Official agent skill

Jetson Promote Image

by NVIDIA in NVIDIA/skills

A skill your agent uses to promote overlay files and built artifacts into the staged BSP image.

OfficialApache-2.0Auto-check: notesAI & LLM Engineering

Install Jetson Promote Image

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

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

GitHub CLI
$ gh skill install NVIDIA/skills jetson-promote-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-promote-image .claude/skills/jetson-promote-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-promote-image
GitHub stars
3.5k
Token cost
~5k tokens
SKILL.md length
2,207 words
Files
8 (incl. references)
Skills in repo
386
Repo updated
First seen
Licence
Apache-2.0

At a glance

A skill your agent uses to promote overlay files and built artifacts into the staged BSP image.

  • Promote overlay files and built artifacts into the staged BSP image
  • SKILL.md covers Purpose, Prerequisites, Overview and When to invoke, plus 5 more sections
  • Calls git
  • Tasks that involve GPU and accelerator computing

What it does

Jetson Promote Image is an agent skill from NVIDIA/skills, published by the product's own GitHub organization. Use to promote overlay files and built artifacts into the staged BSP image. Do NOT use to flash or build. Triggers: promote bsp image.

Its SKILL.md is about 5k 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/build-source-freshness-gate.md`).

It sits in AI & LLM Engineering, covering GPU and accelerator computing. It works with NVIDIA AI Platform, Linux and Git. 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

  • Promote overlay files and built artifacts into the staged BSP image
  • Tasks that involve GPU and accelerator computing

Example prompts

  • “/jetson-promote-image”

What it can do on your machine

Read from SKILL.md and the folder at commit dfdd080. 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

    Shell commands in SKILL.md call:

    • git

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

  • Network

    No URLs in SKILL.md. Its commands use git, 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 Promote Image loads about 5k tokens when it runs, and up to ~9.3k if it reads all its reference files. Until then it costs about 39 tokens; SKILL.md has 2,207 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~39
When it runs · the whole SKILL.md, loaded when a task matches
~5k
With references · SKILL.md plus every file in references/, read only if the agent opens them
~9.3k

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:268
    `rootfs/` were extracted with `sudo tar xpjf` by
  • NoteRuns commands with sudoSKILL.md:270
    bits the flashing toolchain reads back. `sudo cp -p`
  • NoteRuns commands with sudoSKILL.md:316
    n the host | Run on an account that can `sudo cp`; re-run resumes via diff-aware copy. |
  • NoteRuns commands with sudoSKILL.md:320
    st the stale kernel | Force the gate by `sudo touch <LFT_DST>/kernel/Image` + re-run promote, or run the two steps manua

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 dfdd080, republished under its Apache-2.0 licence (© NVIDIA). 2,207 words, ~4,992 tokens.

Download SKILL.mdSave it as .claude/skills/jetson-promote-image/SKILL.md (or your agent's skills folder). This skill also uses 7 other files; get the full folder from GitHub.
name
jetson-promote-image
description
Use to promote overlay files and built artifacts into the staged BSP image. Do NOT use to flash or build. Triggers: promote bsp image.
version
0.0.1
license
Apache-2.0
metadata.data-classification
public
metadata.author
Jetson Team
metadata.team
pts
metadata.tags
bsp, promote, deploy
metadata.domain
meta

Promote BSP Image

Purpose

Stage every Customize-* and Build output into bsp_image so it is ready for /jetson-flash-image. This is the promote leg of Deploy — it copies files, never flashes and never builds.

Prerequisites

  • Active target-platform profile with both source: and bsp_image: resolved (run /jetson-init-source and /jetson-init-image first).
  • <source.root_path>/Linux_for_Tegra/ initialized as a git repo (overlay tracker) with a clean working tree.
  • <bsp_image.root_path>/Linux_for_Tegra/ extracted from a BSP tarball + apply_binaries.sh already run.
  • git, yq, cmp, and sudo (for rootfs/* destinations) on the host.
  • <source.root_path>/.build-manifest.yaml + .build-state.yaml from /jetson-build-source (required when kernel-side repos have customize-* commits).

