| name | content-workbench |
| description | Use when a long-running coding agent needs to inspect, manipulate, render, validate, optimize, restore, or export a USD asset through Content Workbench, including scene sessions, snapshots, rendering, picking, path translation, edit transactions, and workflow evidence. |
content-workbench
Use this skill when a long-running coding agent needs to inspect, manipulate,
render, validate, optimize, or export a USD asset through Content Workbench.
Content Workbench is the scene and asset state service. The agent owns planning,
workflow choices, evidence review, and user interaction.
Normal Loop
- Start or connect to Content Workbench.
- Discover the API from
/agent-api, /agent-api.json, /openapi.json, or
/agent/tool-manifest.
- Load a local USD scene.
- Inspect scene hierarchy, candidates, properties, material bindings, and
diagnostics.
- Render and pick pixels to ground decisions in visual evidence.
- Apply reversible preview edits through Workbench APIs.
- Verify with renders, picks, snapshots, and operation artifacts.
- Restore/export accepted edits back to source space when required.
- Preserve artifacts under the run directory.
Rules
- Do not edit source USD files directly unless the user explicitly asks for
source-file mutation.
- Prefer Workbench APIs over ad hoc USD scripts for scene state that Workbench
owns.
- Prefer stable object or revision handles when available; fall back to paths
only when the current API has not exposed handles yet.
- Treat optimization and restore as Workbench operations, not prompt-time path
rewriting.
- Record evidence artifacts for every meaningful output claim.
- When the API shape is unclear, fetch the Workbench API docs before guessing.
References
references/agent-workbench-api.mdreferences/object-identity-and-mapping.mdreferences/edit-transactions.md
