA's ecosystem, including robotics, simulation, CUDA-X libraries, and AI Blueprints.

How it works

Skills are standardized instruction sets that can be installed into various AI coding agents (such as Claude Code, Codex, Cursor, and Kiro) using a dedicated CLI. When an agent encounters a task related to a specific NVIDIA product, the installed skill guides the agent through the correct APIs and workflows. The repository acts as a central catalog where skills from various NVIDIA product repositories are mirrored daily via an automated pipeline.

Who it’s for

Developers using AI coding agents who need to build applications with NVIDIA software, such as those working with GPU-accelerated dataframes (cuDF), quantum computing (CUDA-Q), or robotics (Isaac).

Highlights

• Broad Ecosystem Support: Covers a wide range of NVIDIA tools including cuOpt for optimization, NeMo for LLM training, and Jetson for edge AI.
• Agent Agnostic: Compatible with multiple coding agents including Claude Code, Codex, Cursor, and Kiro.
• Automated Sync: Skills are mirrored daily from product-specific repositories to ensure they remain up-to-date.
• Easy Installation: Uses a simple CLI flow (npx skills add) to install specific capabilities directly into the agent's environment.