Back to Skills

hsb-setup

Clone the latest NVIDIA Holoscan Sensor Bridge repo, ask which supported devkit is being used, configure the host per platform, build the correct demo container, run it, and verify HSB connectivity by pinging 192.168.0.2. Use for Holoscan Sensor Bridge setup, build, container launch, and first-connectivity bring-up.

2,941stars340forksUpdated 8/14/2026

Security Assessment

Safe(90/100)
Security Score90/100

About hsb-setup

This skill brings up an NVIDIA Holoscan Sensor Bridge demo environment end to end. It clones the latest HSB repository, asks which supported devkit is in use, auto-detects the platform, configures the host per platform, builds the correct demo container, runs it, and verifies HSB connectivity by pinging the board at a fixed address and reading the FPGA version. It solves the multi-step, platform-specific setup problem of getting a sensor bridge devkit from a fresh state to a working, connected demo.

The workflow is interactive and gated. It reads environment variables (SSH_TARGET and REMOTE_ROOT are required; HSB_PLATFORM and a custom HSB_REPO URL are optional), presents an explicit phase plan the user must acknowledge, and runs a mandatory token-budget preflight before any remote command. It then confirms the physical setup, establishes SSH to the devkit, detects the platform by reading the product name and mapping it to a build mode, clones or refreshes the repository from the public NVIDIA repo (or a user-supplied URL), studies the repo user guide, configures the host and network with the correct container build flags per platform (dGPU vs iGPU), builds and runs the demo container, verifies board connectivity, and checks the FPGA version against the host software — suggesting the hsb-flash skill if they mismatch. It reports issues per phase, attempts safe fixes, offers to export a report, and finally stops apps and hands control back to the user.

It targets NVIDIA Holoscan hardware developers setting up sensor bridge devkits (IGX Orin, AGX Orin, AGX Thor, DGX Spark). It has side effects on the host (installing configuration, building containers, using sudo for privileged commands), so it is non-autonomous and runs only when explicitly invoked. A benchmark documents that it passed NVIDIA's NVSkills-Eval.

FAQ

What does it set up?

It clones the HSB repo, detects and confirms the platform, configures the host and network, builds and runs the correct demo container, and verifies connectivity by pinging the board and reading the FPGA version.

What are the prerequisites?

SSH_TARGET and REMOTE_ROOT must be set (it stops and asks otherwise). A powered, connected devkit with the correct OS and network access is assumed. HSB_PLATFORM and a custom HSB_REPO URL are optional.

Which platforms are supported?

IGX Orin (dGPU and iGPU), AGX Orin, AGX Thor, and DGX Spark, each mapped to the correct docker build flag; it asks whether an IGX Orin is iGPU or dGPU when unclear.

Does it modify the host system?

Yes. It configures the host and network, uses sudo for privileged commands, and builds and runs containers, so it has side effects and runs only when the user explicitly invokes it, prompting for confirmation before changes.

What if the FPGA version doesn't match?

If the FPGA version on the sensor does not match the HSB host software, it suggests using the hsb-flash skill to flash the board to the proper FPGA version.

All Files

15 files
linux/bashrc_hsb.sh3.0 KB
View
SKILL.md19.1 KB
View
windows/set-hsb-env.ps11.6 KB
View
docs/failure-playbook.md10.8 KB
View
references/phase-details.md55.4 KB
View
windows/Microsoft.PowerShell_profile.ps12.6 KB
View
BENCHMARK.md3.9 KB
View
linux/set-hsb-env.sh1.9 KB
View
skill.oms.sig6.8 KB
View
evals/evals.json1.0 KB
View
skill-card.md4.1 KB
View
windows/run-hsb.cmd0.3 KB
View
docs/platform-mapping.md2.6 KB
View
scripts/hsb_phase_runner.sh0.8 KB
View
windows/README-WINDOWS.md1.8 KB
View

Install hsb-setup

Download and extract the skill files to your .claude/skills/ directory.

Quick Setup:

  1. Copy the skill folder to .claude/skills/
  2. Claude will automatically detect and use the skill

Repository

nvidia/skills