Official agent skill

Jetson Validate Image

by NVIDIA in NVIDIA/skills

Use after jetson-flash-image to run static BSP checks, on-target smoke/regression tests on a flashed DUT, or both.

OfficialApache-2.0Auto-check: notesAI & LLM Engineering

Install Jetson Validate Image

skills CLI
$ npx skills add NVIDIA/skills --skill jetson-validate-image -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install NVIDIA/skills jetson-validate-image --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/NVIDIA/skills.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/jetson-validate-image .claude/skills/jetson-validate-image && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
jetson-validate-image
GitHub stars
3.5k
Token cost
~3.5k tokens
SKILL.md length
1,457 words
Files
7 (incl. scripts, references)
Skills in repo
380
Repo updated
First seen
Licence
Apache-2.0

At a glance

Use after jetson-flash-image to run static BSP checks, on-target smoke/regression tests on a flashed DUT, or both.

  • Works in 5 steps: Read the active target per the… → Choose validation scope — static… → Static checks (if in scope) → …
  • Tasks that involve GPU and accelerator computing
  • SKILL.md covers Purpose, Prerequisites, When to invoke and Instructions, plus 6 more sections
  • Runs Python scripts from its folder; calls apt and pip

What it does

Jetson Validate Image is an agent skill from NVIDIA/skills, published by the product's own GitHub organization. Use after jetson-flash-image to run static BSP checks, on-target smoke/regression tests on a flashed DUT, or both. Not for build or flash steps. Triggers: validate bsp, on-target validation.

Its SKILL.md is about 3.5k tokens, which your agent loads only when the skill is triggered. The skill folder holds 9 other files, including scripts and reference files (for example `BENCHMARK.md`, `evals/evals.json` and `references/dut-access.md`).

It sits in AI & LLM Engineering, covering GPU and accelerator computing. It works with NVIDIA AI Platform. The repository describes itself as: Agent Skills for NVIDIA products — install into Claude Code, Codex, and other coding agents to run Physical AI, robotics, simulation, CUDA, and RAG workflows end to end. The licence is Apache-2.0.

When your agent uses it

  • Tasks that involve GPU and accelerator computing

Example prompts

  • “/jetson-validate-image”

Requirements

  • Python 3

Workflow steps

5 steps, taken from the first numbered list in SKILL.md.

  1. Read the active target per the prerequisite contract.
  2. Choose validation scope — static (against bsp_image on
  3. Static checks (if in scope)
  4. On-target checks (if in scope)
  5. Summary: per-check pass/fail, overall verdict, where

What it can do on your machine

Read from SKILL.md and the folder at commit 0e0d506. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Ships 1 file in scripts/ (Python), which the agent can run.

    Shell commands in SKILL.md call:

    • apt
    • pip

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md. Its commands use pip, which can reach the network depending on how they are called.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Jetson Validate Image loads about 3.5k tokens when it runs, and up to ~7.3k if it reads all its reference files. Until then it costs about 53 tokens; SKILL.md has 1,457 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~53
When it runs · the whole SKILL.md, loaded when a task matches
~3.5k
With references · SKILL.md plus every file in references/, read only if the agent opens them
~7.3k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check: notes

The automated check noted patterns worth knowing about, such as sudo or a known installer.

  • NoteRuns commands with sudoSKILL.md:101
    and sudo method from the active profile's `dut_access:`
  • NoteRuns commands with sudoSKILL.md:139
    lower the restriction via `sudo sysctl
  • NoteRuns commands with sudoSKILL.md:187
    # run_script: exec a sudo command and capture exit code
  • NoteRuns commands with sudoSKILL.md:254
    om, getty) | Close the holding process (`sudo fuser -k <tty>` only if you know what's there), or rerun with `lock_strate
  • NoteRuns commands with sudoSKILL.md:259
    | `sudo prompt detected but no sudo password configured` (exit 130) | `--use-sudo` passed to `uart_session.py exec` but

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); the scripts in this folder are not scanned.

