NVIDIA-Omniverse

Map unfamiliar or multi-repository codebases before refactoring, integration, or feature work. Use when Codex needs to analyze submodules, ownership boundaries, APIs, duplicate implementations, build/test surfaces, risk areas, and a practical refactor sequence before changing code.

Coordinate multiple AI agents or subagents on software work without duplicating effort or causing conflicts. Use when Codex is explicitly asked to delegate exploration, implementation, review, testing, integration, or parallel analysis across a codebase.

Maintain C ABIs and language bindings as exact, testable contracts. Use when evolving public headers, ctypes/cffi/pybind/Rust/JS bindings, shared libraries, struct layouts, GPU/native APIs, or any cross-language boundary where drift can cause crashes or silent corruption.

Debug rendering, viewer, shader, material, camera, lighting, and GPU integration bugs with reproducible visual evidence. Use when scenes render blank, only appear after camera motion, show incorrect shading, diverge across backends, crash in GUI mode, or need pixel/screenshot verification.

Refactor disjoint or duplicated projects into clear layered architecture with canonical contracts and thin adapters. Use when viewers, renderers, bindings, data libraries, plugins, or services overlap and need maintainable ownership boundaries without losing behavior.

Identify and remove obsolete legacy, fallback, staging, compatibility, duplicate, generated, and reference code before creating clean initial commits or new project history. Use when a repo is being reset, consolidated, or prepared for a cleaner public history.

Load USD scenes through NanoUSD and convert composed USD prims into renderer-ready geometry, material, light, camera, transform, instancing, texture, and diagnostic representations. Use when implementing, debugging, or validating NanoUSD-based scene ingestion, renderer scene structs, material/light extraction, texture path resolution, or loader parity against USD/OpenUSD semantics.

Design, debug, or validate physically based renderer behavior against a reference renderer. Use when Codex works on PBR materials, USD light semantics, shadows, transparency, GI, emissive sampling, tone mapping, AOVs, stochastic sampling, path tracing, ray tracing, or Omniverse Kit/Hydra/OVRTX visual parity.

High-level overview of a typical ovrtx application lifecycle. Use when user asks how to structure an ovrtx program, what the main steps are, or how the pieces fit together.

Asynchronous operation patterns including polling, timeouts, and non-blocking workflows. Use when user asks about async rendering, non-blocking operations, polling, timeouts, or parallel rendering.

Creating persistent attribute bindings for efficient repeated writes to the same prims and attribute. Use when user asks about persistent bindings, repeated writes, efficient animation loops, bind_attribute, or updating transforms every frame with caller-owned tensors. Use mapping-attributes when the hot path needs zero-copy direct writes into ovrtx buffers.

Binding materials to prims at runtime. Use when user asks to assign a material, change a material, set material binding, or swap materials on a prim.

Available camera render outputs for Real-Time Path-Tracing (RT2) mode. Use when user asks what AOVs/render vars are available, what format or dtype an output has, or how to read a specific output like depth, normals, albedo, or distance.

Cloning USD subtrees to create copies at new paths. Use when user asks to clone, duplicate, copy a prim, or create instances of existing geometry.

Authoring and configuring OmniLidar sensor prims and lidar PointCloud render outputs. Use when user asks to create a lidar scene, configure an OmniLidar prim, choose lidar output frame/coordinate behavior, or request lidar PointCloud channels.

Authoring and configuring OmniRadar sensor prims and radar PointCloud render outputs. Use when user asks to create a radar scene, configure an OmniRadar prim, choose radar output frame/coordinate behavior, configure radar scan outputs, or request radar PointCloud channels.

Orchestrate an end-to-end NVIDIA self-managed streaming stack across AWS or Azure. Use when planning, generating, sequencing, deploying, or validating the full path from CSP base infrastructure through NVCF self-managed control plane, NVCF auxiliary services, caches, storage APIs, and post-deploy readiness gates. Trigger when the user asks for orchestration, full-stack setup, bring-up order, runbook, dependency sequencing, AWS/Azure streaming deployment, csp-infra plus NVCF, caches, or storage API coordination. Do not use for a single component task when the user explicitly asks only for one child skill.

Deploy DDCS as a shader cache for Kit/write-render-worker on GKE. Produces Kubernetes manifests, the Helm values override for write-render-worker (drivercachewrapper + UJITSO), and optionally runs a proof-of-life test. Use when setting up DDCS for MEGA shader caching (replacing GX cache).