Overview

This is the promote leg of Deploy — see ../../context/bsp-customization-workflow.md for the pipeline view. The two channels this skill walks are:

ChannelSourceCarrierOwner
Overlay tracker<source.root_path>/Linux_for_Tegra/ (git repo at HEAD)Customize-* outputs that don't require a build (e.g. nvfancontrol.conf, nvpmodel.conf, BPMP DTB hand-edits)Customize customize-* skills commit here
Build manifest<source.root_path>/.build-manifest.yamlRebuilt kernel Image, in-tree .ko, OOT .ko, NVIDIA DTBsBuild jetson-build-source writes here

The skill computes the union of files to copy and writes each into <bsp_image.root_path>/Linux_for_Tegra/ with diff-aware skip-if-identical logic. When the copy pass touches the kernel Image or anything under rootfs/lib/modules/, it also rebuilds the initramfs via NVIDIA's tools/l4t_update_initrd.sh so the freshly promoted kernel + modules ship in the initrd the bootloader actually loads. After it returns, bsp_image carries every Customize and Build output. The skill does not flash and does not modify the workspace.

When to invoke

  • First leg of the typical Deploy chain jetson-promote-image → jetson-flash-image → jetson-validate-image.
  • Standalone, when the user wants bsp_image updated but isn't ready to flash yet (e.g. to inspect resolved files, run an out-of-band build that reads bsp_image, or hand bsp_image to a separate flashing host).

Procedure

Resolve active target + paths

Resolve the active profile per the contract in ../../context/target-platform-contract.md.

Refuse and route in these cases:

ConditionRefuse with
No active profile, or active: NARoute to /jetson-set-target or /jetson-init-target.
Profile lacks bsp_image:Route to /jetson-init-image.
<bsp_image.root_path>/Linux_for_Tegra/ missingRoute to /jetson-init-image.
<source.root_path>/Linux_for_Tegra/ missing or not a git repoRoute to /jetson-init-source.

Resolve paths:

  • <workspace> = parent of the active profile's target-platform/ directory (discovered at load time).
  • <bsp_image.root_path> from bsp_image.root_path: if present, else <workspace>/Image.
  • <source.root_path> from source.root_path: if present, else <workspace>/Source.

Bind shell variables for the rest of the procedure:

bash
LFT_SRC="<source.root_path>/Linux_for_Tegra"   # overlay tracker
LFT_DST="<bsp_image.root_path>/Linux_for_Tegra"
MANIFEST="<source.root_path>/.build-manifest.yaml"   # build outputs
Validate the two channels

The skill needs at least one channel populated. Refuse if the overlay tracker has uncommitted changes (status --porcelain non-empty), if $MANIFEST exists but doesn't parse as YAML, or if both channels are empty. Records OVERLAY_HAS_COMMITS / OVERLAY_HEAD and MANIFEST_PRESENT for downstream steps.

See references/copy-pass-snippets.md for the shell snippet and refuse messages.

Verify build-source freshness

Refuse if .build-state.yaml shows any kernel-side repo in Source/bsp_sources/ dirty since the last /jetson-build-source — otherwise the copy pass would silently ship stale artifacts. Detection rules + shell snippet in references/build-source-freshness-gate.md. Records BUILD_FRESH=1.

Pre-promote collision check (overlay only)

When the overlay tracks a remote, refuse if upstream has commits not yet pulled. Skip gracefully when no remote is configured (the default git init empty tracker from jetson-init-source). Manifest channel has no git remote concept — this check is overlay-only. Records COLLISION_CHECK for the Summary.

See references/copy-pass-snippets.md for the shell snippet.

Enumerate sources (both channels)

Channel A — overlay: git ls-files against $LFT_SRC is the source of truth (transparent to symlink mounts when source.repos.Linux_for_Tegra was overridden, excludes untracked / .gitignored files). Each entry maps src = $LFT_SRC/<rel> → dst = $LFT_DST/<rel>.

Channel B — manifest: parse artifacts[].{src,dst} from $MANIFEST. Refuse if any src is missing on disk (build was interrupted, or manifest stale — re-run /jetson-build-source). The manifest schema is written by jetson-build-source v0.2.0.

See references/copy-pass-snippets.md for both shell snippets and the manifest YAML schema.

