nvidia-omniverse-blueprints/digital-twins-for-fluid-simulation/rtdt-customize
Claude Code and agent skill by nvidia-omniverse-blueprints/digital-twins-for-fluid-simulation
- community
Checking that the SKILL.md file is still in the repository.
Customize an RTDT workflow, AppState schema, USD stage, frontend control, or Kit-CAE 3.0 geometry/result wiring. Use when adding or changing controls, variants, bindings, output fields, visualization selections, or bring-your-own geometry. Read rtdt-core first; use rtdt-aero as the specialization-specific companion for the shipped wind tunnel.
Install
npx vskill@latest install nvidia-omniverse-blueprints/digital-twins-for-fluid-simulation/rtdt-customizeSKILL.md
RTDT Customize
Read rtdt-core before using this skill. Most workflow changes span four surfaces: typed AppState schema, workflow manifest, USD stage, and the hand-built Trame frontend. Changing only one surface is usually incomplete.
Classify the change
- Backend-only control: schema + manifest + binding, used without a new shipped UI widget.
- User-visible control: backend wiring plus the matching Trame widget and bridged state key.
- Visual-only geometry: cosmetic USD composition and visibility/variant wiring; no new result.
- Inference geometry: plugin-backed PLY surface, inference target wiring, model compatibility,
and cache consequences.
- Output field: producer wire, stage result selection, result dataset APIs, manifest metadata,
and visualization selection must all agree.
Authoring loop
- Read the current schema,
workflow.toml, stage, and frontend together. - Add or update the AppState attribute. Preserve its USD type,
allowedTokens, and authored default;
regenerate the specialization schema using the repository's verified codeless-schema workflow.
- Add or update
[controls.<id>]. Keep the control ID equal to the schema attribute short key. - Author only binding kinds implemented by the controller. Use
manifest-reference.md for exact semantics.
- Author the stage-side target: variant set, attribute, relationship, material, callback write, or
Kit-CAE 3.0 dataset/field APIs.
- For a visible control, update
trame-app/app.py. The current UI does not generate widgets from
get_control_manifest. If an inference control participates in lite-cache restrictions, also map its <key>_options in pullAvailableOptionsFromKit() in trame-app/static/app_streamer.js.
- Validate the manifest/schema. Add a composed stage when checking targets, variants, outputs, or
colormap readouts:
``bash tools/validate/rtdt-validate.sh \ --manifest <workflow.toml> \ --schemas <AppStateAPI...> \ --stage <composed-stage.usda> ``
- Run the smallest relevant unit/static checks, then use the appropriate Compose profile for
runtime verification. A clean validator result is necessary but does not prove rendering, streaming, callback behavior, or model accuracy.
Bring-your-own geometry
Use geometry-and-inference.md for the Kit-CAE 3.0 pattern. tools/geometry/prepare_mesh.py remains useful for converting, fitting, centering, repairing, and writing the PLY surface. Treat its scaling notice as a required review item.
prepare_mesh.py emits the current Kit-CAE 3 plugin-backed recipe: an OmniSciDataset payload with OmniSciFileFormatArgsPythonAPI, RtwtModelAPI, and omni:rtwt:model:tag. Its output is ready-to-paste native USDA for a sibling model prim when the command supplies the required --model-tag and a --asset-path relative to the layer that will receive it. For a nested variant tree, apply the APIs once on the base dataset and move only the emitted payload and omni:rtwt:model:tag opinions into each variant leaf.
The directory source/specializations/auto-aero/examples/byo-geometry-cube/ is a static manifest/schema/stage validator fixture. It demonstrates control-to-variant wiring; it is not a runnable frontend, streaming application, or CAE inference example.
Output-field checklist
For every new field, verify all of the following:
- producer tensor and Triton configuration;
RtwtResultFieldSelectionAPI:<wire-id>target andfieldName;- paired
OmniSciArrayAPI:<Field>andOmniSciFieldAPI:<Field>on the one result dataset; [outputs.<wire-id>]metadata;cae:viz:field_selection:<role>:fieldNameson each consumer;- cache data and sidecars for lite mode.
Do not introduce a second result dataset merely to hold another field. Keep fields on the RtwtInferenceResultsAPI dataset unless the runtime contract itself is deliberately redesigned.
Resources
- manifest-reference.md — parsed manifest fields and binding
behavior.
- schema-regeneration.md — exact Kit/OpenUSD 25.11 codeless
schema regeneration commands and review gate.
- geometry-and-inference.md — Kit-CAE 3.0 geometry, domain,
result, visualization, and cache contracts.
- rtdt-aero — exact auto-aero files and geometry recipe.
Security
- Security scanPassed, 95 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/core/rtdt-customize/SKILL.md
Popularity
- Repository stars
- 139
- Repository forks
- 47
- 7-day trend score
- 25
Stars and forks are for the whole GitHub repository, not this skill alone. Refreshed 11 Sep 2026, a month ago.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/nvidia-omniverse-blueprints/digital-twins-for-fluid-simulation/rtdt-customize)