SKILL.md

The full file from NVIDIA/skills at commit 0e0d506, republished under its Apache-2.0 licence (© NVIDIA). 1,457 words, ~3,469 tokens.

Download SKILL.mdSave it as .claude/skills/jetson-validate-image/SKILL.md (or your agent's skills folder). This skill also uses 6 other files; get the full folder from GitHub.
name
jetson-validate-image
description
Use after jetson-flash-image to run static BSP checks, on-target smoke/regression tests on a flashed DUT, or both. Not for build or flash steps. Triggers: validate bsp, on-target validation.
version
0.0.1
license
Apache-2.0
metadata.data-classification
public
metadata.author
Jetson Team
metadata.tags
bsp, validation, test, deploy
metadata.domain
meta

Validate BSP Image

Status: the DUT-access contract is stable; the rest of the validation procedure is a skeleton.

Purpose

Confirm that a freshly customized BSP landed correctly — both as a static artifact on disk and as a running system on the target — without re-promoting or re-flashing. Forms the validation tail of Deploy in the Setup → Customize → Build → Deploy pipeline (see ../../context/bsp-customization-workflow.md for the pipeline view) and is independently re-runnable.

Prerequisites

  • Active target-platform profile with bsp_image: resolved (run /jetson-init-image first).
  • For static-only scope: nothing further; the skill reads <bsp_image.root_path> directly.
  • For on-target scope:
    • /jetson-flash-image already pushed the staged BSP to the DUT.
    • dut_access: block authored in the active profile (or filled interactively at runtime — see ## DUT access).
    • Host tooling per transport: ssh + sshpass for SSH; Python 3.6+ with pyserial for UART.
    • Required env vars resolvable on the host when auth=password or sudo.method=password (password_env / sudo.password_env name the env var; never inline the secret in YAML).

When to invoke

  • After jetson-flash-image has put the BSP onto a target.
  • The user explicitly asks to validate, test, or run smoke / regression checks on a flashed device.
  • As a CI gate before declaring a customization batch shippable.

Instructions

The procedure below is a skeleton.

