Official agent skill

Jetson Flash Image

by NVIDIA in NVIDIA/skills

A skill your agent uses to flash a promoted BSP image to a Jetson DUT in RCM mode via flash.sh or l4tinitrdflash.sh.

OfficialApache-2.0Auto-check: notesAI & LLM Engineering

Install Jetson Flash Image

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

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

GitHub CLI
$ gh skill install NVIDIA/skills jetson-flash-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-flash-image .claude/skills/jetson-flash-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-flash-image
GitHub stars
3.5k
Token cost
~4.9k tokens
SKILL.md length
2,157 words
Files
8 (incl. references)
Skills in repo
380
Repo updated
First seen
Licence
Apache-2.0

At a glance

A skill your agent uses to flash a promoted BSP image to a Jetson DUT in RCM mode via flash.sh or l4tinitrdflash.sh.

  • Works in 2 steps: If the active profile records the boot… → Otherwise, prompt the user with the…
  • Flash a promoted BSP image to a Jetson DUT in RCM mode via flash.sh
  • SKILL.md covers Purpose, Prerequisites, When to invoke and Instructions, plus 3 more sections
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md

What it does

Jetson Flash Image is an agent skill from NVIDIA/skills, published by the product's own GitHub organization. Use to flash a promoted BSP image to a Jetson DUT in RCM mode via flash.sh or l4tinitrdflash.sh. Do NOT use for BSP customization, image promotion, or carrier derivation.

Its SKILL.md is about 4.9k tokens, which your agent loads only when the skill is triggered. The skill folder holds 9 other files, including reference files (for example `BENCHMARK.md`, `evals/evals.json` and `references/default-user-staging.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

  • Flash a promoted BSP image to a Jetson DUT in RCM mode via flash.sh
  • L4tinitrdflash.sh
  • BSP customization
  • Image promotion

Example prompts

  • “/jetson-flash-image”

Workflow steps

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

  1. If the active profile records the boot device, use it.
  2. Otherwise, prompt the user with the choices the per-board conf

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

    No scripts in the folder and no shell commands in SKILL.md (its code samples are bash).

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

  • Network

    No URLs in SKILL.md.

    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 Flash Image loads about 4.9k tokens when it runs, and up to ~7k if it reads all its reference files. Until then it costs about 48 tokens; SKILL.md has 2,157 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~48
When it runs · the whole SKILL.md, loaded when a task matches
~4.9k
With references · SKILL.md plus every file in references/, read only if the agent opens them
~7k

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:178
    using `sudo ./nvautoflash.sh --print_boardid` from
  • NoteRuns commands with sudoSKILL.md:251
    sudo ./<flow-tool> [<resolved flags>] <board> <boot-dev>

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); files beside SKILL.md are not scanned.

SKILL.md

The full file from NVIDIA/skills at commit 0e0d506, republished under its Apache-2.0 licence (© NVIDIA). 2,157 words, ~4,920 tokens.

Download SKILL.mdSave it as .claude/skills/jetson-flash-image/SKILL.md (or your agent's skills folder). This skill also uses 7 other files; get the full folder from GitHub.
name
jetson-flash-image
description
Use to flash a promoted BSP image to a Jetson DUT in RCM mode via flash.sh or l4t_initrd_flash.sh. Do NOT use for BSP customization, image promotion, or carrier derivation.
version
0.0.1
license
Apache-2.0
metadata.data-classification
public
metadata.author
Jetson Team
metadata.tags
bsp, flash
metadata.domain
meta

Flash BSP Image

Purpose

Push a promoted bsp_image to a Jetson DUT by running the NVIDIA flashing toolchain (flash.sh or l4t_initrd_flash.sh) from the in-tree Linux_for_Tegra/. The DUT must be in RCM (recovery) mode at flash time. This is the flash leg of the BSP overlay deploy chain (/jetson-promote-image → /jetson-flash-image); see the BSP overlay workflow at ../../context/bsp-customization-workflow.md.

Design principle. Four invariants govern this skill; each is fully explained in the Instructions step that owns it.

  • Host-side flash variables (<board>, <boot-dev>, flow tool, per-board .conf, boardctl) are resolved from the active profile or the in-tree BSP.
  • Artifact paths (DTB_FILE, BPFDTB_FILE, partition XML, BCTs, DRAM training) are resolved inside flash.sh / l4t_initrd_flash.sh at flash time from board_sku / board_FAB read off the DUT's EEPROM.
  • The DUT's EEPROM is authoritative; the profile is an authoring-time prediction reconciled by the preflight cross-check. Empty EEPROM values are valid, not refusal triggers.
  • The user explicitly confirms a printed resolution before flashing.

