What does this skill do, and when should you use it?
This skill guides Compose-based deployments of NVIDIA’s Video Search and Summarization (VSS) blueprint. It routes requests to base, alerts, lvs, search, warehouse, or edge profiles and covers host preparation, model placement, credentials, Compose resolution, deployment, readiness checks, and troubleshooting. It can also tear down existing deployments, probe remote model endpoints, and remove dangling optional dependencies from resolved configuration. It does not deploy standalone microservices; hardware sizing and minimum GPU requirements belong to each profile’s reference.
It detects the video-search-and-summarization checkout, inspects GPU, Docker, and network context, and validates credentials and remote model endpoints. It copies .env to generated.env, applies deployment overrides, and runs docker compose config to produce resolved.yml. It runs normalize_resolved_yml.py to remove optional depends_on entries for filtered services, checks access to selected NGC images and model resources, starts the Compose stack, and verifies container counts, health endpoints, the Agent API, the UI, and LLM/VLM endpoints. For failures, it follows common troubleshooting and profile-specific debugging procedures.
- A VSS operator needs to deploy the base video search and summarization profile on a host with an NVIDIA GPU, Docker, and the NVIDIA Container Toolkit.
- An infrastructure engineer needs to deploy the search, lvs, alerts, or warehouse profile while matching model placement to available hardware.
- A team needs to deploy VSS on DGX Spark, AGX Thor, or IGX Thor using the documented edge recipes.
- An operator needs to verify that the full deployment path works rather than merely checking whether containers are running.
- An engineer needs to diagnose remote LLM/VLM endpoints, NGC entitlement failures, Compose variables, or service-readiness problems.
How do you install this skill?
- The workflow reaches teardown and data-volume clearing before explicit deployment confirmation; actual use should require explicit consent about data loss and verify recovery options.
- The supplied material omits the references directory, script implementations, and reproducible execution logs, so hardware matrices, credential probes, and readiness gates cannot be verified statically.
- Chinese interaction and mainland-China network reachability are not documented; NGC, Hugging Face, Docker registries, and remote endpoints may require additional network configuration.
- Shell / CLI
- Network access
- Local filesystem
DockerDocker ComposeNVIDIA GPU driverNVIDIA Container Toolkitnvidia-smiNode.js/npmuv
Install the specific skill with the skills CLI documented in the repository README:
npx skills add nvidia/skills --skill vss-deploy-profile --yes
For Codex-specific installation:
npx skills add nvidia/skills --skill vss-deploy-profile --agent codex --yes
The README does not provide a separate manual-copy procedure for this individual skill; the collection is installed through the CLI.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/vss-deploy-profile" ~/.claude/skills/
rm -rf "$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:
- verify the deploy
After installation, give a compatible agent a concrete request such as "deploy vss", "deploy search", "debug warehouse", or "verify the deploy". The skill then requires repository detection, hardware, credentials, and network checks before copying .env to generated.env, applying overrides, running docker compose --env-file generated.env config > resolved.yml, reviewing and normalizing the result, and starting services after confirmation. It does not silently choose remote models merely because endpoint variables already exist.
What are this skill's strengths and limitations?
- Covers the deployment lifecycle from host and credential checks through post-deployment readiness verification.
- Provides explicit routing for base, alerts, lvs, search, warehouse, and edge profiles.
- Includes remote-model probing, NGC artifact entitlement checks, and dangling depends_on normalization.
- Separates the checked-in .env defaults from the per-deployment generated.env working copy.
- Requires an NVIDIA GPU, driver, Docker, the NVIDIA Container Toolkit, and the VSS source repository.
- Deployment may require access to specific NGC artifacts or valid remote model endpoints; missing credentials or entitlement block progress.
- Scope is limited to Compose-based VSS profiles and excludes standalone microservice deployment.
- The supplied material does not show a test suite or actual results across hardware platforms; detailed sizing depends on profile references not included here.
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 |
|---|---|---|---|---|
| VSS Profile Deployment Assistant this page ✓ NVIDIA · Official | 42 · Blocked | ★ 3.8k | today | Apache-2.0 |
| VSS Video Summarizer ✓ NVIDIA · Official | 55 · Use with care | ★ 3.8k | today | Apache-2.0 |
| VSS Standalone Video Analytics API ✓ NVIDIA · Official | 54 · Use with care | ★ 3.8k | today | Apache-2.0 |
| RT-VLM Dense Captioning Deployment ✓ NVIDIA · Official | 45 · Not recommended | ★ 3.8k | today | Apache-2.0 |
| VSS Multi-Camera 3D Detection and Tracking ✓ NVIDIA · Official | 56 · Use with care | ★ 3.8k | today | Apache-2.0 |
How did FollowSkills review this skill?
The skill explicitly provides credential gates, remote-endpoint intent checks, separation of generated.env from the checked-in .env, token non-disclosure, and a pre-deployment review prompt. However, Step 0 requires tearing down an existing deployment and clearing data volumes before the confirmation step, without explicit consent, backup, or recovery evidence in this file. That destructive default is a major deduction and triggers blocking.
The evidence shows a detailed sequence with hardware checks, credential probes, Compose dry-run, configuration validation, container-count gates, readiness checks, and defined stop conditions. However, the referenced files, script implementations, dependencies, and real execution results are not supplied, so static review cannot verify reproducibility or failure behavior. The score stays below the static cap and is reduced for that uncertainty.
The skill identifies its audience and scenarios, routes five main profiles, handles edge hardware, and excludes standalone microservices. However, trigger coverage is incomplete, input/output contracts depend on external references, Chinese-language support is not stated, and mainland-China network compatibility is not evidenced. Points are deducted for these boundary and environment uncertainties.
The evidence includes structured metadata, an Apache-2.0 license declaration, version, product-repository link, routing table, staged workflow, limitations, and a benchmark report. It lacks an author field and Examples section, gives no skill-specific changelog or explicit maintenance owner, and places many installation and troubleshooting details in external references. Points are deducted accordingly.
The skill covers deployment, verification, debugging, and teardown, and defines directly actionable outputs such as resolved.yml and health checks. But no independently verifiable representative output is supplied, and the external profile references are absent from the material, so direct completeness and usability across scenarios remain uncertain. The score is conservatively kept within the static cap.
The evidence includes a pinned revision, product-repository link, committed evaluation report, and multiple evaluation cases, making some claims traceable. It does not include reproducible logs, implemented CI coverage, or independent corroboration, and the core referenced materials are missing. Only limited static evidence is therefore credited.
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 →