复制安装命令
用 Codex 或 Claude 安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它先审查 Skill 页面再帮你安装。
复制前请先查看来源、License 和安全提示。
SkillAtlasSkill 详情
jetson-memory-audit
Official, NVIDIA-verified Agent Skills for Claude Code, Codex, and other coding agents.
审核状态：已审核Quality 80Security 62
项目 README
来源文件：README.md
NVIDIA Agent Skills
Official, NVIDIA-verified Agent Skills for Claude Code, Codex, and other coding agents.
📖 Docs: docs.nvidia.com/skills · 📺 Livestream: From Vulnerable to Verified · 📝 Blog: NVIDIA Verified Agent Skills: Capability Governance for AI Agents
Skills are portable instruction sets that teach AI agents how to use NVIDIA software optimally: Physical AI and robotics workflows, simulation, CUDA-X libraries, RAG and AI Blueprints, and platform tools. This repository is a catalog: skills are maintained in their respective product repos, and mirrored here daily via an automated sync pipeline. Skills are being added continuously, so check back for updates. We are building this infrastructure in the open, and contributions are welcome. See the Roadmap for what is planned next.
Quickstart
Install NVIDIA skills with the default skills CLI flow:
npx skills add nvidia/skills
The CLI runs through npx and prompts you to choose a skill and install destination. You do not need to clone this repo or copy skill folders by hand.
Requires a current
skillsCLI (v1.5.16 or newer). Installing vianpx skills@latest add nvidia/skillsalways uses the latest. On older CLIs (v1.5.15 and earlier), skills may install but not appear in Claude Code — see Troubleshooting.
The skill is available the next time your agent loads skills and encounters a relevant task. For example, ask your agent to "solve a linear programming problem with cuOpt" and the skill guides it through the cuOpt Python API. In Claude Code, run /reload-skills to load newly installed skills in your current session.
Install One Skill Without Prompts
Use this when you already know the skill name and want to skip prompts.
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --yes
Replace cuopt-numerical-optimization-api with any skill name from the Skill Catalog.
Install for a Specific Agent
Use --agent to target a specific AI coding agent. Initially, we'll support common client targets, expanding the list over time. For the full list of clients supported by the spec, see the skills CLI Supported Agents table.
Claude Code
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --agent claude-code
Codex
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --agent codex
Snowflake CoCo
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --agent cortex
Cursor
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --agent cursor
Kiro
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --agent kiro-cli
Use --agent more than once to install the same skill into multiple agents.
npx skills add nvidia/skills \
--skill cuopt-numerical-optimization-api \
--agent claude-code \
--agent codex \
--agent cursor \
--agent kiro-cli
Keep Skills Up to Date
New skills land continuously, and existing ones are revised, renamed, or consolidated as the catalog evolves. Refresh what you have installed with:
npx skills update
Run it interactively and the CLI also flags skills that were removed or merged upstream (for example, when several skills are consolidated into one) and offers to remove the stale local copies. Use npx skills list to see what is installed and npx skills check to preview what is out of date first.
Browse the Catalog
Use this when you want to see available NVIDIA skills before installing anything.
npx skills add nvidia/skills --list
For non-interactive installs, global installs, agent-specific installs, updates, removals, and fallback manual copying, see Advanced installation.
Skill Catalog
Getting Help & Contributing
Where to file an issue depends on what's broken:
- Skill content issues (a specific skill has a bug, missing functionality, or incorrect content) — file in the source repo for that product, using the per-product table below.
- Catalog issues (catalog README errors, sync workflow problems, distribution channels, signing/verification flow, docs in this repo) — file here using the catalog issue templates: Bug Report, Feature Request, or Documentation Request or Correction.
- Questions or general discussion — use Discussions. The issue tracker is reserved for bug reports, feature proposals with a design, and documentation issues.
- Security vulnerabilities — follow the disclosure process in SECURITY.md; do not open a public issue.
Per-product source repo links:
For issues with this catalog repo itself (README, structure, listing a new product): open an issue here.