Out of scope: BSP customization (use the /jetson-customize-* skills), promoting the overlay tracker into bsp_image (use /jetson-promote-image), and producing a custom carrier's flash conf (use /jetson-derive-carrier).

Prerequisites

  • Active target-platform profile resolved per ../../context/target-platform-contract.md with a populated bsp_image: block. Refuse and route to /jetson-init-image if missing.
  • <bsp_image.root_path>/Linux_for_Tegra/ exists on the host and has been through apply_binaries.sh (route to /jetson-init-image otherwise).
  • A per-board flash .conf resolvable via the active-block precedence rule (custom_carrier.flash_config → reference_devkit.flash_config). Refuse and route to /jetson-derive-carrier or /jetson-init-image if absent.
  • For the prior overlay → image leg: /jetson-promote-image has already run (or a standalone re-flash with no promotion changes since the last flash).
  • A DUT cabled to the host that can be driven into RCM mode either via the in-tree boardctl or by manual recovery + reset buttons.

When to invoke

  • After /jetson-promote-image has updated bsp_image to carry the desired customizations.
  • Standalone re-flash with no promotion changes since the last flash.
  • DUT must be in RCM mode before invocation.

Instructions

Resolve target and bsp_image

Resolve the active profile per ../../context/target-platform-contract.md. Refuse if bsp_image: is missing or <bsp_image.root_path>/Linux_for_Tegra/ does not exist (route the user to /jetson-init-image).

Resolve the flash conf path

Pick the per-board .conf using the active-block precedence rule from target-platform-contract.md: custom_carrier.flash_config when present, else reference_devkit.flash_config. Verify the chosen file exists under <bsp_image.root_path>/Linux_for_Tegra/. Refuse if absent — route the user at /jetson-derive-carrier (custom carrier) or /jetson-promote-image / /jetson-init-image (reference devkit).

Bind <board> as the basename of the resolved .conf minus the .conf suffix (e.g. jetson-agx-thor-devkit.conf → <board>=jetson-agx-thor-devkit). the "Invoke the flash" step's command shape consumes this binding directly.

No dispatch preview at this stage. Artifact resolution (DTB_FILE, BPFDTB_FILE, partition XML, MB1 BCTs, DRAM training tables) happens inside flash.sh / l4t_initrd_flash.sh during image generation, using board_sku and board_FAB read from the DUT's EEPROM in recovery mode — not values typed in from the profile. The profile is an authoring-time prediction; the DUT EEPROM is authoritative at flash time. The "Preflight checks" step's EEPROM cross-check reads EEPROM and refuses the flash on profile / EEPROM mismatch, so passing preflight is what guarantees the flash-time dispatch will pick the artifacts the profile expects.

For static analyses outside the flash flow that need predicted artifact paths (KB generation, customize-* skills locating files to edit), see the standalone snippet in per-board conf dispatch — that snippet is valid for BSP-side resolution but not as a flash-time preview.

Select <boot-dev> and flash flow

<boot-dev> is resolved deliberately, never assumed. Source order:

  1. If the active profile records the boot device, use it.
  2. Otherwise, prompt the user with the choices the per-board conf actually supports (internal, external, nvme0n1p1, mmcblk0p1, etc., depending on chip family and conf variant).

Pick the flow tool from the matrix below:

Chip familyDefault boot mediaTool
T234 / OrineMMC / SDflash.sh
T234 / OrinNVMe / USBl4t_initrd_flash.sh
T264 / ThorNVMe / UFSl4t_initrd_flash.sh

Massflash and --flash-only re-runs use the same tool selection but add their respective flags.

Put the DUT into recovery mode

Drive the device into recovery via the in-tree boardctl (preferred) or by manually pressing the recovery button followed by the reset button.

The full procedure — where to find boardctl under Linux_for_Tegra/, how to enumerate -t targets and pick one (recommend topo), the exact recovery verb to use, and the manual fallback — lives in references/recovery-mode-boardctl.md.

Bind the resolved binary path as <boardctl> for the remainder of this skill. Never substitute a $PATH boardctl, never invent a target name not present in <boardctl> -h. The "Preflight checks" step's DUT-recovery verification covers — for both paths — that the DUT actually landed in RCM mode.

