What does this skill do, and when should you use it?
This skill launches AutoMagicCalib’s microservice and web UI from NVIDIA NGC release images. It checks Docker permissions, resolves or requests an AMC repository checkout, configures the Compose environment, and verifies access to the required NGC images. It can optionally download the approximately 4.7GB VGGT model from Hugging Face. After launch, it checks container status, the microservice readiness endpoint, and the UI HTTP response. Users need Docker, an NVIDIA GPU container runtime, an AMC checkout, and NGC access.
Runs docker ps to verify passwordless Docker access; searches for an existing auto-magic-calib checkout and requests confirmation before cloning one; creates or uses a Python virtual environment and installs huggingface_hub; logs in to nvcr.io with an NGC API key; checks for and optionally downloads the VGGT model; finds free ports in ranges 8000-8009 and 5000-5009; writes compose/.env and a backup ignore rule; assigns UID/GID 1000 ownership to projects/ and models/; checks image access using the resolved Compose configuration; runs docker compose up -d; polls /v1/ready; checks the Web UI with curl; and prints the microservice and UI URLs.
- A DeepStream user needs to start AutoMagicCalib on a host with an NVIDIA GPU.
- An engineer has an auto-magic-calib checkout or DeepStream submodule and wants Compose ports, paths, and host networking configured automatically.
- A deployment operator wants to verify NGC API-key access to every required release image before pulling layers.
- A calibration workflow requires VGGT and the user wants to download and verify the model before starting services.
- An operator needs readiness, UI availability, container status, and log-oriented troubleshooting for AutoMagicCalib.
How do you install this skill?
- docker login, HuggingFace downloads, and Docker Compose create external network, image-pull, and local-service side effects; confirm credential scope before use.
- Do not blindly execute cat "$ENV_FILE" because it may print existing configuration values or secrets.
- Recursive sudo chown changes ownership under projects and models, and the document lacks a complete automatic rollback procedure; confirm exact paths and preserve backups first.
- Connectivity to GitHub, NGC, and HuggingFace and license-acceptance state were not verified statically; mainland-China network conditions may block core workflow steps.
- Shell / CLI
- Network access
- Local filesystem
DockerDocker ComposeNVIDIA Docker RuntimeGitPython 3 with venv supportHugging Face CLI (only for optional VGGT download)NGC account and API keycurllsof
Use the repository README’s CLI installation command: npx skills add nvidia/skills --skill amc-setup-calibration-stack --yes. The command installs the skill to the destination selected by the CLI; the README does not specify another client-specific installation directory.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/amc-setup-calibration-stack" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- deploy auto calibration,
- set up auto-magic-calib.
With the skill loaded, prompt the Agent with “deploy auto calibration,” “launch AMC,” “start MS+UI,” or “set up auto-magic-calib.” The Agent will check the Docker, repository, and credential prerequisites, then ask for an NGC API key. For VGGT, the user must also accept the Hugging Face model license and provide a Hugging Face read token. If no usable AMC checkout exists, the Agent must request confirmation before cloning into ~/auto-magic-calib.
What are this skill's strengths and limitations?
- Covers the startup path from repository resolution and NGC login through Compose configuration and readiness checks.
- Reads image definitions from the active Compose configuration instead of hardcoding release tags.
- Includes preflight image-access checks, microservice readiness polling, and UI HTTP verification.
- Addresses Docker permissions, directory ownership, port conflicts, VGGT download, and GPU runtime issues.
- Requires Docker, the NVIDIA Docker Runtime, an NGC account, and an available AMC checkout.
- Requires confirmed sudo chown operations that recursively assign UID/GID 1000 ownership to projects/ and models/.
- The optional VGGT download is approximately 4.7GB and requires a Hugging Face account, license acceptance, and read token.
- The source provides no automated test suite or validated platform matrix.
- The skill launches the calibration stack but does not perform sample, MP4, or RTSP calibration itself.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| AutoMagicCalib Calibration Stack Launcher this page ✓ NVIDIA · Official | 54 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| VSS Multi-Camera 3D Detection and Tracking ✓ NVIDIA · Official | 56 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| VSS Standalone Behavior Analytics ✓ NVIDIA · Official | 51 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| VSS Video Auto-Calibration ✓ NVIDIA · Official | 51 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| VSS Profile Deployment Assistant ✓ NVIDIA · Official | 42 · Blocked | ★ 3.6k | 13d ago | Apache-2.0 |
How did FollowSkills review this skill?
The document requires --password-stdin and HF_TOKEN handling, confirmation before recursive chown, and explicit consent before cloning; however, it invokes sudo chown, Docker, NGC, and HuggingFace, may expose existing values through cat "$ENV_FILE", and lacks complete rollback and data-flow disclosure, so 8 points are deducted.
Prerequisites, sequencing, readiness checks, and failure messages are fairly detailed; however, the workflow assumes lsof, curl, hostname, Docker Compose, and a specific repository layout, with environment-sensitive port and venv detection. Key paths were not reproduced during this static review, so the score is limited to 8.
Trigger phrases, audience, dependencies, success criteria, optional VGGT behavior, and related skills are clear; however, non-Linux and no-GPU boundaries are incomplete, Chinese-language support is not described, and core cloning, registry, and model-download flows depend on GitHub, NGC, and HuggingFace, so 5 points are deducted.
The material is well structured and includes prerequisites, procedures, troubleshooting, shutdown, owner, version, and Apache-2.0 licensing; however, it lacks the recommended Examples section, a changelog, explicit maintenance responsibility/update path, and a clear explanation of the repository-wide dual-license relationship, so 6 points are deducted.
The command sequence theoretically covers checkout resolution, authentication, optional model download, Compose configuration, startup, and readiness verification; however, representative successful output is not statically verified and environment, credential, and permission operations may still require manual correction. The score is 6 under the static ceiling, with 1 point deducted for uncertainty.
Concrete commands, success criteria, evaluation tasks, and a repository benchmark report provide some auditability; however, the benchmark claims were not independently reproduced here, and there is no committed test suite covering the deployment key paths or corroborating evidence from multiple independent sources, so 6 points are deducted.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →