Traction scoreGitHub stars can be faked, so popularity alone can be misleading. Traction Score looks for broader signs of recent attention, adoption, and active maintenance.
This repository provides structured implementation guides for developers leveraging NVIDIA hardware to accelerate compute-intensive tasks. It focuses on transitioning data workloads from CPUs to GPUs, offering specific patterns for high-performance data manipulation and mathematical optimization.
Data engineers can use these resources to refine ETL processes, joins, and Parquet/CSV pipelines using cuDF. Quant developers can apply the guidance to solve routing problems, linear programming, and portfolio optimization via cuOpt and cuFOLIO.
Compatible AgentsThe repository documents support for these agents. The skills may also work with other agents that can load SKILL.md files, but they may need some setup or small changes.
Follow the documented setup, then try a first task.
I’m using
Interactive NVIDIA skill selection
Installs: One user-selected skill from the NVIDIA skills catalog, with an interactively selected installation destination. · Claude Code, Codex, other compatible coding agents
Before you start
Current skills CLI v1.5.16 or newer.
Run this terminal command, then choose a skill and installation destination when prompted.
npx skills add nvidia/skills
Allow the agent to load the installed skill when it next loads skills. To use it in the current Claude Code session, enter `/reload-skills` in the Claude Code agent session.
Allow the agent to load the installed skill when it next loads skills. If the destination is Claude Code and you need it in the current session, enter `/reload-skills` in the Claude Code agent session.
The skill will be available when Claude Code next loads skills. To load it in the current session, enter `/reload-skills` in the Claude Code agent session.
Repository containing the skill and its skill_manifest.yaml, opened at the repository root
<PATH_TO_DICOM>: path to one DICOM file
<CALLER_PROVIDED_RUN_DIRECTORY>: directory where output may be written
Python packages declared in runtime.side_effects.pip_packages
Use the dicom-metadata-extract skill to extract selected metadata from the single DICOM file at <PATH_TO_DICOM>. Keep any output under <CALLER_PROVIDED_RUN_DIRECTORY>, report the transfer syntax, modality, phi_present value, and phi_tags_found, and clearly note that this is neither anonymization nor clinical review. Read skill_manifest.yaml first and use the documented wrapper rather than writing a replacement.
<FILE_PATH>: path to the proposed NeMo-RL documentation or source file
The surrounding docs/index.md and relevant existing documentation for placement comparison
Use the nemo-rl-docs skill to review the proposed documentation or docstring change at <FILE_PATH>. Give a short list of required corrections: whether docs/index.md must be updated, whether the document is in the most appropriate section, and whether public class or function docstrings follow Google style. Do not edit files yet.
<CODE_PATH>: path to the relevant concurrent request, worker, or agent implementation
<LANGUAGE>: Python, Rust, Go, or Node.js
A brief description of the request, task, thread, worker, goroutine, or agent boundaries
Use the nemo-relay-instrument-context-isolation skill to review the concurrent request handling in <CODE_PATH> for <LANGUAGE>. Identify any shared mutable scope stack, missing propagation across execution boundaries, or scope-local middleware that could leak between requests. Return a concise risk list and recommended isolation pattern without changing the code.
Claude
Codex
Cursor
OpenClaw
OpenCode
Kiro
Hermes