Preflight checks

Performing the following preflight checks in the specified order, and never skip each check. Host-side checks run first (fail early before asking the user to flip recovery), then DUT-side checks once the user has put the device in RCM mode. EEPROM-dependent checks come after RCM detection, since the recovery channel is what makes the read possible.

5.1 bsp_image readiness (host-side).

  • Per-board conf exists at <bsp_image.root_path>/Linux_for_Tegra/<flash_config>.
  • Artifact subtrees populated: kernel/dtb/, bootloader/, bootloader/generic/BCT/, rootfs/.
  • apply_binaries.sh has been run — check for rootfs/etc/nv_tegra_release and the chip's nvidia-l4t-bsp-* marker package files.
  • The exact resolved DTB_FILE / BPFDTB_FILE / BCT filenames are intentionally not checked here — those are picked from EEPROM at flash time (see the "Resolve the flash conf path" step).
  • Kernel Image mirror invariant. When both <LFT_DST>/kernel/Image and <LFT_DST>/rootfs/boot/Image exist, they must be byte-identical (cmp -s); drift means /jetson-promote-image's Mirror step was skipped — route back. Initramfs presence. l4t_initrd_flash.sh requires <LFT_DST>/bootloader/l4t_initrd.img and <LFT_DST>/rootfs/boot/initrd; absent → route to /jetson-init-image. (Freshness vs. rootfs/lib/modules/<ver>/ is not checked here — owned by /jetson-promote-image's Refresh initramfs step.)

5.2 DUT in RCM mode.

  • Verifies the outcome of the "Put the DUT into recovery mode" step (either the user-selected <boardctl> -t <target> invocation or the manual fallback).

  • lsusb -d 0955: must report at least one device matching the active chip family's recovery VID:PID pair:

    Chip familyRecovery VID:PID
    T23x / Orin0955:7X23 (X is any hex digit — module variant)
    T26x / Thor0955:7026

    For T23x, match on the trailing 23 (e.g. lsusb -d 0955: | grep -E ' 0955:7.23 '); a literal 0955:7023 check will miss valid T23x variants.

  • Absent → if the "Put the DUT into recovery mode" step used boardctl, surface its output and refuse; if the "Put the DUT into recovery mode" step used the manual path, re-prompt the user to confirm the jumper / button and power-cycle.

  • This is the gate for everything downstream. flash.sh's image-generation phase reads EEPROM over the same recovery channel; a device that's not in RCM here will fail there too.

5.3 EEPROM cross-check vs. active profile.

Read board_sku and board_FAB (and any additional dispatch inputs the active chip family uses) from the DUT's EEPROM in recovery mode using sudo ./nvautoflash.sh --print_boardid from <bsp_image.root_path>/Linux_for_Tegra/. The full reference — sample output, label-to-dispatch-input mapping, empty-value semantics, and the EEPROM-vs-profile reconciliation table — lives in references/eeprom-cross-check.md.

Refusal trigger is a real non-empty disagreement, never a missing value. Empty EEPROM values are valid and not a refusal trigger. The cross-check is the primary defense against the wrong-target / wrong-SKU class of failures whenever both sides supply enough information to disagree.

5.4 Default user staging (host-side, interactive).

A freshly applied bsp_image has no Linux user pre-staged. Detect via <bsp_image.root_path>/Linux_for_Tegra/rootfs/home/ and rootfs/etc/passwd UID ≥ 1000. If none, issue one AskUserQuestion with four click-to-select options: ubuntu / ubuntu, nvidia / nvidia, custom (sub-prompt for username + password), or skip (OEM wizard on first boot). Non-skip picks run l4t_create_default_user.sh --autologin --accept-license. Full invocation + rationale in references/default-user-staging.md.

Record the resolution so the "Confirm resolution" step can display it. This step is not a refusal gate — it is a user-interaction point.

