NVIDIA Jetson Device Skills
13 NVIDIA skills for an agent on a Jetson device: diagnostics, memory, LLM serving, packages and video and inference benchmarks
- 168 ★ on the repository
- Last commit
Broad access
Why this level
- The jetson-headless-mode skill disables the desktop and background services, and the video skills install SDKs
- Scripts run on behalf of the agent on a live board, where a mistake can break the device
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/nvidia-jetson-device-skillsDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The jetson-headless-mode skill disables the desktop and background services, and the video skills install SDKs
- Scripts run on behalf of the agent on a live board, where a mistake can break the device
This is third-party code. Review the repository files before installing.
What it does
A set of NVIDIA agent skills for working with a Jetson device that has already booted. The agent takes hardware data from the live device with helper scripts and relies on it. The skills fall into three groups. Diagnostics and system: jetson-diagnostic takes a read-only snapshot of device identity, memory, GPU, thermals, power, storage and services, jetson-memory-audit measures memory and checks whether reclamation helped, jetson-headless-mode plans disabling the desktop and background services and applies the plan. Models: jetson-inference-mem-tune picks a runtime and memory settings, jetson-llm-serve configures LLM serving for Jetson, jetson-llm-benchmark and jetson-speculative-decoding measure and explore speed, jetson-package selects compatible packages, wheels and containers. Video: five skills check codec capabilities, install the Video Codec SDK and PyNvVideoCodec, build configurations and benchmark encoding and decoding. The install.sh script installs the skills into Claude Code, Codex and Cursor directories.
Who it is for. Embedded AI developers on NVIDIA Jetson who set up the device and model serving with Claude Code, Codex or Cursor right on the board.
Good fit when
- A Jetson is slow or hot and you want to see what holds memory and the GPU
- Freeing gigabytes of memory for a model is possible by disabling the desktop on a headless device
- A runtime and memory settings for inference are being chosen for a specific board
- You want an LLM speed measurement on llama.cpp, Ollama or vLLM with results recorded
- To see which video formats the board really encodes and decodes, the capability skill helps
Not a fit when
- Flashing and BSP customization from a host computer: NVIDIA has a separate Jetson BSP Skills set for that
- The device has not booted into a working system, so the scripts have nothing to read
- The hardware is not NVIDIA: the skills rely on tegrastats and Jetson parameters
Example request
Take a diagnostic of this Jetson: board model, free memory, temperature and services that could be turned offLimitations
The skills expect a running Jetson and run on the device itself, not on a host machine. After installation the agent must be restarted to pick up the new skills. The project is not accepting contributions for now, bugs can be reported in Issues. Some skills change the system: jetson-headless-mode disables the desktop and services and the video skills install SDKs, so confirm those actions. Installation creates symlinks, and copies need the `--copy` flag. Code is Apache-2.0, docs are CC-BY-4.0.
How to disable. Remove the links or copies of the skills from `~/.claude/skills`, `~/.codex/skills`, `~/.agents/skills` and `~/.cursor/skills`. Services disabled by jetson-headless-mode are restored with systemd tools.
FAQ
How do I install Jetson Device Skills?
On the device clone the repository and run `./install.sh`. By default it links into the Claude Code, Codex and Cursor directories, and targets are chosen with `--targets claude,cursor`. Then restart the agent.
How is this different from Jetson BSP Skills?
Device Skills work with a Jetson that has already booted: diagnostics, memory, models and video. BSP Skills handle BSP customization and flashing from a host computer.
Which skills only read device state?
jetson-diagnostic takes a read-only snapshot and jetson-video-capability checks live encode and decode capabilities against documentation. Setup skills such as headless-mode and SDK installs change the system.
Does the set work with NemoClaw and OpenClaw?
Yes. Their sandboxes have an install target: `./install.sh --targets nemoclaw --nemoclaw-sandbox jetson-skills`.
Does the project accept contributions?
Not at the moment: the README states that contributions are not accepted. Bugs and documentation problems go through Issues, and security issues follow SECURITY.md.
Discussion
No questions or reviews yet.
Related
SkillsOfficial AMD skills teach an agent to set up ROCm, serve LLMs on Instinct and EPYC, profile GPUs and quantize models.
A GPU profiler for AI agents: the agent runs vLLM or SGLang under graphsignal-run, reads the metrics and tunes flags for speed
A Nix and Home Manager module: skill sources are pinned in a flake and chosen SKILL.md folders are laid out into agent directories
Official Dynatrace skills for agents: DQL, traces, logs, Kubernetes, cloud and security work inside Claude Code and Cursor.
A large base of DevOps, security, infrastructure and compliance skills for AI agents
Official Datadog skills: monitors, logs, APM, cloud integrations, RUM and LLM observability from inside the agent
README in short
The README presents a catalog of 13 NVIDIA skills for working with a live Jetson, a purpose table and an installer with target selection. Skills live in skills/, each has a SKILL.md and sometimes scripts/ and references/. It links to a walkthrough, the NVIDIA Agent Skills catalog and the BSP set. Dual licence: Apache-2.0 for code, CC-BY-4.0 for docs.
README badge
Original SKILL.md
--- name: jetson-diagnostic description: Read-only Jetson health snapshot for identity, memory, GPU, thermal, power, storage, services, and top processes. version: 0.0.1 license: "Apache-2.0" metadata: author: "Jetson Team" tags: [jetson, diagnostic, telemetry] languages: [bash] data-classification: public --- # Jetson Diagnostic A unified, agent-friendly view of a running Jetson device. Replaces the need to remember which of `tegrastats`, `jtop`, `procrank`, `/sys/kernel/debug/nvmap`, `nvpmodel`, `free`, `swapon`, `df`, and `systemctl list-units` produces which slice of the truth. ## Purpose Capture a read-only health snapshot from the Jetson host so agents can answer device identity, memory, GPU, thermal, power, storage, and service-state questions using live data instead of guesses. ## When to use Activate when the user asks: - "What is this Jetson? What SKU? How much memory?" - "What's running on this Jetson right now?" - "Why is my Jetson slow / hot / out of memory?" - "Give me a snapshot of GPU / CPU / power usage." - "What does my tegrastats output mean?" - "Which services are running that I could turn off?" - The user has installed `jetson-memory-audit`, `jetson-headless-mode`, `jetson-inference-mem-tune`, `jetson-llm-benchmark`, `jetson-llm-serve`, or `jetson-package` and needs a baseline measurement before running them. Do not use this skill to change power modes, drop caches, stop services, install packages, serve models, or tune inference flags. Report the observed state, then hand off to the action-oriented skill. ## Prerequisites - Run on the Jetson host, or in a sandbox/container with host-visible Jetson system paths and process data. ## Available Scripts | Script | Purpose | Arguments | |--------|---------|-----------| | `scripts/snapshot.sh` | Emits the all-in-one JSON snapshot for identity, memory, GPU, thermal, power, disk, top processes, and candidate services. | `--human`, `--tegra-secs N`, `--top-procs N`. | | `scripts/mem_summary.sh` | Emits a compact human-readable RAM/GPU/swap summary. | `--short`, `--watch`, `--interval N`. | | `scripts/detect_jetson.sh` | Exports or prints canonical Jetson SKU/generation/product-line fields for this repo. | No arguments. |