  1. Read the active target per the prerequisite contract.
  2. Choose validation scope — static (against bsp_image on disk, no DUT needed) and/or on-target (DUT must be booted from the just-flashed image).
  3. Static checks (if in scope):
    • Required artifacts present at expected paths in <bsp_image.root_path>/Linux_for_Tegra/.
    • DTB / module checksum or signature verification.
    • Partition-layout sanity vs. the per-board .conf's XML.
    • Cross-check overlay-staged outputs against bsp_image (they should be identical post-promotion).
    • Initramfs ↔ kernel + rootfs module coherence. Extract <bsp_image>/Linux_for_Tegra/bootloader/l4t_initrd.img and <bsp_image>/Linux_for_Tegra/rootfs/boot/initrd; verify three invariants against the promoted bsp_image state: (a) <bsp_image>/Linux_for_Tegra/kernel/Image matches <bsp_image>/Linux_for_Tegra/rootfs/boot/Image byte-for- byte — drift here means the "Mirror kernel Image into rootfs" step was skipped and any subsequent initramfs refresh built against the previous kernel. (b) for each module path the two initrd images ship, the bytes / md5 match the file under <bsp_image>/Linux_for_Tegra/rootfs/lib/modules/<ver>/ — any drift means modules will be shadowed at early boot. (c) the vermagic stamped on every initramfs *.ko matches the UTS_RELEASE reachable from <bsp_image>/Linux_for_Tegra/kernel/Image (e.g. parse the Linux version … string with strings) — a vermagic skew means the kernel Image was refreshed without rerunning l4t_update_initrd.sh, and modules will fail to load with "disagrees about version of symbol …". All three failure modes are closed by /jetson-promote-image's gate on either kernel/Image or rootfs/lib/modules/ plus the kernel-Image mirror; surface drift here and route the user back to a clean promote. See ../jetson-promote-image/SKILL.md.
  4. On-target checks (if in scope):
    • Connect to the DUT per the ## DUT access section below — resolve transport (ssh / uart), credentials, and sudo method from the active profile's dut_access: block (with interactive fallback when fields are missing / marked prompt), then run the connection probe and refuse if it fails.
    • Confirm boot reached userspace.
    • Run the selected test suite (smoke, regression, focused per-customization, ad-hoc).
    • Loaded-module srcversion drift. For modules a customization is known to have rebuilt, compare cat /sys/module/<name>/srcversion on the DUT against the modinfo /lib/modules/$(uname -r)/.../<name>.ko | awk '/srcversion/ {print $2}' reading. A mismatch means the kernel is running an older copy than the rootfs ships — almost always a stale initramfs (the bootloader-side initrd shipped a pre-customize module, it loaded first, and the rootfs copy cannot replace a live module). Recommend re-running /jetson-promote-image and re-flashing. Note: some modules don't emit srcversion; fall back to an md5 check against the binary the kernel loaded by extracting the region under /sys/module/<name>/sections/ or by comparing behaviorally (printk / sysfs nodes / DT properties the new version is known to expose).
    • Running kernel vs. rootfs kernel Image. Compare cat /proc/version (or uname -v) on the DUT against the Linux version … string extracted from /boot/Image (strings /boot/Image | grep -m1 'Linux version'). A mismatch — usually the build timestamp / LOCALVERSION — means the bootloader is running an older kernel Image than the rootfs holds, almost certainly because a fresh Image was promoted but the initramfs / extlinux.conf / QSPI boot partition wasn't refreshed. Modules in the rootfs will then have a different vermagic and any subsequent modprobe of a built-against-the-new-kernel .ko will fail. Recommend /jetson-promote-image and re-flash.
    • Userspace dmesg readability. Ubuntu 22.04 sets kernel.dmesg_restrict=1; non-root dmesg reads return "Operation not permitted" and silently zero hits. Every dmesg-based check must run with sudo (or temporarily lower the restriction via sudo sysctl kernel.dmesg_restrict=0). Surfacing this in the validate layer keeps printk-based customization checks honest.
    • Collect results, logs, artifacts.
  5. Summary: per-check pass/fail, overall verdict, where logs and artifacts landed.

DUT access

The on-target leg needs a way to reach the just-flashed DUT. Two transports are supported as full peers: ssh (primary) and uart (fallback for DUTs with no network).

The contract is locked in but lives in references/dut-access.md to keep this SKILL.md under the agent-routing budget. That reference covers:

  • The dut_access: profile schema (ssh, uart, sudo, workdir).
  • Resolution order (profile → env var → interactive prompt) with the full refusal-trigger table.
  • The mandatory connection probe (uname -r + cat /etc/nv_tegra_release) and its output-validation rules.
  • File transfer per transport (scp vs. base64-over-tty, with the >100 KB warning).
  • Sudo invocation matrix per sudo.method × transport.
  • UART implementation contract that scripts/uart_session.py honors (state machines for login / exec / push / pull, exit codes, robustness notes).
  • Security notes (password handling, host-key pinning).
Show full SKILL.md (593 more words)Show less

Available Scripts

ScriptPurposeArguments
scripts/uart_session.pyUART transport black box for the on-target leg: login, command exec (with optional sudo), and base64-over-tty file transfer. Replaces ssh when the DUT has no network or ssh is broken.--tty <dev> [--baud <n>] --user <name> --password-env <ENVVAR> [--sudo-password-env <ENVVAR>] [--shell-prompt <regex>] [--lock-strategy refuse|wait] <probe|exec|push|pull> [...]