Diff-aware copy into bsp_image

Iterate the union of overlay files and manifest entries. For each dst: if byte-identical, skip; otherwise cp -p (with sudo for rootfs/* destinations, where the sample rootfs was extracted as root). Tag INITRD_DIRTY=1 on any rootfs/lib/modules/* or kernel/Image write — the "Refresh initramfs" step gates on this flag. Counts / FIRST / LAST are recorded for the Summary.

Fail-fast: if any cp fails, surface the failed path and stop. bsp_image may be left partially updated — re-running after fixing the cause resumes via the diff-aware skip. Channel order is overlay first, then manifest: on a dst collision the manifest wins (freshly built artifact beats the older overlay copy).

See references/copy-pass-snippets.md for the copy_one() function and the two driving loops.

Mirror kernel Image into rootfs (when kernel changed)

The kernel Image lives in two paths inside bsp_image: <LFT_DST>/kernel/Image (read by the flash tool) and <LFT_DST>/rootfs/boot/Image (the rootfs-side copy, visible as /boot/Image from inside the rootfs chroot the refresh tool will run in). The build manifest only carries the kernel/Image dst, so this step mirrors kernel/Image → rootfs/boot/Image (diff-aware, no-op when already in sync) so the chrooted refresh tool resolves the kernel against the freshly promoted binary, not the stale rootfs copy. The mirror also sets INITRD_DIRTY=1 so a kernel-only promote (no rootfs/lib/modules/* writes) still triggers the refresh.

See references/kernel-image-and-initramfs.md for the shell snippet, the failure mode this prevents, and the INITRD_DIRTY corner case.

Refresh initramfs (when kernel or modules changed)

Run tools/l4t_update_initrd.sh from <LFT_DST>/ whenever INITRD_DIRTY=1 (set by the diff-aware copy or the mirror step above). The tool chroots into rootfs/, runs NVIDIA's nv-update-initrd, and writes both <LFT_DST>/bootloader/l4t_initrd.img (used by the flash tool) and <LFT_DST>/rootfs/boot/initrd (/boot/initrd on the DUT). Idempotent; ~30 s. Skip when INITRD_DIRTY=0 (overlay-only edits). DUT-side workarounds (update-initramfs -u + manual cp) are out of scope — fix the gap here so flash ships a coherent image.

See references/kernel-image-and-initramfs.md for the shell snippet, refuse paths, the "module shadowing" and "vermagic skew" failure modes the rebuild closes, and why bootloader/initrd (a different file) is left alone.

Summary

Report:

  • Overlay scope: overlay HEAD ($OVERLAY_HEAD) or "(empty)".
  • Manifest scope: mode=<...>, bsp_version=<...>, rebuilt_at=<...>, N artifacts or "(absent)".
  • Collision check: $COLLISION_CHECK.
  • Counts:
    • overlay: $COPIED_OVERLAY copied, $IDENTICAL_OVERLAY identical
    • manifest: $COPIED_MANIFEST copied, $IDENTICAL_MANIFEST identical
  • Kernel Image mirror: $KIMG_MIRRORED and initramfs: $INITRD_STATUS (copied … / rebuilt when triggered by kernel/Image or rootfs/lib/modules/* writes; skipped … otherwise).
  • First / last paths copied (omit if both COPIED totals are 0).
  • Resolved <source.root_path>, <bsp_image.root_path>.
  • Next step: /jetson-flash-image (or /jetson-validate-image if the user only wanted bsp_image refreshed for inspection / static validation).

