Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/nvidia/nvcf/add-featurenpx skills add NVIDIA/nvcf --skill add-featuregit clone --depth 1 https://github.com/NVIDIA/nvcfWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00093 | $0.01561 |
| Opus 5 | $0.00046 | $0.00781 |
| Sonnet 5 | $0.00019 | $0.00312 |
| Haiku 4.5 | $0.00009 | $0.00156 |
Grade A, and why
add-feature scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 135 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Adding a UI Feature
For all component selection, layout decisions, and styling — invoke the kaizen-ui skill.
Gotchas
- This project uses code-based routing, not file-based. Do not use
createFileRouteor the TanStack Router Vite plugin — usecreateRouteandcreateLazyRoute. - View components are named by view type (e.g.
WidgetsList,WidgetDetail), never with aPagesuffix. router.tsxis pure composition — it only imports and wires route trees. Do not put route definitions or loaders inrouter.tsx.rootRoute.tsxis separate fromrouter.tsxto avoid circular imports. Layout routes that span multiple features live atsrc/root, not in a feature folder.- Generated hooks, schemas, and mocks live in
ui/src/generated/— never edit these files. If the endpoint you need doesn't have generated hooks, update the appropriate spec inspec/(control-plane-openapi.yaml,nvcf-openapi.yaml, orsis-openapi.yaml) then runtask generate. - All source files need Apache 2.0 license headers. Run
task license:addbefore committing.
When to Create a New Feature Folder
- New domain (new data, new concepts, new API surface) → new feature folder
- Composition of existing domains (e.g. a home page assembling widgets) → new thin feature folder
- Extension of an existing domain (e.g. a "Create Function" page) → new route/component in the existing feature folder
Feature Folder Structure
A feature is a domain boundary, not a route. Each feature is self-contained under ui/src/features/<name>/:
features/<name>/
├── components/ # Smaller building blocks (badges, cards)
├── hooks/ # Feature-scoped hooks
├── types/ # Feature-scoped types
├── routes.tsx # Route definitions, loaders, search validation (optional)
├── <View>.tsx # Route-level view components (PascalCase, named by view type)
└── utils.ts # Feature-scoped helpers (when needed)
- Not every subfolder is required — add
components/,hooks/,types/only when needed - A feature may not have routes at all — it may only export hooks and components for other features
- Helpers go inline in the component file by default. Extract to a feature-root file only when shared across multiple files in the same feature
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 135 lines · 93 tokens per session scan A 81eb978401d9
add-feature is a skill published in the GitHub repository NVIDIA/nvcf (202 stars, last pushed 2d ago), licensed Apache-2.0. It adds 93 tokens to every session and 1,561 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
vmware-privateai
Use this skill whenever the user needs the GPU / AI-infrastructure layer of VMware Private AI Foundation with NVIDIA (PAIF-N) on vSphere 9.x / VCF 9.1: inventory GPU hosts and physical GPU devices, see which VMs consume a vGPU and the profile each holds, read real-time GPU utilization, list the vGPU and DirectPath…
kubernetes-mesh-provisioner
Kubernetes Mesh Provisioner atomic skill. Stands up an RKE2 cluster (server + agents) with Cilium CNI and the NVIDIA GPU device plugin, the Kubernetes parallel of swarm-mesh-provisioner. Idempotent — re-runnable.
accelerated-computing-cudf
Official NVIDIA-authored guidance for NVIDIA cuDF GPU DataFrames, pandas acceleration, dask-cuDF, ETL, joins, groupby, CSV/Parquet I/O, nullable semantics, and multi-GPU DataFrame workloads.
llm-torch-profiler-analysis
Unified LLM torch-profiler triage skill for sglang, vllm, TensorRT-LLM, and TokenSpeed. Use it to inspect an existing trace.json(.gz) or profile directory, or to drive live profiling against a running server when supported and return one three-table report with kernel, overlap-opportunity, and fuse-pattern tables.
cookbook-review-pr
Review a pull request against the SGLang Cookbook (docs/, Mintlify) contribution checklist — the config-driven format (per-model config + benchmarks JSX consumed by the shared deployment.jsx / playground.jsx engines). Run with /cookbook-review-pr .
speculative-naming
Naming conventions for SGLang speculative decoding identifiers. Use when adding, renaming, or reviewing identifiers in speculative decoding code — anything under python/sglang/srt/speculative/, related attention backends, scheduler accumulators, IPC fields, observability metrics, or CLI flags.