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.
Set up the BSP source workspace: LinuxforTegra overlay tracker, bspsources, Crosstool-NG toolchain.
$ npx skills add NVIDIA/skills --skill jetson-init-source -a claude-codeProject install by default; add -g for ~/.claude/skills/.
$ gh skill install NVIDIA/skills jetson-init-source --agent claude-codeProject scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).
$ git clone --depth 1 https://github.com/NVIDIA/skills.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/jetson-init-source .claude/skills/jetson-init-source && rm -rf skills-srcUse ~/.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/
Install the "jetson-init-source" agent skill from https://github.com/NVIDIA/skills/tree/main/skills/jetson-init-source into .claude/skills/jetson-init-source/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "jetson-init-source", then confirm the skill loads.Claude Code copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$skill-installer install https://github.com/NVIDIA/skills/tree/main/skills/jetson-init-sourceType this inside Codex. $skill-installer <name> installs a curated skill from openai/skills. The installer writes to $CODEX_HOME/skills (default ~/.codex/skills). Restart Codex if the skill does not show up.
$ npx skills add NVIDIA/skills --skill jetson-init-source -a codexProject install goes to .agents/skills/; add -g for ~/.codex/skills/.
$ gh skill install NVIDIA/skills jetson-init-source --agent codexProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/NVIDIA/skills.git skills-src && mkdir -p .agents/skills && cp -r skills-src/skills/jetson-init-source .agents/skills/jetson-init-source && rm -rf skills-srcUse ~/.agents/skills/ instead of .agents/skills for a personal install.
Codex skills documentation · loads skills from .agents/skills/
Install the "jetson-init-source" agent skill from https://github.com/NVIDIA/skills/tree/main/skills/jetson-init-source into .agents/skills/jetson-init-source/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "jetson-init-source", then confirm the skill loads.Codex copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add NVIDIA/skills --skill jetson-init-source -a cursorProject install goes to .agents/skills/; add -g for ~/.cursor/skills/.
$ gh skill install NVIDIA/skills jetson-init-source --agent cursorProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/NVIDIA/skills.git skills-src && mkdir -p .cursor/skills && cp -r skills-src/skills/jetson-init-source .cursor/skills/jetson-init-source && rm -rf skills-srcUse ~/.cursor/skills/ instead of .cursor/skills for a personal install.
Cursor skills documentation · loads skills from .cursor/skills/, .agents/skills/, .claude/skills/, .codex/skills/
Install the "jetson-init-source" agent skill from https://github.com/NVIDIA/skills/tree/main/skills/jetson-init-source into .cursor/skills/jetson-init-source/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "jetson-init-source", then confirm the skill loads.Cursor copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gemini skills install https://github.com/NVIDIA/skills.git --path skills/jetson-init-source--scope user (default) or --scope workspace; --path is the subfolder of the repo that holds the skill; --consent skips the security confirmation prompt.
$ npx skills add NVIDIA/skills --skill jetson-init-source -a gemini-cliProject install goes to .agents/skills/; add -g for ~/.gemini/skills/.
$ gh skill install NVIDIA/skills jetson-init-source --agent gemini-cliProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/NVIDIA/skills.git skills-src && mkdir -p .gemini/skills && cp -r skills-src/skills/jetson-init-source .gemini/skills/jetson-init-source && rm -rf skills-srcUse ~/.gemini/skills/ instead of .gemini/skills for a personal install, then run /skills reload.
Gemini CLI skills documentation · loads skills from .gemini/skills/, .agents/skills/
Install the "jetson-init-source" agent skill from https://github.com/NVIDIA/skills/tree/main/skills/jetson-init-source into .gemini/skills/jetson-init-source/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "jetson-init-source", then confirm the skill loads.Gemini CLI copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gh skill install NVIDIA/skills jetson-init-sourceInstalls for Copilot at project scope by default; add --scope user for a personal install. Preview a skill first with gh skill preview. Needs GitHub CLI 2.90.0 or later (public preview).
$ npx skills add NVIDIA/skills --skill jetson-init-source -a github-copilotProject install goes to .agents/skills/; add -g for ~/.copilot/skills/.