Limitations

  • Two channels, one destination. bsp_image/Linux_for_Tegra/ is written by both passes. Overlay carries customize-* outputs (overlay-only edits like nvfancontrol.conf); manifest carries rebuilt binaries (kernel/OOT/DT). The two are intentionally disjoint by construction: build outputs don't go into the overlay, and customize-* edits to non-build files don't enter the manifest.
  • Build manifest is the trace-to-dirty contract. Anything in the manifest came from a dirty source repo (per jetson-build-source's "Write the build manifest" step trace policy). Promoting the manifest is therefore safe: every entry is a customization-bearing artifact, not toolchain-divergence noise. The skill does not re-derive the trace — it trusts the manifest.
  • Manifest entries can outlive their build outputs. If the user wipes Source/.build/ or bsp_sources/'s build artifacts between jetson-build-source and jetson-promote-image, the manifest will reference missing files. The "Enumerate sources (both channels)" step refuses in that case and points the user at /jetson-build-source to rebuild.
  • Manifest absence is fine when only overlay edits happened. A purely overlay-side customization (e.g. customize-fan) produces no build outputs and writes no manifest — the "Enumerate sources (both channels)" step is a no-op, the "Diff-aware copy into bsp_image" step promotes only overlay files. The skill prints "manifest: (absent)" in the summary and continues.
  • Diff-aware, idempotent. Re-running with no overlay commits or manifest changes since the last promote is a no-op (all files identical). Use this to confirm bsp_image is in sync without side effects.
  • Symlink-mount transparency. When source.repos.Linux_for_Tegra was overridden in jetson-init-source, the canonical mount is a symlink into <source.root_path>/.repos/Linux_for_Tegra/<subdir>. git -C, cp -p, and cmp -s all follow it transparently — no special handling needed at this layer. Manifest src paths are absolute, so symlinks under bsp_sources/ don't matter for the manifest channel.
  • sudo is scoped to rootfs/ destinations. Files under rootfs/ were extracted with sudo tar xpjf by jetson-init-image, so they carry root ownership and special mode bits the flashing toolchain reads back. sudo cp -p preserves them. Everything else (bootloader/, kernel/, kernel/dtb/, tools/, etc.) is user-owned and does not need sudo. This applies to both channels.
  • Channel-overlap precedence. If the same dst appears in both overlay and manifest, manifest wins (later in the "Diff-aware copy into bsp_image" step's loop). This is the desired semantic — manifest entries are freshly built, overlay entries may be older state. Hand-editing binary files into the overlay is discouraged (Build's job is to rebuild them); the precedence rule makes such mistakes recoverable.
  • bsp_image is read-only outside Deploy. This skill is the only writer in the normal flow (matches the workflow invariant). Hand-edits to <bsp_image.root_path>/Linux_for_Tegra/ outside Deploy will be silently overwritten on the next promote run if the same path exists in either channel; conversely they will not be reverted if no entry shadows them. Both behaviors are wrong for the diff trail — never hand-edit upstream.
  • Scope is overlay HEAD only (channel A). Named tags / manifests / commit ranges are deferred (see below). To promote a historical state, git -C $LFT_SRC checkout <ref> first, then re-run. The manifest channel has no ranged scope — it reflects whatever jetson-build-source's last run produced.
  • No automatic rollback on partial failure. If cp fails partway through, bsp_image is left in an intermediate state. Fix the underlying cause (usually permissions / disk full) and re-run — the "Diff-aware copy into bsp_image" step will resume by skipping already-promoted files.
  • Kernel Image mirror + initramfs refresh. Gated on copy-pass writes to kernel/Image or rootfs/lib/modules/*; the mirror feeds the refresh's chroot. Both are diff-aware and skipped on pure-overlay edits. tools/l4t_update_initrd.sh must exist in bsp_image (ships with apply_binaries.sh); a missing tool refuses and routes to /jetson-init-image. See references/kernel-image-and-initramfs.md for the full contract and failure modes.
Show full SKILL.md (616 more words)Show less