Invocation (the skill calls scripts as a subprocess — run_script() in agent-runtime terminology):

bash
# run_script: probe the DUT over UART
DUT_UART_PASSWORD_ENV=DUT_UART_PWD \
DUT_UART_PWD="$(read -rs -p 'UART login pw: '; echo "$REPLY")" \
  scripts/uart_session.py \
    --tty /dev/ttyACM0 --baud 115200 \
    --user ubuntu --password-env DUT_UART_PASSWORD_ENV \
    probe

# run_script: exec a sudo command and capture exit code
scripts/uart_session.py --tty /dev/ttyACM0 --user ubuntu \
  --password-env DUT_UART_PASSWORD_ENV \
  --sudo-password-env DUT_SUDO_PWD \
  exec --use-sudo 'dmesg | tail -200'

The script's exit code is the contract — see the exit-code table in UART implementation contract.

Examples

Static-only validation (no DUT needed):

/jetson-validate-image
> static checks only against the staged BSP

On-target validation over SSH after a freshly flashed DUT:

/jetson-flash-image
   ↓
/jetson-validate-image
> on-target checks via dut_access.ssh

On-target validation over UART (no network on the DUT):

/jetson-validate-image
> use the uart transport at /dev/ttyACM0; the dut_access.uart block
  in the profile already has the tty and login_password_env wired up

Limitations

  • Placeholder skill — only the DUT-access contract and the uart_session.py helper are locked in. The static-check list, test-suite selection, result-sink layout, and pass/fail policy are tracked under ## Open items and may change.
  • UART file transfer is byte-banged base64 at ~10 KB/s on 115200 baud — emits a warning for sources > 100 KB but proceeds. For high-volume transfers, switch to the SSH transport.
  • uart_session.py opens and closes the tty per subcommand invocation (~1–2 s login per call). Validation passes running

    10 commands amortize poorly on UART; prefer SSH.

  • SSH uses StrictHostKeyChecking=accept-new with a per-profile known_hosts file. A fingerprint change refuses — typically means the DUT was reflashed (host keys regenerated) or the IP was reassigned. Remove the per-profile entry manually rather than auto-accepting.
  • Passwords are never inlined in the profile YAML — only password_env (env-var name) is persisted. auth=prompt / sudo.method=prompt exposes the password in the conversation log, which is the user's responsibility to manage.
  • transport=uart with lock_strategy=steal is not implemented (would require sending control characters that could corrupt the holder's state).

Troubleshooting

ErrorCauseSolution
no DUT transport configured (refuse)On-target scope requested but dut_access.transport unresolved across profile / env / interactive promptAuthor dut_access: in the active profile, or rerun with static-only scope.
$PASSWORD_ENV is unset (refuse)auth=password (or sudo.method=password) names an env var that isn't exported on the hostexport <ENVVAR>=... before invoking the skill; do NOT inline the password in YAML.
tty held by another process (refuse, exit 3)lock_strategy=refuse and fuser reports another holder (minicom, picocom, getty)Close the holding process (sudo fuser -k <tty> only if you know what's there), or rerun with lock_strategy=wait.
pyserial import failed (exit 4)UART transport selected but pyserial not installed in the skill's Pythonapt install python3-serial or pip install pyserial.
SSH fingerprint refusedDUT's host keys changed since the per-profile known_hosts was pinned (typically a reflash)Remove the matching line from <workspace>/target-platform/<profile-stem>.known_hosts and rerun — the new key will be accepted on first connect.
login failed (exit 128)UART probe couldn't match a login or shell prompt within timeoutConfirm the DUT is powered + booted to userspace; check --shell-prompt override if the DUT carries an unusual PS1; verify the --user matches a real account.
DUT not booted from the just-flashed BSP (warn / refuse)/etc/nv_tegra_release on the DUT doesn't match bsp_image.versionRe-run /jetson-flash-image and confirm the DUT actually power-cycled into the new BSP, not the previous one.
sudo prompt detected but no sudo password configured (exit 130)--use-sudo passed to uart_session.py exec but --sudo-password-env not set, and DUT user lacks NOPASSWDSet sudo.method=password + sudo.password_env in the profile, or grant NOPASSWD on the DUT via /etc/sudoers.d/.
command timed out (exit 129)Long-running DUT command exceeded the script's timeoutRun the command directly via ssh transport, or split into shorter steps.