Deploy NVIDIA Omniverse Content Cache (OVCC) alongside a MEGA simulation stack. Covers evaluation, sizing (NGINX queue-depth model), configuration for GCS/S3/Azure backends, Helm install from NGC (chart ovcontentcache v4.0.4), TLS setup, in-place upgrade from UCC 3.0.3, and wiring to write-render-worker and storage-service. MEGA does not use Nucleus; always disable it. Use when deploying OVCC for MEGA — whether exploring fit, sizing replicas, or ready to install.

Use for installing, debugging, or modifying NVCF GXCache across self-managed Kubernetes environments. Trigger on explicit GXCache, helm-gxcache, gxcache-init, gxcache-webhook, distributed shader cache, NVCA cache feature gates, or GXCache provider adaptation for Kubernetes clusters.

Generate the AWS network and IAM foundation for a KAS/NVCF EKS deployment: VPC, public/private subnets, route tables, Internet/NAT gateways, and base IAM/tagging inputs. Use as Stage 01 of aws-infra before eks-cluster.

Orchestrate the AWS KAS/NVCF EKS foundation by delegating to the component AWS infrastructure skills: aws-tf-conventions, aws-iam-vpc, eks-cluster, eks-nodes, eks-storage, ecr, eks-gpu-operator, eks-validate, and optionally ecr-mirror. Use when setting up, reviewing, or tearing down the full AWS foundation end-to-end; not for runtime installation, functions, or stream validation.

Project-specific Terraform rules for the AWS KAS/NVCF EKS foundation. Use when writing or reviewing AWS Terraform for VPC, EKS, node groups, IAM, ECR, state, tags, provider configuration, or before terraform apply.

Mirror every KAS/NVCF manifest image and optional Helm OCI chart from NGC into AWS ECR using generated regsync config.

Start one NVIDIA Content Agents REST service locally with Docker Compose. Use when the user asks for /quickstart, local POC setup, or single-service Material, Physics, and Texture startup.

Manage Brev instances safely from the Brev CLI. Use when a workflow needs to list, create, access, copy to, execute on, port-forward, stop, or delete Brev instances for Content Agents dependency endpoints.

Deploy the full Content Agents package with a provider-neutral, endpoint-driven, multi-instance strategy. Use when the user asks to deploy all content agents, deploy material/physics/texture together, configure shared OVRTX rendering, choose VLM/LLM/image-gen/embedding endpoints, test the full package, or use Brev as an optional self-hosted dependency provider.

Deploy or smoke-test a Brev-hosted NVIDIA embedding NIM endpoint for Content Agents. Use when the user asks to deploy embeddings on Brev, host the llama-nemotron-embed-vl model, enable Material Agent prim clustering embeddings, test embedding sidecars, or create a Brev embedding endpoint for the collection deployment.

Deploy or smoke-test a Brev-hosted FLUX image-generation NIM endpoint for Content Agents. Use when the user asks to deploy image generation on Brev, host FLUX.2 Klein for Texture Agent, enable image-gen sidecars, test image-gen endpoints, or create a Brev image-gen endpoint for the collection deployment.

Deploy or smoke-test the Material Agent with Brev-hosted dependency endpoints. Use when the user asks to test material agent on Brev, deploy material agent with Brev, use RTX/L40S for rendering and A100/H100-grade GPU for VLM, run a Brev hybrid material-agent test, or recreate the Brev material-agent deployment. This workflow keeps the main pipeline local and uses Brev port-forwards to reach OVRTX and Qwen-family VLM endpoints.

Deploy the material-agent-service locally using Docker Compose with OVRTX rendering. Use when user wants to run material agent with docker, docker compose, set up local deployment, run the service locally with GPU rendering, start material agent containers, or configure VLM provider for docker deployment. Trigger phrases include "docker compose", "docker deploy", "run locally with docker", "start material agent docker", "docker up", "local deployment", "compose up".

Deploy just the OVRTX rendering API standalone via Docker Compose so one or more agent CLIs / services can point at it as a shared rendering endpoint. Use when user wants to run OVRTX standalone, deploy a shared rendering service, separate the rendering endpoint from an agent service, expose OVRTX to multiple callers, or set RENDER_ENDPOINT to a dedicated render host. Trigger phrases include "deploy ovrtx", "run ovrtx standalone", "shared rendering service", "ovrtx rendering docker", "render endpoint deploy", "ovrtx docker compose", "standalone rendering".