Verifying Skills
Every published skill ships with a detached OMS signature (skill.oms.sig). The sync pipeline drops any skill missing the required artifacts before publishing, so every skill in the catalog carries:
SKILL.md— the skill instructions consumed by the agentskill-card.md— skill identity and governance cardskill.oms.sig— detached OMS signature (verifiable againstnv-agent-root-cert.pem)- A Tier-3 evaluation dataset — accepted at
evals/evals.json,evals/*.json,eval/*.json, orbenchmark/evals.json BENCHMARK.md— generated benchmark report capturing verifiable uplift data
Verify a skill against the NVIDIA trust anchor nv-agent-root-cert.pem:
pip install model-signing
model_signing verify certificate SKILL_DIR \
--signature SKILL_DIR/skill.oms.sig \
--certificate_chain nv-agent-root-cert.pem \
--ignore_unsigned_files
A successful verification confirms that the skill contents have not been modified since signing by NVIDIA.
See Verify Signed Agent Skills for signature layout, the trust pipeline, and policy options.
Roadmap
- ✅ Public skills catalog with NVIDIA-verified skills across multiple products
- ✅ Automated sync pipeline with skills mirrored from product repos daily
- ✅ Security scanning for all published skills covering instruction safety and supply-chain integrity
- ✅ Skills signing so every published skill carries a verifiable NVIDIA signature
- ✅ Skills universal evaluation criteria and task-specific criteria
- ✅ Skill Card with machine-readable metadata for identity, provenance, quality, and behavioral boundaries
- ✅ Sync-time compliance gates — signature drift detection and missing-artifact enforcement
- ✅ Syndication to external marketplaces — Skills.sh, Codex plugin, Claude Code plugin, ClawHub, Hermes Hub
- 🔲 Syndication to additional MCP hubs and partner channels
Repository Structure
NVIDIA/skills/
├── skills/ # NVIDIA-verified skills (count grows continuously),
│ │ synced from upstream product repos
│ ├── README.md # Browser-facing install guidance
│ ├── <product-prefix>-*/ # Flat layout — one dir per skill, product-prefixed
│ │ # e.g. aiq-*, cuopt-*, cupynumeric-*,
│ │ # dali-*, deepstream-*, dicom-*, digital-health-*,
│ │ # dynamo-*, earth2studio-*, holoscan-*, hsb-*,
│ │ # jetson-*, launch-nemo-rl, mcore-*,
│ │ # nemo-automodel-*, nemo-data-designer-plugin,
│ │ # nemo-evaluator-plugin, nemo-mbridge-* (20 skills),
│ │ # nemo-retriever, nemo-rl-* (4 skills),
│ │ # nemoclaw-user-guide, nemotron-*, nemotron-speech,
│ │ # nv-* (medical AI), physicsnemo-*, rag-*,
│ │ # skill-card-generator, tao-*, tilegym-*,
│ │ # vss-* (15 skills), accelerated-computing-cudf,
│ │ # cudaq-guide, portfolio-optimization
│ ├── omniverse-*/ # Physical AI — manually staged (see manual-components.yml)
│ └── physical-ai-*/ # Physical AI — manually staged
├── components.d/ # Product registry — one file per component, teams onboard here
│ ├── README.md # Schema and onboarding instructions
│ └── <product>.yml # one file per registered product
├── plugins/ # Packaged plugin distributions
│ └── nvidia-skills/ # Curated NVIDIA skills bundle (Claude Code, Codex)
├── plugins.d/ # Plugin build registry — config for `build-plugins.py`
│ ├── README.md
│ ├── _defaults.yml
│ └── nvidia-skills.yml
├── .claude-plugin/ # Claude Code marketplace metadata
│ └── marketplace.json
├── .agents/plugins/ # Agent marketplace metadata (other clients)
│ └── marketplace.json
├── docs/ # Long-form documentation (published via Fern)
│ ├── README.md # How to build the docs locally
│ ├── index.mdx
│ ├── advanced-install.mdx
│ ├── agent-skill-trust-pipeline.mdx
│ ├── release-checklist.mdx
│ ├── scanning-agent-skills.mdx
│ ├── signing-agent-skills.mdx
│ └── skill-cards.mdx
├── fern/ # Fern docs site configuration
├── .github/
│ ├── workflows/ # Sync pipeline, plugin validation, DCO check, author verify
│ └── scripts/ # regenerate-readme.sh, build-plugins.py,
│ # manual-components.yml (temp Physical AI catalog
│ # exception, removed after Computex 2026),
│ # marketplace/metadata.json (skill metadata sidecar)
├── nv-agent-root-cert.pem # Trust anchor for OMS signature verification
├── skills.sh.json # Skills.sh marketplace grouping config
├── CHANGELOG.md
├── CONTRIBUTING.md # Contribution guidelines
├── SECURITY.md # Security reporting policy
├── CODE_OF_CONDUCT.md # Community code of conduct
├── LICENSE-APACHE # Apache 2.0 (source code)
└── LICENSE-CC-BY-4.0 # CC BY 4.0 (documentation/skills)
Skills are maintained in their respective product repos (see the Source column in the Skill Catalog) and synced to this repo daily. Products only appear under skills/ after the sync pipeline confirms each skill carries:
skill.oms.sig— detached OMS-format signature (verifiable againstnv-agent-root-cert.pem)skill-card.md— skill identity and governance card- A Tier-3 evaluation dataset — accepted at
evals/evals.json,evals/*.json,eval/*.json, orbenchmark/evals.json
When evaluation runs produce a BENCHMARK.md, it ships alongside the skill so consumers can see verifiable benchmark uplift data.