$ git clone --depth 1 https://github.com/NVIDIA/skills.git skills-src && mkdir -p .github/skills && cp -r skills-src/skills/jetson-init-source .github/skills/jetson-init-source && rm -rf skills-srcUse ~/.copilot/skills/ instead of .github/skills for a personal install. Commit .github/skills so cloud agent and code review can use it.
GitHub Copilot skills documentation · loads skills from .github/skills/, .claude/skills/, .agents/skills/
Install the "jetson-init-source" agent skill from https://github.com/NVIDIA/skills/tree/main/skills/jetson-init-source into .github/skills/jetson-init-source/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "jetson-init-source", then confirm the skill loads.GitHub Copilot copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add NVIDIA/skills --skill jetson-init-source -a opencodeOpenCode documents no install command of its own. Project install goes to .agents/skills/; add -g for ~/.config/opencode/skills/.
$ gh skill install NVIDIA/skills jetson-init-source --agent opencodeProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/NVIDIA/skills.git skills-src && mkdir -p .opencode/skills && cp -r skills-src/skills/jetson-init-source .opencode/skills/jetson-init-source && rm -rf skills-srcUse ~/.config/opencode/skills/ instead of .opencode/skills for a personal install.
OpenCode skills documentation · loads skills from .opencode/skills/, .claude/skills/, .agents/skills/
Install the "jetson-init-source" agent skill from https://github.com/NVIDIA/skills/tree/main/skills/jetson-init-source into .opencode/skills/jetson-init-source/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "jetson-init-source", then confirm the skill loads.OpenCode copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
jetson-init-sourceSet up the BSP source workspace: LinuxforTegra overlay tracker, bspsources, Crosstool-NG toolchain.
Jetson Init Source is an agent skill from NVIDIA/skills, published by the product's own GitHub organization. Set up the BSP source workspace: LinuxforTegra overlay tracker, bspsources, Crosstool-NG toolchain. Use after jetson-init-image; not for fetching inputs.
Its SKILL.md is about 4.3k tokens, which your agent loads only when the skill is triggered. The skill folder holds 7 other files, including reference files (for example `BENCHMARK.md`, `evals/evals.json` and `references/branch-a-extraction.md`).
It sits in AI & LLM Engineering, covering GPU and accelerator computing. It works with NVIDIA AI Platform and Linux. 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.
5 steps, taken from the first numbered list in SKILL.md.
Read from SKILL.md and the folder at commit 0e0d506. It shows what the files ask for, not the result of running them.
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.
Shell commands in SKILL.md call:
gitbashFrom the folder's file list and the shell code blocks in SKILL.md.
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.
Names no API keys, tokens, secrets or passwords.
From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.
Jetson Init Source loads about 4.3k tokens when it runs, and up to ~5.4k if it reads all its reference files. Until then it costs about 44 tokens; SKILL.md has 1,689 words of instructions outside code blocks.
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.
The automated check found no risky patterns in SKILL.md.
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.
The full file from NVIDIA/skills at commit 0e0d506, republished under its Apache-2.0 licence (© NVIDIA). 1,689 words, ~4,250 tokens.
.claude/skills/jetson-init-source/SKILL.md (or your agent's skills folder). This skill also uses 5 other files; get the full folder from GitHub.This skill bootstraps the source-side workspace that customize-* / build
skills depend on: the Linux_for_Tegra overlay tracker (git repo for
pristine + customization commits), the bsp_sources/ mono-tree (kernel,
OOT, nvgpu, display, hwpm, hardware DTs), and a working NVIDIA
Crosstool-NG cross-compile prefix. It owns only the source: block in
the active profile and the on-disk source workspace under
<source.root_path> (default: <workspace>/Source).
Responsibilities:
source.root_path.Linux_for_Tegra overlay tracker.bsp_sources using the precedence in the "Materialize the BSP-sources baseline" step.source.toolchain.source.repos:.<source.root_path>/Linux_for_Tegra/".jetson-init-image (Setup's next step on a fresh target).Follow the shared
quick_start_prefill contract.
This skill has source-specific mappings:
quick_start_prefill.source.public_sources_archive maps to the
Branch-A archive candidate.quick_start_prefill.source.repos maps to proposed source.repos:
entries; validate reserved keys and url: / archive: mutual
exclusion before writing.quick_start_prefill.source.toolchain may be a cross-compile prefix,
a gcc path, a containing bin/ directory, an x-tools.tbz2 archive
path, or skip.This skill remains the only owner of source.root_path, source.repos:,
and source.toolchain profile writes.