Deploy and develop against NVIDIA Omniverse Storage APIs on any Kubernetes cluster (MicroK8s, EKS, AKS, GKE, bare metal). Two storage adapters: (1) Example Storage Adapter — Python filesystem reference implementation + Discovery for learning and custom adapter development; (2) S3/Azure Production Storage Adapter — NVIDIA pre-built S3/Azure adapter + Discovery as minimum, expandable with Event Aggregation, Event Consumer, RabbitMQ, Envoy Auth Extension, Storage Navigator, and Contour ingress. Can deploy on the developer's behalf (generating scripts + .env) or guide manual deployment. Custom adapter development covers three APIs: Storage, Notifications, and Permissions. Use when a developer asks about deploying, configuring, validating, troubleshooting, or building any component of the Omniverse Storage APIs stack.

Use when adding or reviewing an authz plugin for ovstorage-broker or ovstorage-rest - covers the AuthzPlugin trait, the 21 operations, the policy-epoch model, and the cdylib cancellation contract.

Use when adding or reviewing a storage backend plugin for ovstorage.

Use when Rust changes affect ovstorage C or C++ headers and checked-in generated headers must be refreshed.

Use when starting an ovstorage-broker locally for development or integration testing - covers a minimal TOML, the cdylib plugin path, and how to hit the listener with broker-client or curl.

Use when checking the Storage API wire contract that ovstorage-plugin-services-client depends on.

Use before opening or merging an ovstorage PR to run the project verification gate and inspect risky diffs.

Use when authoring or rotating the broker's authz policy - covers the [authz] TOML block, plugin selection, policy-epoch advance/invalidate, and grace-window freshness mode.

Reference for Usd Optimize's validator infrastructure (registration, CLI, logging, REQUIRES_MESH cache). Do not use for ad-hoc validation runs — use run-validators instead.

Build Usd Optimize from source via repo.sh (Linux) or repo.bat (Windows). Use when compiling the repo, switching configs, or selecting a USD flavor.

Diff two USD stages by prim/mesh/vertex/material count, file size, and validator summary. Use for before/after optimization comparisons.

Create a USD proxy mesh sibling. Use to generate decimated, bbox, or LOD stand-ins, with optional render/proxy purpose variant set.

Triage a failing Usd Optimize operation. Use when an op errors, silently no-ops, or returns unexpected output.

Collapse duplicate prim hierarchies into instanceable internal references. Use when deduplicating subtrees or folding repeated prims into prototypes.

Read saved validator artifacts and present a tier-classified report with per-rule prim lists and fix recommendations. Use when interpreting a saved run.

Scaffold a new Usd Optimize operation plugin (C++ source, premake, test, guide, optional validator). Use to add a new op or plugin.

Wiring connection / message / input callbacks for WebRTC, native, SHM, and CUDASHM streams. Use when user asks about receiving input events, handling messages, detecting client connect/disconnect, or sending messages back.

Writing a CUDASHM reader/consumer that attaches to an ovstream CUDASHM producer over CUDA IPC. Use when user asks about consuming GPU-resident frames, reading from a CUDA IPC ring, writing a simulation/post-processing kernel that ingests rendered output, or running a consumer in a separate container.

Writing a native (StreamSDK) reader/consumer that attaches to an ovstream NATIVE server over the network. Use when user asks about consuming a native stream, writing a StreamSDK client viewer, decoding frames client-side via NvStreamingMedia, or streaming between two different machines.

Setting up a new CMake C/C++ project that uses ovstream. Use when user asks to create a new C project, set up CMake with ovstream, scaffold a C++ streaming app, or configure build dependencies.

Choosing between WebRTC, RTSP, the native protocol, SHM, and CUDASHM. Use when user asks which transport to use, the trade-offs, what each protocol supports, or how to view a stream.

Writing an SHM reader/consumer that attaches to an ovstream SHM producer. Use when user asks about consuming SHM frames, reading from shared memory, writing an OpenCV / Electron / custom viewer, or multi-reader patterns.

High-level overview of a typical ovstream application lifecycle. Use when user asks how to structure an ovstream program, what the main steps are, or how the pieces fit together.

CUDA buffer allocation, pitch alignment, BGRA channel order, and ctypes-bound cudart usage. Use when user asks about CUDA interop, pitched allocations, frame layout, or zero-copy GPU integration.

