Inspect
Read the actual module, release, memory, thermals, and services.
← Jetson AI Lab Give your coding agent
NVIDIA Jetson Agent Skills
Give your coding agent
Jetson instincts.
Build, optimize, and deploy NVIDIA Jetson applications with AI coding agents guided by Jetson Device Skills and Jetson BSP Skills.
From general-purpose agent to Jetson collaborator.
More than a decade of Jetson engineering knowledge now helps coding agents support development on Jetson. Agent Skills organize that foundation into reusable workflows and combine it with live context such as JetPack, memory, power mode, and the selected BSP target.
Decide
Match runtimes, packages, and settings to the device in front of you.
Execute
Use repeatable helpers for serving, benchmarking, video, and BSP work.
Verify
Measure the result instead of assuming that a change worked.
Choose the Skills You Need
Most developers should start with Jetson Device Skills. Add Jetson BSP Skills only if you customize or flash the Board Support Package.
A / LIVE DEVICE Application workflows
Recommended for most developers
Jetson Device Skills
Run on a booted Jetson. Diagnose the system, recover memory, select packages, serve and benchmark models, and build accelerated video pipelines.
- RUNS ON
- Your Jetson
- BEST FOR
- Application development
- AGENTS
- Cursor · Codex · Claude · OpenClaw
B / PRE-FLASH Platform workflows
For custom hardware and production bring-up
Jetson BSP Skills
Run on a host workstation against a Linux_for_Tegra tree. Configure pinmux, PCIe, UPHY, cameras, power, memory, builds, flashing, and validation.
- RUNS ON
- Host workstation
- BEST FOR
- Platform customization
- AGENTS
- Cursor · Codex · Claude
See what each skill adds.
jetson-diagnostic
DEVICERead-only health snapshot for identity, memory, GPU, thermals, power, storage, and services.
jetson-print-device-info
DEVICEReport module model, L4T, kernel, operating system, and current power mode.
jetson-memory-audit
DEVICEMeasure DRAM and NvMap use, then verify whether a change reclaimed memory.
jetson-headless-mode
DEVICEPlan and apply bounded headless changes that recover GUI and daemon memory.
jetson-inference-mem-tune
DEVICESelect a serving stack and tune memory flags for the live Jetson configuration.
jetson-llm-serve
DEVICELaunch vLLM or SGLang with compatible images and device-aware settings.
jetson-llm-benchmark
DEVICEBenchmark LLM serving with repeatable metrics and structured JSON output.
jetson-speculative-decoding
DEVICEAdd EAGLE-3 or a draft model when output-token latency is the bottleneck.
jetson-package
DEVICEChoose Jetson-compatible containers, runtime images, and Python package sources.
jetson-video-setup
DEVICEInstall, repair, probe, and verify Video Codec SDK or PyNvVideoCodec.
jetson-video-capability
DEVICEResolve codec, profile, chroma, bit-depth, dimensions, and engine support.
jetson-video-recipe
DEVICETurn an encoder use case into one validated configuration and command.
jetson-video-benchmark
DEVICEMeasure encode/decode throughput and compare presets or worker capacity.
jetson-video-pipeline
DEVICEExecute and verify encode, decode, transcode, segmentation, and AV1 workflows.
jetson-quick-start
BSPCollect core target inputs and dispatch the right BSP setup workflow.
jetson-init-target
BSPCreate a target-platform profile and make it the active BSP target.
jetson-download-bsp
BSPDownload release-matched BSP, rootfs, sources, toolchain, and guides.
jetson-derive-carrier
BSPFork reference carrier files and scaffold a custom carrier overlay.
jetson-customize-pinmux
BSPApply per-pin SFIO, direction, and initial state from pinmux data.
jetson-customize-usb
BSPEnable or disable USB2 and USB3 ports through a device-tree overlay.
jetson-customize-pcie
BSPConfigure PCIe controllers, lanes, and link speed for the target.
jetson-customize-uphy
BSPAllocate UPHY lanes across PCIe, USB3, and MGBE.
jetson-customize-camera
BSPEnable MIPI or GMSL camera sensors from in-tree sensor definitions.
jetson-customize-nvpmodel
BSPAdd, edit, and select the boot-default nvpmodel power mode.
jetson-build-source
BSPRebuild device tree, modules, kernel, or the full source workspace.
jetson-flash-image
BSPFlash a promoted BSP image with the appropriate NVIDIA workflow.
jetson-validate-image
BSPRun static BSP checks and on-target validation after flashing.
Need the source? Device Skills ↗ · BSP Skills ↗
Install from the NVIDIA Skill Catalog.
Discover and install selected Jetson skills from the official NVIDIA Agent Skills catalog. The source repositories remain available when you want the complete collection.
Browse the NVIDIA Skills catalog on GitHub ↗Open the NVIDIA catalog
npx skills@latest add nvidia/skillsBrowse without installing
npx skills@latest add nvidia/skills --listInstall one skill for Cursor
npx skills@latest add nvidia/skills --skill jetson-diagnostic --agent cursor --yesReason broadly. Act with device context.
Skills do not replace the agent or NVIDIA documentation. They connect reasoning to verified Jetson procedures and live measurements.
- 01
YOU
Describe the outcome
Use natural language. Set boundaries and approve material changes.
- 02
CODING AGENT
Plans and orchestrates
Cursor, Codex, or Claude inspects the project and selects relevant skills.
- 03
JETSON SKILL
Applies platform procedure
Instructions and helpers gather live facts and execute bounded workflows.
- 04
MEASURED RESULT
Returns evidence
Review output from the target hardware before deciding what comes next.
Install once. Use from every project.
Run this on your Jetson. The installer links the complete Device Skills catalog into the location your selected coding agent reads.
- 01 Clone the skills
git clone https://github.com/NVIDIA-AI-IOT/jetson-device-skills.git - 02 Open the directory
cd jetson-device-skills
03
Choose a target
./install.sh --targets cursor./install.sh --targets codex./install.sh --targets claude./install.sh --targets cursor-project --project /path/to/project./install.sh --targets nemoclaw --nemoclaw-sandbox jetson-skills NEXT STEPS