hsb-app

This skill discovers and runs NVIDIA Holoscan Sensor Bridge (HSB) example applications on a connected devkit. It solves the problem of matching the many available HSB example apps to a user's specific hardware: it filters apps by platform, HSB software version, board type (for example Lattice), and connected sensors (for example dual IMX274 or VB1940), then runs the chosen app inside the demo container, monitors output, analyzes failures, suggests code or environment edits, and supports iterative re-runs.

The workflow is heavily gated and interactive. It requires SSH_TARGET and REMOTE_ROOT environment variables, prints resolved settings, and presents a phase plan (verify connectivity, discover setup and select app, run with monitoring and debugging, generate a session report) that the user must explicitly confirm before each phase. It connects to the devkit over SSH in batch mode, pings the HSB board at a fixed address, verifies the demo container image, runs `hololink enumerate` to read board identity, and launches Python example apps via `docker run` inside the demo container. It supports a session fast-path that skips re-verification when prior state is recorded, timed execution with a watchdog timeout, and a saved session report of issues and fixes. It is explicitly non-autonomous: model invocation is disabled and it runs only when the user invokes it.

It targets NVIDIA Holoscan hardware developers and QA engineers working with sensor bridge devkits who want a guided way to find and run compatible example applications. It assumes the devkit is already set up and defers initial bring-up to a companion hsb-setup skill. A benchmark file documents that it passed NVIDIA's NVSkills-Eval, including a 100% security dimension score.