Standards & Compatibility
This repository adheres to the Agent Skills specification:
- Skills are portable directories with a
SKILL.mdfile at their root. - Metadata uses YAML frontmatter with required
nameanddescriptionfields. - Skills follow a progressive disclosure model — lightweight metadata loads at startup, full instructions load on activation.
- Validate your skill using the
skills-refreference library.
License
Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
This code is dual-licensed with documentation/skills under the CC-BY-4.0 AND source code under Apache-2.0 license terms. The full license texts can be found in LICENSE-APACHE and LICENSE-CC-BY-4.0 respectively.
数据与 AI
高风险
- 来源需自行核对维护者身份。
- 包含脚本或命令调用，安装前请复核。
- 未检测到明显外部权限要求。
- 存在潜在风险命令，请谨慎安装。
- 扫描发现：3 条。
Codex — Git Clone 安装
- 安装前请先查看来源仓库和风险报告。
- 克隆仓库：git clone https://github.com/NVIDIA/skills.git
- 将 "skills/jetson-memory-audit" 文件夹复制到 Codex 的 skills 目录中。
- 重启 Codex 让新的 skill 生效。
Codex — 手动复制安装
- 安装前请先查看来源仓库和风险报告。
- 从源仓库下载 SKILL.md 及相关文件。
- 在 Codex 的 skills 目录中创建新文件夹。
- 将所有 skill 文件复制到新文件夹中。
- 重启 Codex 让新的 skill 生效。
Claude Code — Git Clone 安装
- 安装前请先查看来源仓库和风险报告。
- 克隆仓库：git clone https://github.com/NVIDIA/skills.git
- 将 "skills/jetson-memory-audit" 文件夹复制到 Claude Code 的 skills 目录中。
- 重启 Claude Code 让新的 skill 生效。
Claude Code — 手动复制安装
- 安装前请先查看来源仓库和风险报告。
- 从源仓库下载 SKILL.md 及相关文件。
- 在 Claude Code 的 skills 目录中创建新文件夹。
- 将所有 skill 文件复制到新文件夹中。
- 重启 Claude Code 让新的 skill 生效。
Cursor — Git Clone 安装
- 安装前请先查看来源仓库和风险报告。
- 克隆仓库：git clone https://github.com/NVIDIA/skills.git
- 将 "skills/jetson-memory-audit" 文件夹复制到 Cursor 的 skills 目录中。
- 重启 Cursor 让新的 skill 生效。
Cursor — 手动复制安装
- 安装前请先查看来源仓库和风险报告。
- 从源仓库下载 SKILL.md 及相关文件。
- 在 Cursor 的 skills 目录中创建新文件夹。
- 将所有 skill 文件复制到新文件夹中。
- 重启 Cursor 让新的 skill 生效。
GitHub Copilot — Git Clone 安装
- 安装前请先查看来源仓库和风险报告。
- 克隆仓库：git clone https://github.com/NVIDIA/skills.git
- 将 "skills/jetson-memory-audit" 文件夹复制到 GitHub Copilot 的 skills 目录中。
- 重启 GitHub Copilot 让新的 skill 生效。
GitHub Copilot — 手动复制安装
- 安装前请先查看来源仓库和风险报告。
- 从源仓库下载 SKILL.md 及相关文件。
- 在 GitHub Copilot 的 skills 目录中创建新文件夹。
- 将所有 skill 文件复制到新文件夹中。
- 重启 GitHub Copilot 让新的 skill 生效。
Windsurf — Git Clone 安装
- 安装前请先查看来源仓库和风险报告。
- 克隆仓库：git clone https://github.com/NVIDIA/skills.git
- 将 "skills/jetson-memory-audit" 文件夹复制到 Windsurf 的 skills 目录中。
- 重启 Windsurf 让新的 skill 生效。
Windsurf — 手动复制安装
- 安装前请先查看来源仓库和风险报告。
- 从源仓库下载 SKILL.md 及相关文件。
- 在 Windsurf 的 skills 目录中创建新文件夹。
- 将所有 skill 文件复制到新文件夹中。
- 重启 Windsurf 让新的 skill 生效。
查看 SKILL.md 原文
name: jetson-memory-audit
description: Measure Jetson DRAM/NvMap usage and verify before/after memory reclamation with live audit data.
