nvidia-omniverse/usd-content-agents/content-sdf-operations
Claude Code and agent skill by nvidia-omniverse/usd-content-agents
- community
Checking that the SKILL.md file is still in the repository.
Use when a content agent needs bounded signed- or unsigned-distance-field operations such as mesh rasterization, field booleans, offsets, filtering, sampling, topology conversion, surface extraction, or admitted field output through an in-process backend.
Install
npx vskill@latest install nvidia-omniverse/usd-content-agents/content-sdf-operationsSKILL.md
Content SDF Operations
Use the backend-neutral sdf_tools contract for signed- and unsigned-distance field work. Select semantic capabilities first; treat the implementation that provides them as recorded execution evidence, not as the capability surface.
When To Use
- Convert a world-space polygon mesh to a signed or unsigned distance field.
- Apply bounded field booleans, offsets, smoothing, normalization, or rebuilds.
- Sample field values or gradients at declared world-space points.
- Convert active topology to an SDF, extract an enclosed region, or polygonize
an isosurface.
- Write a backend-supported field artifact with an explicit format.
- Inspect which admitted backend and exact implementation executed an operation.
Use a qualified Geometry Repair workflow instead when the goal is to accept or publish repaired geometry. A successful SDF operation does not establish source fidelity, protected-feature preservation, correspondence, or repair acceptance.
Ownership Boundary
This is an atomic capability, not a workflow. It owns one admitted in-process SDF session and the bounded operations requested from that session. It does not own run state, artifact lifecycle, refinement policy, geometry acceptance, or a terminal workflow outcome. The calling workflow owns those decisions and must bind this capability's execution evidence into its own durable handoff.
Core Contract
backend="auto" is for exploratory work. For a benchmark, qualified geometry route, or durable artifact, select the exact backend admitted by the governing policy and persist backend_info. The first production-qualified driver is openvdb, backed by the source-locked OpenVDB 13 runtime; future admitted drivers may implement the same operations.
Rules
- Keep world units explicit. Signed fields use negative values inside.
- Declare voxel size, band width, interpolation, isovalue, thread count, and
resource limits rather than relying on an implementation default.
- Treat
Fieldas an opaque, backend-owned in-process value. Do not unwrap its
payload, mix fields from different backends, or serialize a live field handle.
- Request every operation and artifact format needed when creating a session,
using require_formats together with READ_FIELDS or WRITE_FIELDS so the direction is explicit. An unavailable capability must fail admission; it is not permission to substitute another algorithm or lower-level API silently.
- Do not import
openvdboropenvdb_runtimefrom content-agent code. - Do not install, import by module path, or register a model- or caller-supplied
backend. Installed discovery admits only exact entries in checked-in policy.
- Driver calls execute inside the importing CPython process. Do not add an SDF
helper process, daemon, RPC service, JSON transport, or file-based bridge.
- Do not admit an SDF dependency delta that introduces or bundles an LGPL
component. Require the structured license manifest and native-closure gate; a top-level package license alone is insufficient.
- Preserve resource-limit, unavailable, invalid-geometry, and backend-operation
failures as distinct outcomes.
Evidence
For durable outputs, record the requested operations, selected backend ID, backend_info, execution limits, units, voxel and band controls, input/output digests, and any selection rejections. Record the explicit artifact format for field output. Backend identity is provenance, not a replacement for geometric validation.
References
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingagentic/.agents/skills/content-sdf-operations/SKILL.md
Popularity
- Repository stars
- 205
- Repository forks
- 25
- 7-day trend score
- 27
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
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/usd-content-agents/content-sdf-operations)