Resolve the active profile + workspace defaults per the contract in
../../context/target-platform-contract.md.
<bsp_image.root_path> does not contain
Linux_for_Tegra/ (BSP not extracted — route to /jetson-init-image).source.root_path:, use it. Otherwise
<source.root_path> defaults to <workspace>/Source; use that default
silently and do not write source.root_path: to the profile. Ask only
for an explicit custom path, unrelated content at the default path, or an
unwritable parent.Read source.repos: (if present) into a map keyed by entry name,
each carrying optional url, ref, subdir, path. Reserved keys:
Linux_for_Tegra (overlay tracker), bsp_sources (kernel-source
repo). Every other key is an extra user-defined repo.
source.root_path overrideOnly when source.root_path is absent from the profile and one of the
override conditions above applies:
source.root_path: default =<workspace>/Source. Press Enter to accept, or enter an absolute path to override.
target-platform/<active>.yaml in place to add/update
source.root_path:. Preserve all other blocks, comments, and
quoting — use a round-tripping YAML loader (e.g. ruamel.yaml).Fires at most once per profile. Otherwise create <workspace>/Source as
needed and continue without prompting.
Linux_for_TegraMount path is canonical: <source.root_path>/Linux_for_Tegra/.
Default (no source.repos.Linux_for_Tegra entry):
LFT="<source.root_path>/Linux_for_Tegra"
mkdir -p "$LFT"
[ -d "$LFT/.git" ] || git -C "$LFT" initEmpty tracker. Do not commit anything here — pristine imports happen file-by-file when customization skills run.
Override (url, ref, optional subdir):
# Clone the user's repo to a side location, then mount the
# expected tree (subdir or repo root) at the canonical path.
CLONE="<source.root_path>/.repos/Linux_for_Tegra"
git clone <url> -b <ref> "$CLONE"
ln -s "$CLONE/<subdir or .>" "<source.root_path>/Linux_for_Tegra"If the mount already exists with valid git state, skip; refuse if it exists with unrelated content.
Three branches, dispatched in precedence order against the
profile entry source.repos.bsp_sources:
| Order | Profile state | Branch |
|---|---|---|
| 1 | url: set | C. Customer git clone (explicit override always wins) |
| 2 | archive: set, OR entry absent AND <workspace>/Downloads/public_sources.tbz2 exists | A. Local archive extraction (default) |
| 3 | Entry absent AND no local archive | B. source_sync.sh (fallback) |
url: and archive: are mutually exclusive — refuse if both are
set in the same entry.
Branch A is the preferred default because it sidesteps NVIDIA git egress entirely (the most common Setup failure mode). Branch B exists for fresh workspaces with no pre-downloaded tarball. Branch C is for customer forks of the whole BSP layout.
Default branch: extract a pre-downloaded public_sources.tbz2 into
<source.root_path>/bsp_sources/ as a single mono-repo (git init +
pristine commit). See
references/branch-a-extraction.md
for the full archive shape, path-resolution rules, and the extraction
script (including the Tegra OOT Makefile force-replace workaround for
R36.x).
Branches B and C may produce per-component repos instead; downstream
build logic still walks the canonical sub-paths under
<source.root_path>/bsp_sources/.
source_sync.sh (fallback)Runs only when no local archive is found and no url: is set.
Create the bsp_sources/ mount directory under <source.root_path>
and run source_sync.sh from the extracted BSP with two flags:
mkdir -p "<source.root_path>/bsp_sources"
bash "<bsp_image.root_path>/Linux_for_Tegra/source/source_sync.sh" \
-d "<source.root_path>/bsp_sources" \
-t "jetson_<major.minor>"-d <source.root_path>/bsp_sources — write clones into the
bsp_sources/ subdir of the workspace, so the on-disk folder
matches the schema key. Without -d, the script writes under its
own directory (the BSP itself) — wrong for the overlay model.-t jetson_<major.minor> — pin the tag to the BSP release line.