Troubleshooting

ErrorCauseSolution
Overlay has uncommitted changes at <LFT_SRC>Customize-* edits not committed before promoteRun git -C $LFT_SRC commit (or stash), then re-run.
origin has N unpulled commits on <upstream>Remote overlay diverged from localgit -C $LFT_SRC pull, resolve conflicts, then re-run.
Both overlay and manifest are empty — nothing to promoteNo Customize-* commits and no Build manifestRun a customize-* skill or /jetson-build-source first.
Kernel-side source(s) changed since last /jetson-build-sourceFreshness gate detected unprocessed customize-* edits under Source/bsp_sources/Commit pending edits, run /jetson-build-source, re-run promote.
Manifest entry references missing build output: <src>bsp_sources/ build outputs wiped or stale manifestRe-run /jetson-build-source to regenerate.
Build manifest at <MANIFEST> is not valid YAMLManifest hand-edited or partially writtenRe-run /jetson-build-source to rewrite the manifest.
cp: permission denied under rootfs/Missing sudo privilege on the hostRun on an account that can sudo cp; re-run resumes via diff-aware copy.
Profile lacks bsp_image: / source:Workspace not bootstrappedRun /jetson-init-image and/or /jetson-init-source.
tool not found at <LFT_DST>/tools/l4t_update_initrd.shtools/ was pruned, or bsp_image extracted from a non-NVIDIA tarballRe-run /jetson-init-image to repopulate.
l4t_update_initrd.sh exited non-zeroInsufficient sudo, broken rootfs (missing lib/modules/<ver>/modules.dep), or out-of-space /tmpRun depmod -a -b <LFT_DST>/rootfs <ver> against the rootfs first; verify /tmp headroom; rerun promote.
DUT boots with stale kernel / modules after promote, modules fail to load with disagrees about version of symbol …, or initramfs ships pre-customize modules even after the refresh ranThe mirror / refresh gate didn't fire (manual hand-edit under <LFT_DST> outside the skill), or rootfs/boot/Image drifted from kernel/Image so the chrooted refresh built against the stale kernelForce the gate by sudo touch <LFT_DST>/kernel/Image + re-run promote, or run the two steps manually: sudo cp -p <LFT_DST>/kernel/Image <LFT_DST>/rootfs/boot/Image && cd <LFT_DST> && sudo ./tools/l4t_update_initrd.sh. Then re-flash. See references/kernel-image-and-initramfs.md.

Spec status

Locked in for v0.2.0:

  • Two-channel scope — overlay HEAD + build manifest, both diff-aware, both copying into <bsp_image.root_path>/Linux_for_Tegra/.
  • Channel-overlap precedence — manifest wins on dst collision.
  • Source-repo collision check — overlay only; manifest has no remote concept and source repos under bsp_sources/ are not fetched (their state was sealed when jetson-build-source wrote the manifest).
  • Atomicity — fail-fast, no rollback. Diff-aware copy makes resume natural.
  • Audit trail — stdout-only at promote time. The overlay tracker's git log is the canonical record for channel A; the manifest itself is the canonical record for channel B.
  • Kernel Image mirror + initramfs refresh. Locked in as a paired step. The mirror copies kernel/Image → rootfs/boot/Image whenever the copy pass touched kernel/Image; the refresh runs tools/l4t_update_initrd.sh whenever kernel/Image or any rootfs/lib/modules/* was promoted, rebuilding both bootloader/l4t_initrd.img and rootfs/boot/initrd. Inseparable because the refresh chroots into rootfs/ and resolves the kernel through /boot/Image — the mirror has to run first. Closes both module-shadowing and vermagic-skew failure modes; both diff-aware, both skipped on overlay-only edits. Full contract in references/kernel-image-and-initramfs.md.

Still deferred:

  • Named-tag / commit-range scope for the overlay channel. Revisit when a "promote release X" use case appears.
  • Manifest history. Currently only the last build's manifest exists; if a user wants to roll bsp_image back to a previous build state, they'd need to re-run /jetson-build-source at the prior commit. A manifest archive (saved per-build-mode or per-commit) would enable rollback without rebuild.
  • Sidecar manifest in bsp_image. Revisit when promotion happens on a host that does not have access to the overlay tracker repo (or the workspace's manifest file).

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-promote-image of NVIDIA/skills.

  • SKILL.md
  • BENCHMARK.md
  • evals/evals.json
  • references/build-source-freshness-gate.md
  • references/copy-pass-snippets.md
  • references/kernel-image-and-initramfs.md
  • skill-card.md
  • skill.oms.sig

Open the folder on GitHubat commit dfdd080

Compare with similar skills

Jetson Promote 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 Promote Image compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Jetson Promote Image this skillNVIDIA/skills3.5k—~5kAutomated safety check: NotesApache-2.0
Graphsignalgraphsignal/graphsignal257—~6.2kAutomated safety check: PassApache-2.0
LLM Torch Profiler Trace AnalysisBBuf/AI-Infra-Auto-Driven-SKILLS925—~2.8kAutomated safety check: PassNone
Cv DeployLMIXR/CV_Deployment_skill170—~547Automated safety check: PassNone
TensorRT-LLM InferenceOrchestra-Research/AI-Research-SKILLs13k4 repos~1.3kAutomated safety check: PassMIT
Triton SkillslowlyC/agent-gpu-skills169—~1.3kAutomated safety check: PassMIT

Similar skills

  • Graphsignal

    graphsignal/graphsignal

    Profile AI inference workloads (vLLM, SGLang, TensorRT-LLM, PyTorch, any GPU application) with the Graphsignal profiler and read the results from its local /signals JSON endpoint.

