Guides Holoscan SDK installation: inspects the host, assesses platform compatibility, recommends an install method, and delegates to the matching install skill.
The holoscan-setup skill is a decision-and-delegation guide that determines the correct Holoscan SDK installation method for a given host and then hands off to the matching method-specific install skill. It solves the problem of choosing among five install paths (NGC container, Debian/apt, pip wheel, Conda, source) across a wide matrix of platforms including Ubuntu, RHEL, IGX Orin, Jetson, and DGX Spark / Grace-Hopper, where the right choice depends on hardware, OS, CUDA driver, and existing tooling.
The skill runs a conversational, step-by-step workflow: it first fetches the official docs, then inspects the machine with read-only commands (uname, lsb_release, nvidia-smi, nproc, free), assesses platform compatibility against a support matrix, checks installed tooling, and verifies Docker GPU passthrough itself rather than asking the user. It ships two helper scripts, check_conda.sh (detects Conda installs even when not on PATH by searching common install directories and shell rc files) and check_ngc_image.sh (checks whether an NGC container image for a given CUDA tag is pulled or available). The workflow ends with a single bolded recommendation, then stops and asks the user which method to use before any install command is issued, deliberately deferring the actual install to the delegated skill.
Target users are developers and engineers standing up Holoscan on supported NVIDIA platforms who want host inspection and install-method selection automated and grounded in current documentation rather than hardcoded assumptions.
No. It inspects the host, recommends an install method, then stops and asks which method to use, delegating the actual install to a method-specific skill. It explicitly avoids pasting docker pull, apt install, or pip install commands in the recommendation turn.
Ubuntu 22.04/24.04 and RHEL 9.x on x86_64, IGX Orin, Jetson AGX Orin / Orin Nano / Thor, and DGX Spark / Grace-Hopper, each with a documented set of available install methods.
For a first-time user on a supported x86_64 host with Docker available, the recommendation must be the NGC Container, since it bundles all dependencies and is the fastest path to a working install.
check_conda.sh detects Conda installs even when not on PATH by searching common directories and shell rc files and reporting which envs have holoscan importable; check_ngc_image.sh checks whether the NGC Holoscan image for a CUDA tag suffix (cuda13, cuda12-dgpu, cuda12-igpu) is pulled or available.
A Linux host on a supported platform, an NVIDIA GPU with a working driver, network access to docs.nvidia.com and NGC, and one of Docker + NVIDIA Container Toolkit, apt, Python 3.10-3.13 with pip, Conda, or a build toolchain depending on the chosen method.
Quick Setup:
.claude/skills/Repository
nvidia/skills