Stand up a USD Search deployment the user controls. Branches on local docker-compose vs. helm on Kubernetes. Local branch covers the full quickstart compose stack with GPU/VLM/storage-backend questionnaires. Helm branch covers the usdsearch chart at helm/usdsearch. Use when: "deploy usdsearch", "run the stack myself", "helm install usdsearch", "start the stack", "run usdsearch locally", "host my own".

Run the locally-built Explorer UI against a REMOTE, fully-hosted USD Search instance by proxying every API call through the local gateway. No local deepsearch-api / OpenSearch / SigLIP2 / Asset Graph Service — only the Explorer + gateway run locally, so search, thumbnails, dependency graphs, and downloads all work exactly as on the hosted instance (which already has a populated index + AGS + storage). Use when: you want to test/demo the local Explorer with WORKING dependency graphs and downloads but don't want to (or can't) build a local index + Asset Graph Service. Default target is the public staging instance (api.staging.deepsearch-horde.nvidia.com).

Deep-inspect a single 3D asset by URL. Fetches multi-view thumbnails, per-plugin indexing status, USD scene summary (prim counts, geometry, scale, up-axis, default prim), prim hierarchy, forward + inverse dependency graphs, and optionally runs VLM relevance validation against a description. Use when the user asks to inspect / deep-dive / describe / learn about / report on a specific asset, or asks "what's in this USD", "tell me about this asset", "where is this asset used", "what does this asset reference", "is this asset indexed".

Get a first-time user to a real 3D asset in seconds. Asks one question — "where do you want to search?" — with three backend lanes (NVIDIA-hosted default, your own URL, or a local deployment), then hands off to /search for the actual query. The journey is the SVG at docs/quickstart-user-journey.svg: Stage 1 (entry) + Stage 2 (backend pick), then /search drives Stages 3-5. Use when: "quickstart", "try usd search", "show me what this does", "getting started", "what is usd search", "tour", "first time".

Spatial and scene-graph queries inside a USD scene via the Asset Graph Service. Answers: what's near a point or prim (radius), what's inside a bounding box, what prims of a given type or property exist in this scene, what does this scene reference, where is this asset used, what's the prim hierarchy. Optionally chains to /search_hybrid with image_similarity_search to find assets similar to a prim's source asset. Use when the user has a scene URL plus a spatial, structural, or dependency question.

Search USD Search for 3D assets. One skill, three input modes: text query, reference image, or both combined. Hits POST /search_hybrid + GET /images. Iterates on intermediate matches for visual-similarity refinement when the user wants "more like this". Use when: "search assets", "find a 3D model", "find a USD", "find similar", "more like this", "visually similar".

Discover which USD properties a USD Search deployment's corpus actually carries, and turn that into local config: hits GET /search/stats/usd_properties, then builds a ranked usd_property_catalog.yaml (the grounding source for the built-in LLM query parser) plus ready-to-use search_fields.generated.yaml filter stanzas. Has a P0 "gap audit" mode that reports which target properties are present/absent and pulls sample assets for parsing validation. Use when: "what USD properties can I filter on", "discover usd properties", "build a property catalog", "which metadata is in the index", "audit P0 metadata", "ground the query parser on real properties".

Retrieves full NVCF function and deployment information for a given function ID (and optionally a function version ID) via the NVCF and NGC APIs, including status, container image, inference and health endpoints, ports, and deployment specifications. When only a function ID is provided, lists the available versions and asks the user to confirm or select one — even when only one version exists. Use when the user asks to inspect, verify, or debug an NVCF function or function version, or needs details such as ACTIVE status, container image, inference port, or cluster/GPU settings.

Retrieves full details for a published streaming application on the Omniverse portal by locating it via app_id or NVCF function_id and function_version_id, authenticating like publish-streaming-app, and calling GET /api/apps/{app_id}. Returns portal metadata, NVCF runtime status, and deployment details when available. Use when the user asks to inspect, verify, or look up a portal streaming app or its NVCF linkage.

Triages Omniverse Kit streaming failures on NVCF and the Web Streaming Portal by matching symptoms to issue guides, verifying portal and NVCF state via check-streaming-app and check-nvcf-function, and summarizing fixes. When the root cause stays unclear, collects every piece of evidence (portal registration, NVCF deployment details, function logs, client error text) into a single diagnostic bundle folder for handoff or escalation. Use when a stream fails to start, the portal shows connection errors, NVCF deployment stalls, or the user asks about livestream, WebRTC, NVCF, Kit streaming, or OVC/DGXC.