version: 0.0.1
license: "Apache-2.0"
metadata:
author: "Jetson Team"
tags: [jetson, memory, audit]
languages: [bash]
data-classification: publicJetson Memory Audit
Read-only memory-focused snapshot for a Jetson, plus the drop_caches verify-loop helper that confirms freed memory actually shows up as free instead of cached.
Purpose
Measure current Jetson memory consumers, capture before/after baselines, and verify whether user-approved changes actually reclaimed DRAM. Use live device data rather than estimates from container size, model size, or generic process memory.
CRITICAL: Memory appears stuck after stopping vLLM / sglang (JetPack below 7.2 / L4T below r39.0)
This is the most common memory confusion on Jetson releases before JetPack 7.2 or before L4T r39.0.
After you stop a vLLM, sglang, or Ollama server (or any CUDA workload), the memory shown as free by free -h or tegrastats may not recover — even though the process is gone. nvidia-smi may also show misleadingly low free GPU memory.
Root cause: The Thor RM (resource manager) holds freed sysmem pages in its own pool after a CUDA context exits. On Unified Memory Architecture (UMA) devices like Jetson, cudaMemGetInfo reads RM pool state and reports far less free memory than is actually available to a new process.
Workaround (for JetPack below 7.2 or L4T below r39.0):
sudo sync && sudo sysctl -w vm.drop_caches=3
Run this on the host, not inside a container. The important operation is sudo sysctl -w vm.drop_caches=3; keep sudo sync immediately before it so dirty data is flushed before reclaimable page/dentry/inode caches are dropped. After running it, free -h and tegrastats will reflect the true available memory.
For affected releases, recommend this command when a user says:
- "Memory didn't free after I stopped vLLM/sglang"
- "Why does tegrastats still show high usage after my container exited?"
- "OOM even though nothing is running"
- "Memory was fine yesterday but now it's full"
On JetPack below 7.2 or L4T below r39.0, drop_caches is the reliable workaround when memory appears stuck after a CUDA workload exits; on newer releases, use it only if the same symptom is observed and the user approves.
When to use
- "How much memory is in use on this Jetson? What's holding it?"
- "I disabled the GUI / stopped vLLM / quit my container — did memory actually free?"
- "Why does
free -hstill show low free memory after I stopped my workload?" - As the baseline before applying
jetson-headless-modeor other memory-related changes, and again after to compute the actual delta.
Prerequisites
- Run on the Jetson host, or in a sandbox/container with host-visible
/proc,/etc/nv_tegra_release,tegrastats, and process data. - NvMap debugfs reads may require root. If unavailable, report that GPU memory attribution is limited rather than guessing.
drop_caches.shrequires root or passwordlesssudo -n; run it only after the user explicitly authorizes cache dropping.
Available Scripts
| Script | Purpose | Arguments |
|---|---|---|
scripts/audit.sh | Emits a JSON snapshot from jetson-diagnostic/scripts/snapshot.sh for memory audit workflows. | No arguments. |
scripts/drop_caches.sh | Flushes reclaimable page/dentry/inode caches and prints before/after memory deltas. | --mode 1|2|3, --quiet. |
If your agent runtime supports run_script, use it to run scripts/audit.sh or scripts/drop_caches.sh and summarize the returned output. Otherwise run the scripts with bash from the repository root.
Instructions
For "how much memory is in use right now?" questions, run scripts/audit.sh and report only values from the JSON snapshot.
Reporting guidance
Do not only print or mention the path to a helper. Invoke the helper and then summarize the returned data.