Confirm resolution

Print the resolved plan and require explicit acceptance. The format is the resolution, not the shell command — the user is approving what will flash, not the string that will be executed:

Target:         <reference_devkit.name> [+ custom_carrier.name]
Profile:        target-platform/<active>.yaml
bsp_image:      <bsp_image.root_path>/Linux_for_Tegra  (version <X>)
Flash conf:     <flash_config>   (path verified in the "Resolve the flash conf path" step)

DUT EEPROM  →   board_sku=<value-or-(empty)>  board_FAB=<value-or-(empty)>
Profile     →   module.sku=<value-or-(any)>    module.revision=<value-or-(any)>
                (reconciled per the "Preflight checks" step's EEPROM cross-check)

Boot device:    <boot-dev>
Flow tool:      flash.sh  |  l4t_initrd_flash.sh
boardctl:       <bsp_image.root_path>/Linux_for_Tegra/tools/board_automation/boardctl
                (or the path the "Put the DUT into recovery mode" step resolved)
RCM entry:      <boardctl> -t <user-selected target> recovery  |  manual recovery + reset buttons
Post-flash:     <boardctl> -t <user-selected target> reset   (T26x / Thor)
                not required — flash tool resets internally    (T23x / Orin)
Default user:   <username> (autologin)
              | already staged in rootfs (kept)
              | none — OEM config wizard on first boot

Artifact paths (DTB_FILE, BPFDTB_FILE, partition XML, BCTs) are not shown — they are resolved by flash.sh from EEPROM at flash time, not by this skill. The "Preflight checks" step's EEPROM cross-check is what guarantees those flash-time picks will line up with what the profile expects.

This is the user-acceptance gate. Refuse to fall back to a raw "paste this command" workflow — that path is how stale doc snippets, wrong dashes, and prompt-character paste artifacts make it into production flashes.

Show full SKILL.md (822 more words)Show less
Invoke the flash

Construct the command from the resolved variables — never accept a verbatim command from the user, the docs, or memory:

bash
cd <bsp_image.root_path>/Linux_for_Tegra
sudo ./<flow-tool> [<resolved flags>] <board> <boot-dev>

Abort and surface the failed step on the first non-zero exit. Do not auto-retry on transient USB errors.

Post-flash reset (T26x / Thor only)

T26x platforms do not auto-reboot from the freshly flashed image when flash.sh / l4t_initrd_flash.sh returns. Run, using the <boardctl> resolved in the "Put the DUT into recovery mode" step and the same target the user selected for RCM entry:

bash
<boardctl> -t <user-selected target> reset

T23x / Orin issues the reset internally; skip this step on Orin. If the "Put the DUT into recovery mode" step used the manual path, prompt the user to remove the force-recovery jumper / button and power-cycle by hand. Gate this step on chip family resolved from the active profile.

Summary

Report: command line(s) used (flash + post-flash reset if it ran), exit code, log location (if teed). Persist the resolved plan and outcome where validation can re-read it.

Limitations

  • DUT must be in RCM mode. Image-generation (EEPROM read) and the flash itself both use the recovery USB channel; not-in-RCM at the gate fails image-gen too.
  • Artifact paths resolved at flash time. DTB_FILE, BPFDTB_FILE, partition XML, BCTs, DRAM training are picked from EEPROM (board_sku / board_FAB) by flash.sh / l4t_initrd_flash.sh; this skill validates only host-side scaffolding.
  • EEPROM is authoritative over the profile. Real non-empty disagreement refuses; empty EEPROM values are valid.
  • No raw-command bypass. Refuses to fall back to user-supplied "paste this command" — that path is how stale doc snippets and prompt-character paste artifacts reach production flashes.
  • No transient-error auto-retry. USB hiccups abort; re-enter RCM and re-invoke.
  • T26x needs an explicit post-flash reset. T26x doesn't auto-reboot from a freshly flashed image; <boardctl> -t <target> reset (or manual power-cycle) required. T23x resets internally.
  • Massflash / --flash-only uses the same tool selection + respective flags; massflash topology setup is out of scope here.