Derive from bsp_image.version by truncating to the first two
dotted components: "38.4.0" → jetson_38.4. Tag-format
fallback: if rejected, try jetson_<bsp_image.version> (older
L4T sometimes uses the full form). If that also fails, surface the
error and stop — never fall back to "latest" silently.Refuse if source_sync.sh does not exist: re-run /jetson-init-image
to repopulate Linux_for_Tegra/source/.
source_sync.sh exits 0 even when every clone failed — verify by
counting Failed to clone lines in its output and refuse if
non-zero. The most likely cause of universal failure is blocked
git egress to gitlab.com/nvidia/nv-tegra /
nv-tegra.nvidia.com; surface that explicitly and route the user
to download public_sources.tbz2 via /quick-start for Branch A.
url: override)Triggered by an explicit url: field. Clone the customer repo once
and expose its canonical kernel-side sub-paths under
<source.root_path>/bsp_sources/. The canonical sub-path list is
read from source_sync.sh's SOURCE_INFO at runtime — do not
hard-code it, so future NVIDIA additions/removals propagate
automatically:
# Parse canonical sub-paths from source_sync.sh's SOURCE_INFO
# (only the kernel-side entries marked `k:` in the second field).
SUBPATHS=$(grep -oP '^\s*k:[^:]+:' \
"<bsp_image.root_path>/Linux_for_Tegra/source/source_sync.sh" \
| sed 's/^\s*k://; s/:$//')
mkdir -p "<source.root_path>/bsp_sources"
CLONE="<source.root_path>/.repos/bsp_sources"
git clone <url> -b <ref> "$CLONE"
ROOT="$CLONE/<subdir or .>"
for SUB in $SUBPATHS; do
[ -d "$ROOT/$SUB" ] && \
ln -s "$ROOT/$SUB" "<source.root_path>/bsp_sources/$SUB"
doneReport any canonical sub-path expected for the active chip family but not present inside the customer repo (warn, don't refuse — customer may legitimately not have all repos).
The downstream jetson-build-source reads source.toolchain from
this profile and exports it as CROSS_COMPILE. This step must
land a valid prefix before init-source returns, or any subsequent
kernel / OOT / DT build will refuse.
NVIDIA's official Crosstool-NG Toolchain gcc is the canonical
toolchain for L4T. jetson-download-bsp owns any network fetch of
x-tools.tbz2; this skill only discovers, extracts, validates, and
writes the resolved prefix. Resolution follows a three-step ladder:
Look under <workspace>/toolchain/x-tools/ for the Crosstool-NG
layout — typically one of:
<workspace>/toolchain/x-tools/aarch64-none-linux-gnu/bin/aarch64-none-linux-gnu-gcc
<workspace>/toolchain/x-tools/aarch64-buildroot-linux-gnu/bin/aarch64-buildroot-linux-gnu-gccGlob: <workspace>/toolchain/x-tools/aarch64-*-linux-gnu/bin/aarch64-*-linux-gnu-gcc.
If exactly one match, bind:
TC_PREFIX=<absolute path to that .../bin/<triple>-> # trailing dash mandatorySkip to "Write to profile" below. If zero matches, fall through
to "Auto-extract from Downloads/x-tools.tbz2" below. If multiple, refuse with the list and ask the user to
remove the unwanted ones (we never pick one silently among
ambiguous installs — different Crosstool-NG flavors produce ABI-
incompatible binaries).
Downloads/x-tools.tbz2If <workspace>/Downloads/x-tools.tbz2 exists (mirrors the
public_sources.tbz2 Branch-A pattern in the "Materialize the BSP-sources baseline" step — air-gapped /
no-egress users drop archives there):
file -b "<workspace>/Downloads/x-tools.tbz2" | grep -q "bzip2 compressed" || \
refuse "<workspace>/Downloads/x-tools.tbz2 is not a bzip2 tarball"
mkdir -p "<workspace>/toolchain"
tar xjf "<workspace>/Downloads/x-tools.tbz2" -C "<workspace>/toolchain"Then re-run the "Auto-discover" pass above. Refuse if extraction succeeds
but no x-tools/aarch64-*-linux-gnu/bin/ is produced (archive
content doesn't match the Crosstool-NG layout).