    257 GitHub stars~6.2k tokensUpdated 11 days ago
    AI & LLM EngineeringAuto-check passed
  • LLM Torch Profiler Trace Analysis

    BBuf/AI-Infra-Auto-Driven-SKILLS

    Analyzes Torch Profiler traces from SGLang, vLLM and TensorRT-LLM servers into kernel attribution, overlap and fusion tables.

    925 GitHub stars~2.8k tokensUpdated 4 days ago
    AI & LLM EngineeringAuto-check passed
  • Cv Deploy

    LMIXR/CV_Deployment_skill

    基于 helpfile 工程经验,协助 agent 配置 CV 主机和边缘设备环境、编译视觉与推理依赖、接入摄像头视频并打包部署服务。适用于 Ubuntu、CentOS、Windows、macOS、Jetson、树莓派和 RK3399 的 CV 工程实施与故障排查,以及相关移动端配套工具;模型训练和纯算法设计不属于本技能主线。

    170 GitHub stars~547 tokensUpdated 10 days ago
    AI & LLM EngineeringAuto-check passed
  • TensorRT-LLM Inference

    Orchestra-Research/AI-Research-SKILLs

    Optimizes and serves LLMs on NVIDIA GPUs with TensorRT-LLM, covering quantization, in-flight batching, multi-GPU parallelism and the trtllm-serve command.

    13k GitHub starsUsed in 4 repos~1.3k tokens
    AI & LLM EngineeringAuto-check passed
  • Triton Skill

    slowlyC/agent-gpu-skills

    Write, debug, and optimize Triton and Gluon GPU kernels from local upstream tutorials, production kernels, language definitions, and compiler source.

    169 GitHub stars~1.3k tokensUpdated 2 mo ago
    AI & LLM EngineeringAuto-check passed
  • Megatron-Core LLM Training

    Orchestra-Research/AI-Research-SKILLs

    Sets up large-scale LLM training with NVIDIA Megatron-Core, choosing tensor, pipeline, data, context and expert parallelism for a given model size and GPU count.

    13k GitHub starsUsed in 2 repos~2.4k tokens
    AI & LLM EngineeringAuto-check passed

More from NVIDIA/skills

All 386 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 Promote Image

What does Jetson Promote Image do?

A skill your agent uses to promote overlay files and built artifacts into the staged BSP image. Jetson Promote Image is an agent skill from NVIDIA/skills, published by the product's own GitHub organization. Use to promote overlay files and built artifacts into the staged BSP image.

When should I use Jetson Promote Image?

Jetson Promote Image fits situations like: promote overlay files and built artifacts into the staged BSP image; tasks that involve GPU and accelerator computing.

How do I install Jetson Promote Image in Claude Code?

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

How do I install Jetson Promote Image in Codex?

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

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

What does Jetson Promote Image need to run?

Going by SKILL.md and its folder, Jetson Promote Image needs the command-line tools its instructions call (git).

Does Jetson Promote Image access the network?

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

Is Jetson Promote 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 Promote Image use?

Jetson Promote 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 Promote Image use?

About 5k 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 4.3k tokens, read only when the agent opens those files.

What are the alternatives to Jetson Promote Image?

Skills that share tags, products or a category with Jetson Promote Image: Graphsignal (graphsignal/graphsignal, 257 stars), LLM Torch Profiler Trace Analysis (BBuf/AI-Infra-Auto-Driven-SKILLS, 925 stars), Cv Deploy (LMIXR/CV_Deployment_skill, 170 stars) and TensorRT-LLM Inference (Orchestra-Research/AI-Research-SKILLs, 13k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Jetson Promote Image?

NVIDIA (a GitHub organization, an official publisher) maintains it in NVIDIA/skills, which has 3,546 GitHub stars. The repository holds 386 skills in this directory. The repository was last updated on October 9, 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.