Troubleshooting

ErrorCauseSolution
lsusb -d 0955: reports nothing matching 0955:7X23 (T23x) or 0955:7026 (T26x)DUT did not enter RCM mode — boardctl recovery failed or the manual recovery + reset sequence didn't take.If boardctl was used, surface its output and re-prompt; manual path, re-confirm the recovery jumper / button and power-cycle. Re-run the "DUT in RCM mode" preflight.
Preflight EEPROM cross-check refuses with board_sku / board_FAB mismatch vs. profileEEPROM holds a real non-empty value that disagrees with module.sku / module.revision in the active profile.Fix the profile (/jetson-set-target or /jetson-init-target) to match the DUT's actual EEPROM. Do not override EEPROM from the profile — EEPROM is authoritative.
Preflight refuses with "per-board conf not found"Active profile points at a flash_config that does not exist under <bsp_image.root_path>/Linux_for_Tegra/.For custom carriers: run /jetson-derive-carrier to produce the conf. For reference devkits: re-run /jetson-promote-image / /jetson-init-image to repopulate the BSP image.
Preflight refuses because rootfs/etc/nv_tegra_release is missingapply_binaries.sh has not been run against <bsp_image.root_path>/Linux_for_Tegra/.Re-run /jetson-init-image (which invokes apply_binaries.sh) or run apply_binaries.sh manually from Linux_for_Tegra/.
boardctl -t <target> recovery errors out or has no effectWrong boardctl (a $PATH binary instead of the in-tree one) or a target name not enumerated by <boardctl> -h.Use the in-tree <bsp_image.root_path>/Linux_for_Tegra/tools/board_automation/boardctl; pick the target from <boardctl> -h. Fall back to manual recovery + reset buttons if needed.
Flash succeeds on T26x but the DUT stays in recovery / does not boot the new imageT26x does not auto-reset out of recovery after flash.sh / l4t_initrd_flash.sh returns.Run <boardctl> -t <target> reset (same target used for RCM entry), or for the manual path remove the force-recovery jumper / button and power-cycle. T23x / Orin needs no extra step.
First boot lands on Ubuntu's OEM configuration wizard despite expecting autologinNo default user was staged into rootfs before flashing.Either re-flash after staging via l4t_create_default_user.sh --autologin --accept-license (see the default-user-staging step), or complete the OEM wizard on the DUT this once.
flash.sh exits non-zero on an artifact-not-found error (DTB / BPFDTB / partition XML)EEPROM-driven dispatch resolved a filename that doesn't exist in bsp_image — usually /jetson-promote-image didn't copy the customized artifact, or the EEPROM SKU is unsupported by the BSP.Re-run /jetson-promote-image. If the SKU is unsupported, the DUT needs a different BSP version.
kernel/Image ↔ rootfs/boot/Image drift, missing bootloader/l4t_initrd.img / rootfs/boot/initrd, or modprobe "disagrees about version of symbol …" on the DUT/jetson-promote-image's mirror / refresh steps were skipped, or bsp_image was hand-edited outside Deploy.Re-run /jetson-promote-image and re-flash. If initramfs files are entirely absent, run /jetson-init-image first. See promote-image's kernel-image-and-initramfs reference for the manual escape hatch.

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 7 other files (references) in skills/jetson-flash-image of NVIDIA/skills.

  • SKILL.md
  • BENCHMARK.md
  • evals/evals.json
  • references/default-user-staging.md
  • references/eeprom-cross-check.md
  • references/recovery-mode-boardctl.md
  • skill-card.md
  • skill.oms.sig

Open the folder on GitHubat commit 0e0d506

Compare with similar skills

Jetson Flash 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 Flash Image compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Jetson Flash Image this skillNVIDIA/skills3.5k—~4.9kAutomated 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 today
    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 today
    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 today
    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 today
    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 today
    Auto-check: notes

Questions about Jetson Flash Image

What does Jetson Flash Image do?

A skill your agent uses to flash a promoted BSP image to a Jetson DUT in RCM mode via flash.sh or l4tinitrdflash.sh. Jetson Flash Image is an agent skill from NVIDIA/skills, published by the product's own GitHub organization.sh.

When should I use Jetson Flash Image?

Jetson Flash Image fits situations like: flash a promoted BSP image to a Jetson DUT in RCM mode via flash.sh; L4tinitrdflash.sh; BSP customization; image promotion.

How do I install Jetson Flash Image in Claude Code?

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

How do I install Jetson Flash Image in Codex?

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

Can I use Jetson Flash 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-flash-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-flash-image, .gemini/skills/jetson-flash-image, .github/skills/jetson-flash-image and .opencode/skills/jetson-flash-image in your project.

What does Jetson Flash Image need to run?

SKILL.md names no scripts, command-line tools or credentials: Jetson Flash Image is instructions for the agent only.

Does Jetson Flash Image access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Jetson Flash 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. Review the folder before installing.

What licence does Jetson Flash Image use?

Jetson Flash 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 Flash Image use?

About 4.9k tokens (SKILL.md is roughly 20k 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 2.1k tokens, read only when the agent opens those files.

What are the alternatives to Jetson Flash Image?

Skills that share tags, products or a category with Jetson Flash 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 Flash 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.