If both Auto-discover and Auto-extract came up empty, ask:
No Crosstool-NG toolchain found at
<workspace>/toolchain/or in<workspace>/Downloads/x-tools.tbz2.Reply with one of:
- absolute path to your
aarch64-*-linux-gnu-gccbinary or its containingbin/directory,cancelto abort.To fetch the archive instead, cancel this run, run
/jetson-download-bsp, then re-run/jetson-init-source.
For a path reply, validate via [ -f "${TC_PREFIX}gcc" ]. Refuse
and re-prompt on failure.
Once $TC_PREFIX resolves and ${TC_PREFIX}gcc exists, write it
into the active profile using a round-tripping YAML loader:
source:
toolchain: <TC_PREFIX> # absolute, with trailing dashIf source: is otherwise empty (no root_path override, no
repos: entries), the source: block is now non-empty and stays
in the profile. Future jetson-init-source runs skip the "Resolve cross-compile toolchain" step
if source.toolchain is already set and points at a working gcc.
For each entry under source.repos: whose name is not
Linux_for_Tegra or bsp_sources:
MOUNT="<source.root_path>/<entry.path or entry.name>"
if [ -n "<entry.subdir>" ]; then
CLONE="<source.root_path>/.repos/<entry.name>"
git clone <entry.url> -b <entry.ref> "$CLONE"
ln -s "$CLONE/<entry.subdir>" "$MOUNT"
else
git clone <entry.url> -b <entry.ref> "$MOUNT"
fiRefuse if a mount path already exists with unrelated content.
Print:
<workspace>, <bsp_image.root_path>, and
<source.root_path>.bsp_sources: branch selected plus key evidence (archive path,
source_sync.sh failure count, or clone URL/ref).${TC_PREFIX}gcc --version
first line.<source.root_path>/Linux_for_Tegra/; promote is what later copies
committed overlay changes into bsp_image.If a downstream skill triggered this run, tell the user to re-issue their original request.
Linux_for_Tegra and bsp_sources mount paths are canonical.
path: applies only to extra user-defined repos.Linux_for_Tegra tracker is intentionally empty; do not
pre-populate it.bsp_sources precedence is url: → archive: → auto-discovered
Downloads/public_sources.tbz2 → source_sync.sh. url: and
archive: are mutually exclusive.source.repos.bsp_sources.archive:.$DEST/Makefile collision. Inner tarballs in
public_sources.tbz2 ship two files named Makefile: the Tegra
orchestrator (kernel_oot_modules_src.tbz2) and the dGPU/OpenRM
proprietary Makefile (nvidia_kernel_display_driver_source_without_ root_dir.tbz2). Alphabetical extraction order lets the dGPU one
win on R36.x; downstream arm64 cross-builds then fail with
'-mlittle-endian' unrecognized. Step 3a force-replaces from
<bsp_image>/Linux_for_Tegra/source/Makefile when the Tegra
modules: hwpm nvidia-oot nvgpu nvidia-display signature is
missing. R38+ extractions already match; the check is a no-op there.source_sync.sh
sub-path layout, optionally shifted by subdir:.source_sync.sh tag from bsp_image.version as
jetson_<major.minor> first; never fall back to an unpinned latest.jetson-download-bsp owns network downloads of public_sources.tbz2
and x-tools.tbz2; this skill consumes local archives only.source.toolchain must be an NVIDIA Crosstool-NG prefix with trailing
dash and a working ${prefix}gcc. Never silently use $PATH.../../context/target-platform-contract.md./jetson-init-image already run so bsp_image.version is recorded
(Branch B source_sync.sh tag derives from it).public_sources.tbz2 (and optionally
x-tools.tbz2) staged under Downloads/.source: block; never edits bsp_image,
reference_devkit, custom_carrier, or documents.source_sync.sh) and Branch C
(customer Git clone); Branch A is fully offline.${toolchain}gcc not found — re-stage x-tools.tbz2 under
Downloads/ and rerun, or pass a verified absolute prefix path.source_sync.sh cannot resolve jetson_<major.minor> tag — the
recorded bsp_image.version is wrong; re-run /jetson-init-image
to refresh it.Linux_for_Tegra/.git shows uncommitted hand-edits — abort and
ask the user to commit or stash; this skill expects a clean tracker.source_sync.sh; set subdir: to the right sub-root or fall
back to multi-repo overrides under source.repos:.../../references/platform_template.yaml — source: schema, including the repos: map.../../context/target-platform-contract.md — target-platform contract.../../context/bsp-customization-workflow.md — Workspace edit protocol.../jetson-init-target/SKILL.md — authors the profile this skill consumes.../jetson-init-image/SKILL.md — extracts the BSP and back-fills bsp_image.version; run before this skill.© 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
SKILL.md and 5 other files (references) in skills/jetson-init-source of NVIDIA/skills.
