NVIDIA Skill Finder
Finds the right NVIDIA agent skill for product, hardware, and workflow requests.
The skill explicitly forbids installing or modifying capabilities without approval, instructs users not to expose secrets, and requires a catalog check before recommendations. However, it relies on unpinned npx, live external catalogs, and network access without documented rollback, isolation, or finer-grained permission controls, so points are deducted.
Trigger rules, negative triggers, catalog fallbacks, and failure behavior are comparatively clear. However, the skill has no dedicated executable tests for its key paths, live catalog contents can drift, and diagnostics for abnormal input or network failure remain limited, so points are deducted.
The skill targets NVIDIA products, hardware, and workflows with detailed positive and negative semantic boundaries and taxonomy routing. It provides no Chinese-language guidance, and its core catalog lookup depends on GitHub, build.nvidia.com, or npx without addressing reachability from mainland-China networks, so points are deducted.
The documentation is readable and layered, with front matter, routing reference, installation guidance, limitations, security notes, ownership, and licensing. However, the recommended Examples section is missing, version information appears only in the skill card without an included changelog, and dependency and maintenance details are incomplete, so points are deducted.
The core task—discovering and recommending NVIDIA skills—is directly addressed through catalog checking, matching, and approval-before-installation guidance. The committed 17-task evaluation report indicates plausible benefit, but static review cannot confirm current-version behavior and results depend on a live catalog, so full marks are not justified.
The repository includes a committed 17-task evaluation summary with metrics, providing some auditable evidence. It lacks CI or a dedicated test suite tied to this skill's key paths and lacks independent reproduction records, so coverage and independence are limited.
- Core functionality depends on live external catalogs and npx; discovery may fail when those services are unreachable or change.
- The skill recommends installation commands without documenting version pinning, artifact verification, isolation, or recovery after failure.
- The committed evaluation report is not a substitute for independent reproduction of this revision; it used one attempt per task and no skill-specific CI tests are shown.
- Chinese-language behavior and mainland-China network reachability are not documented and should be addressed for the stated audience.
What does this skill do, and when should you use it?
NVIDIA Skill Finder is a discovery and routing skill for NVIDIA’s agent-skill catalog. It covers requests involving CUDA, GPUs, Jetson, NIM, NeMo, Omniverse, TensorRT, and other NVIDIA products and workflows. It checks the live catalog before recommending specific skills and can return up to three candidates with reasons, install commands, and an initial prompt. It recommends capabilities but never installs or changes them without explicit user approval.
Detects NVIDIA-related products, hardware, software, SDK, driver, setup, and workflow signals; checks whether a relevant skill is already installed or in context; queries the live catalog with npx skills add nvidia/skills --list or the specified NVIDIA catalog URLs; matches candidates using current names, descriptions, product groups, and skill cards; returns a skill name, fit rationale, installation command, and first useful prompt; and avoids inventing concrete skill names or install targets when catalog lookup fails.
- A Jetson developer needs help with driver, JetPack, L4T, or BSP setup and wants to find the current matching skill first.
- An engineer working with CUDA, cuDF, TensorRT, or GPU acceleration wants specialized NVIDIA guidance.
- A user building with NVIDIA NIM, NeMo, Omniverse, DeepStream, or Holoscan needs to discover relevant workflow skills.
- A team wants to extend an agent for NVIDIA tasks but does not yet know which catalog skill to install.
What are this skill's strengths and limitations?
- Broad detection coverage across NVIDIA products, hardware, SDKs, drivers, and catalog lanes.
- Uses the live catalog as the source of truth for current skill names, descriptions, and availability.
- Checks for skills already installed or present in context before recommending installation.
- Defines a clear approval boundary and fallback behavior when catalog lookup is unavailable.
- Reliable recommendations for specific current skills depend on access to the live catalog.
- It discovers and recommends skills rather than performing the underlying NVIDIA product work itself.
- The catalog changes continuously, so skills may later be renamed, consolidated, or removed.
- The supplied SKILL.md shows no dedicated scripts or standalone test suite.
How do you install this skill?
The README supports installing NVIDIA skills through the skills CLI:
npx skills add nvidia/skills
The CLI prompts for a skill and installation destination; cloning the repository or copying folders manually is not required. The supplied README does not list nvidia-skill-finder as a single-skill catalog entry, so no specific --skill slug should be inferred for it.
How do you use this skill?
Give the agent an NVIDIA-related request, for example: "Find a skill for Jetson driver installation." To browse the catalog directly, run:
npx skills add nvidia/skills --list
The skill becomes useful when the agent loads skills and encounters a relevant task. Installation should occur only after explicit user approval.