References

© NVIDIA, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 6 other files (scripts, references) in skills/jetson-validate-image of NVIDIA/skills.

  • SKILL.md
  • BENCHMARK.md
  • evals/evals.json
  • references/dut-access.md
  • scripts/uart_session.py
  • skill-card.md
  • skill.oms.sig

Open the folder on GitHubat commit 0e0d506

Compare with similar skills

Jetson Validate Image next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Jetson Validate Image compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Jetson Validate Image this skillNVIDIA/skills3.5k—~3.5kAutomated safety check: NotesApache-2.0
Fla Triton To Gluonfla-org/flash-linear-attention5.8k—~4.2kAutomated safety check: PassMIT
DGX Spark Memory and Thermal Opswshobson/agents40k1 repos~2kAutomated safety check: PassMIT
DGX Spark Training Gotchaswshobson/agents40k1 repos~2kAutomated safety check: PassMIT
Megatron-LM on SLURMNVIDIA/Megatron-LM18k—~1.8kAutomated safety check: PassApache-2.0
OpenVLA-OFT Fine-TuningOrchestra-Research/AI-Research-SKILLs13k1 repos~3.7kAutomated safety check: PassMIT

Similar skills

  • Fla Triton To Gluon

    fla-org/flash-linear-attention

    Workflow for porting an existing Triton kernel in fla/ops/ to Gluon (triton.experimental.gluon) to gain explicit control over tensor layouts, shared memory, async data movement (cp.async / TMA), MMA…

    5.8k GitHub stars~4.2k tokensUpdated yesterday
    AI & LLM EngineeringAuto-check passed
  • Plans memory headroom, works through out-of-memory failures and watches temperature and power during long ML training jobs on NVIDIA DGX Spark.

    40k GitHub starsUsed in 1 repo~2k tokens
    AI & LLM EngineeringAuto-check passed
  • Preflight checks and diagnosis for ten known failure modes of ML training on NVIDIA DGX Spark's GB10, spanning launch errors, memory, thermals, bandwidth and precision.

    40k GitHub starsUsed in 1 repo~2k tokens
    AI & LLM EngineeringAuto-check passed
  • Megatron-LM on SLURM

    NVIDIA/Megatron-LM

    Official

    Shows how to launch distributed Megatron-LM training on a SLURM cluster: sbatch skeleton, torch.distributed.run setup, CUDA_DEVICE_MAX_CONNECTIONS rules and failure diagnosis.

    18k GitHub stars~1.8k tokensUpdated today
    AI & LLM EngineeringAuto-check passed
  • OpenVLA-OFT Fine-Tuning

    Orchestra-Research/AI-Research-SKILLs

    Fine-tunes and evaluates OpenVLA-OFT and OFT+ robot policies with LoRA and continuous action heads on LIBERO simulation and ALOHA real-robot setups.

    13k GitHub starsUsed in 1 repo~3.7k tokens
    AI & LLM EngineeringAuto-check passed
  • Cosmos Policy Evaluation

    Orchestra-Research/AI-Research-SKILLs

    Sets up and runs NVIDIA Cosmos Policy evaluations on the LIBERO and RoboCasa simulators, including headless GPU rendering and inference latency profiling.

    13k GitHub stars~3.7k tokensUpdated 3 mo ago
    AI & LLM EngineeringAuto-check passed

More from NVIDIA/skills

All 380 skills in this repo
  • Official

    A skill your agent uses when the user wants to deploy, run, debug, tear down, or call the REST API of the RTVI-CV 2D detection / tracking microservice.

    3.5k GitHub starsUsed in 1 repo~4.5k tokens
    Auto-check passed
  • Official

    Generates, validates, compares and explains HOLOLINK_def.svh macro files for the HSB IP, using bundled Python scripts and asking before it writes anything.

    3.5k GitHub stars~2.9k tokensUpdated yesterday
    Auto-check passed
  • Official

    Runs and validates an end-to-end Mission Control demo in a locally installed Isaac Sim, with a Nova Carter robot driven through a Python server.