Open the folder on GitHubat commit 0e0d506
Jetson Init Source 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.
| Skill | Stars | Used in | Tokens | Auto-check | Licence | Repo updated |
|---|---|---|---|---|---|---|
| Jetson Init Source this skillNVIDIA/skills | 3.5k | — | ~4.3k | Automated safety check: Pass | Apache-2.0 | |
| Graphsignalgraphsignal/graphsignal | 257 | — | ~6.2k | Automated safety check: Pass | Apache-2.0 | |
| LLM Torch Profiler Trace AnalysisBBuf/AI-Infra-Auto-Driven-SKILLS | 900 | — | ~2.8k | Automated safety check: Pass | None | |
| TensorRT-LLM InferenceOrchestra-Research/AI-Research-SKILLs | 13k | 5 repos | ~1.3k | Automated safety check: Pass | MIT | |
| Cv DeployLMIXR/CV_Deployment_skill | 126 | — | ~547 | Automated safety check: Pass | None | |
| Megatron-Core LLM TrainingOrchestra-Research/AI-Research-SKILLs | 13k | 3 repos | ~2.4k | Automated safety check: Pass | MIT |
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.
BBuf/AI-Infra-Auto-Driven-SKILLS
Analyzes Torch Profiler traces from SGLang, vLLM and TensorRT-LLM servers into kernel attribution, overlap and fusion tables.
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.
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。
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.
slowlyC/agent-gpu-skills
Write, debug, and optimize Triton and Gluon GPU kernels from local upstream tutorials, production kernels, language definitions, and compiler source.
NVIDIA/skills
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.
NVIDIA/skills
Generates, validates, compares and explains HOLOLINK_def.svh macro files for the HSB IP, using bundled Python scripts and asking before it writes anything.
NVIDIA/skills
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.
NVIDIA/skills
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.
NVIDIA/skills
Orchestrates video data augmentation and auto-labeling workflows on OSMO, from flow selection and preflight checks to submission, monitoring and output download.
NVIDIA/skills
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.
Works with
Categories
Set up the BSP source workspace: LinuxforTegra overlay tracker, bspsources, Crosstool-NG toolchain. Jetson Init Source is an agent skill from NVIDIA/skills, published by the product's own GitHub organization. Set up the BSP source workspace: LinuxforTegra overlay tracker, bspsources, Crosstool-NG toolchain.
Jetson Init Source fits situations like: tasks that involve GPU and accelerator computing.
Run `npx skills add NVIDIA/skills --skill jetson-init-source -a claude-code`. Or copy the skill folder (skills/jetson-init-source in NVIDIA/skills) into .claude/skills/jetson-init-source in your project. Claude Code loads it when a task matches its description.
Run `npx skills add NVIDIA/skills --skill jetson-init-source -a codex`. Or copy the skill folder (skills/jetson-init-source in NVIDIA/skills) into .agents/skills/jetson-init-source in your project. Codex loads it when a task matches its description.
Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add NVIDIA/skills --skill jetson-init-source -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-init-source, .gemini/skills/jetson-init-source, .github/skills/jetson-init-source and .opencode/skills/jetson-init-source in your project.
Going by SKILL.md and its folder, Jetson Init Source needs the command-line tools its instructions call (git and bash).
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.
Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.
Jetson Init Source 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.
About 4.3k tokens (SKILL.md is roughly 17k 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 1.1k tokens, read only when the agent opens those files.
Skills that share tags, products or a category with Jetson Init Source: Graphsignal (graphsignal/graphsignal, 257 stars), LLM Torch Profiler Trace Analysis (BBuf/AI-Infra-Auto-Driven-SKILLS, 900 stars), TensorRT-LLM Inference (Orchestra-Research/AI-Research-SKILLs, 13k stars) and Cv Deploy (LMIXR/CV_Deployment_skill, 126 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.
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.