Publish or update a streaming application on the Omniverse portal by collecting app metadata, authenticating against the portal's identity setup, and calling `PUT /api/apps/{app_id}`. Use when Codex needs to add, publish, register, or update a streaming function or application on the portal and the user can provide an NVCF `function_id` and `function_version_id`.

Remove a published streaming application from the Omniverse portal by locating the app via `app_id` or via `function_id` plus `function_version_id`, authenticating against the portal's identity setup, previewing the matched app, and calling `DELETE /api/apps/{app_id}` only after explicit user approval. Use when Codex needs to unpublish, delete, or remove a streaming application from the portal.

Capture clean render-only images and movies from Kit-CAE visualizations. Produces output without application UI - just the rendered scene. Supports high-resolution PNG/EXR screenshots, MP4 video with NVENC hardware encoding, and animated camera orbits. Triggers on requests for "capture image", "render screenshot", "export movie", "high-res render", "save visualization", "record animation", "camera pan", or any request to produce clean output from a Kit-CAE scene.

Shared foundation for Kit-CAE agent skills. Contains import APIs, visualization commands, stage discovery, format references, and utility scripts used by cae-visualization and cae-capture. Not invoked directly by users - loaded as a dependency by other CAE skills.

Run Kit-CAE as a WebRTC-streamed application that remote clients can view and control via data-channel messages. Provides a streaming-enabled .kit dep set, the wire-protocol reference, the messaging-API reference, a small reusable Python helper (`OmniverseAPI`), and a runnable reference handler (`load_scene` + scenes registry) that demonstrates the request/response pattern end-to-end. Triggers on requests involving "stream Kit-CAE", "WebRTC", "remote viewer", "browser viewport", "application streaming", "data-channel messages", or "remote control of Kit-CAE".

Visualize and analyze simulation data using NVIDIA Kit-CAE (Omniverse). Supports CGNS, VTK (.vti/.vtu/.vts/.vtp/.vtk), EnSight Gold (.case/.encas), OpenFOAM (.foam), NumPy (.npz/.npy), EDEM (.dem), and custom formats. Covers volume rendering, faces, slices, streamlines, glyphs, points, flow animation, bounding boxes, field statistics, multi-domain composition, and time-varying animation. Triggers on any request involving simulation post-processing, CAE visualization, scientific data rendering, or field data analysis.

Use when creating, modifying, inspecting, or reasoning about ovui UI from low-level atomic ovui widgets, including layout, Pixel/Percent/Fraction sizing, model-view widgets, TreeView/stage hierarchy trees, inputs, menus, windows, docking, viewport-like ovrtx panels, styling, or strict UI validation. Use when an agent must rebuild Stage-Browser-like, Property-Inspector-like, or Viewport-like UI from primitives without reusing high-level composite widgets. Do not use for ovwidgets composite-widget application work; use the omniverse-ui-widgets skill instead.

Use this skill when the user asks to inspect, screenshot, automate, click, type into, or drive a running ovui/ovwidgets application through the ovui-inspect FastAPI inspector. The skill provides an importable `ovuiinspect` module, an `ovui-inspect` CLI, and a screenshot-first mouse/keyboard workflow for controlling ovui apps launched with PYTHONPATH.

Use this skill for ovui and ovwidgets styling work: palettes, colour / float / URL constants, theme shades, ui.style.default, style dictionaries, selector grammar, widget name variants, style_type_name_override values, naming conventions, style hierarchy, the centralised style module, and styling reviews. This skill is the authoritative target style guide.

Use this skill when a user asks to create or validate a disposable standalone ovui/ovwidgets application from the existing widget set. The application source should live outside the repo unless the user explicitly asks to productize it. The skill gives exact imports, runtime environment, ovrtx proof requirements, widget wiring, menu/file-open wiring, and screenshot-first QA expectations. Do not use this skill for raw `omni.ui` atomic widget construction; use the omniverse-ui-apis skill instead.

Set up Python NVIDIA USD Validation plugins: entry points, custom rules, uv/pip install, first CLI run. Do NOT use for requirements.

Use when setting up a local usd-validation-nvidia venv for builds, tests, or nvidia_usd_validate. Do NOT use for plugin scaffolding.

Create USD Validation features with RequirementRef and feature dependencies. Do NOT use for authoring new requirements.

Author and validate USD Validation requirements: Markdown specs, codegen, rule mapping, --requirement CLI. Do NOT use for basic plugins.