    3.5k GitHub stars~4.8k tokensUpdated yesterday
    Auto-check passed
  • Orchestrates defect image generation for PCBA, metal surface and glass inspection with NVIDIA Cosmos AnomalyGen on OSMO, from cold-start Day 0 to real-photo Day 1 labeling.

    3.5k GitHub stars~5k tokensUpdated yesterday
    Auto-check: notes
  • Orchestrates video data augmentation and auto-labeling workflows on OSMO, from flow selection and preflight checks to submission, monitoring and output download.

    3.5k GitHub stars~4.7k tokensUpdated yesterday
    Auto-check: notes
  • Official

    Runs NVIDIA TAO Data Services KPI analysis on object detection results, comparing predictions to ground truth and writing per-class precision, recall and AP to a CSV.

    3.5k GitHub stars~2.7k tokensUpdated yesterday
    Auto-check: notes

Questions about Jetson Validate Image

What does Jetson Validate Image do?

Use after jetson-flash-image to run static BSP checks, on-target smoke/regression tests on a flashed DUT, or both. Jetson Validate Image is an agent skill from NVIDIA/skills, published by the product's own GitHub organization. Use after jetson-flash-image to run static BSP checks, on-target smoke/regression tests on a flashed DUT, or both.

When should I use Jetson Validate Image?

Jetson Validate Image fits situations like: tasks that involve GPU and accelerator computing.

How do I install Jetson Validate Image in Claude Code?

Run `npx skills add NVIDIA/skills --skill jetson-validate-image -a claude-code`. Or copy the skill folder (skills/jetson-validate-image in NVIDIA/skills) into .claude/skills/jetson-validate-image in your project. Claude Code loads it when a task matches its description.

How do I install Jetson Validate Image in Codex?

Run `npx skills add NVIDIA/skills --skill jetson-validate-image -a codex`. Or copy the skill folder (skills/jetson-validate-image in NVIDIA/skills) into .agents/skills/jetson-validate-image in your project. Codex loads it when a task matches its description.

Can I use Jetson Validate Image in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add NVIDIA/skills --skill jetson-validate-image -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/jetson-validate-image, .gemini/skills/jetson-validate-image, .github/skills/jetson-validate-image and .opencode/skills/jetson-validate-image in your project.

What does Jetson Validate Image need to run?

Going by SKILL.md and its folder, Jetson Validate Image needs Python for the scripts in its folder and the command-line tools its instructions call (apt and pip). Our summary lists: Python 3.

Does Jetson Validate Image access the network?

SKILL.md contains no URLs. Its commands use pip, which can reach the network depending on how they are called. This is read from the text; nothing was executed.

Is Jetson Validate Image safe to install?

Our automated static check of SKILL.md found notes only (runs commands with sudo), nothing it rates as a warning. It is not a guarantee. The check reads SKILL.md only: the scripts in the folder are not scanned, so read them before running anything.

What licence does Jetson Validate Image use?

Jetson Validate Image is published under the Apache-2.0 licence (declared in SKILL.md). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Jetson Validate Image use?

About 3.5k tokens (SKILL.md is roughly 14k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 3.8k tokens, read only when the agent opens those files.

What are the alternatives to Jetson Validate Image?

Skills that share tags, products or a category with Jetson Validate Image: Fla Triton To Gluon (fla-org/flash-linear-attention, 5.8k stars), DGX Spark Memory and Thermal Ops (wshobson/agents, 40k stars), DGX Spark Training Gotchas (wshobson/agents, 40k stars) and Megatron-LM on SLURM (NVIDIA/Megatron-LM, 18k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Jetson Validate Image?

NVIDIA (a GitHub organization, an official publisher) maintains it in NVIDIA/skills, which has 3,534 GitHub stars. The repository holds 380 skills in this directory. The repository was last updated on October 7, 2026.

Source: NVIDIA/skills on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.