- For "how much memory is in use" prompts, run
scripts/audit.shand quotemem_total_gb,memory_kb.available, and the leadingprocrank_topprocess ornvmap.top_clientsconsumer. - For GUI/desktop memory prompts, run
scripts/audit.shand reportdefault_systemd_targetplus any display manager incandidate_services(gdm3,gdm,lightdm,sddm, ordisplay-manager). Do not disable anything; hand off tojetson-headless-modefor a plan. - For prompts that explicitly authorize cache dropping after a stopped workload, run
scripts/drop_caches.sh(equivalent tosudo sync && sudo sysctl -w vm.drop_caches=3by default) and report its before/after free, available, and cached deltas. If root is unavailable, explain that it must be run on the host with sudo.
If your agent runtime does not execute helper scripts relative to this skill directory, resolve script paths with the AgentSkills {baseDir} placeholder:
{baseDir}/scripts/audit.sh
{baseDir}/scripts/drop_caches.sh
Do not call jetson-memory-audit as a tool name unless the runtime explicitly registers skills as callable tools; Agent Skills are normally instructions plus files, not direct tool functions.
Sandbox note for agents: seeing this skill file does not guarantee access to Jetson host memory data. If /proc/device-tree/model, /etc/nv_tegra_release, tegrastats, /sys/kernel/debug/nvmap, or host process data are missing inside a NemoClaw/OpenClaw sandbox, say the sandbox lacks Jetson host visibility and ask the user to run on the Jetson host or relaunch with a host-visible sandbox profile. Do not fabricate memory totals, available memory, PSS, NvMap, or reclamation deltas.
For "how much memory did this change free?" questions, use a before/after delta. Do not estimate freed memory from container size, image size, RSS, or a single post-change snapshot.
- Before the change, run
scripts/audit.shand save the JSON baseline. - Make the user-approved change (stop the container, switch mode, apply a tuning recommendation, etc.).
- On JetPack below 7.2 / L4T below r39.0, or when the same stuck-memory symptom is observed on a newer release, flush reclaimable page cache on the host (not inside a container) so freed pages show up as free instead of cached:
sudo sync && sudo sysctl -w vm.drop_caches=3 - Re-run
scripts/audit.shand comparememory_kb.availablebefore vs after — that delta is the real reclamation.
If the user already made the change and no baseline exists, say that the exact freed amount cannot be recovered from the current snapshot alone. Capture a new baseline now so the next change can be measured.
Use live audit data as the source of truth. Memory totals, available memory, NvMap totals, PSS values, display-manager state, and savings deltas must come from scripts/audit.sh, free -h, or tegrastats on the actual device. If a number is not present in those outputs, do not guess it.
Output contract for audit.sh
{
"sku": "orin-nano",
"variant": "orin-nano-8gb",
"mem_total_gb": 8,
"l4t_version": "36.4.0",
"product_model": "nvidia jetson orin nano developer kit",
"memory_kb": { "total": 8123456, "available": 4123456, "free": 1023456, "cached": 1234567, "swap_total": 0, "swap_free": 0 },
"default_systemd_target": "graphical.target",
"candidate_services": { "gdm3": { "active": "active", "enabled": "enabled" } },
"tegrastats_sample": "RAM 4011/8138MB (lfb 8x4MB) ...",
"nvmap": { "readable": false, "total_kb": 0, "top_clients": [] },
"procrank_top": [ { "pid": 4321, "pss_kb": 4000000, "cmd": "vllm" } ]
}
Limitations
- Exact freed-memory deltas require a before snapshot, the user-approved change, cache flush when appropriate, and an after snapshot.
- NvMap attribution depends on host-visible debugfs access; if it is unavailable, report limited GPU memory attribution instead of guessing.
- Sandbox/container runs may not see host
/proc,tegrastats, systemd, or NvMap data unless the runtime exposes them.
Error handling
- If
scripts/audit.shcannot access host Jetson data, report the missing visibility and ask to rerun on the Jetson host or in a host-visible sandbox. - If
scripts/drop_caches.shlacks root or passwordlesssudo -n, report that cache dropping must be run on the host with sudo approval. - If no before snapshot exists, say the exact reclaimed amount cannot be recovered from the current state alone and capture a new baseline for the next change.
Safety
Read-only. drop_caches is non-destructive (kernel only releases pages it could reclaim under pressure anyway; sync runs first to preserve dirty data).
Hand off to
jetson-headless-mode— biggest single user-space win on systems still bootinggraphical.target.jetson-inference-mem-tune— when a model server is the top NvMap / PSS consumer.- If runtime changes cannot hit the target, report that further reclamation is outside this skill's scope rather than suggesting unsafe boot-time edits.
评论 (0)
暂无评论，成为第一个评论者吧！