In the DeepSeek Harness (DSH) plugin ecosystem, standardizing scene data is critical for the model inference stage. Developers often need to generate structured tour_robot YAML from raw point coordinates and narration materials while ensuring the accuracy of field definitions. rein-scene-forge-dsh-plugin is an automation tool designed precisely for this purpose.

What It Is

This is a DSH plugin maintained by peng2048. Its core responsibility is to generate and validate tour_robot YAML files based on the authoritative points.json and narration materials. The plugin is open source under the MIT License.

Core Features

The plugin has a robust generation and validation mechanism built in, with the following key characteristics:

  1. Field Authority: The plugin uses references/model-catalog.md as the basis for field selection, ensuring that the output structure conforms to established standards.
  2. Strict Validation: It internally incorporates schemas/tour-robot/*.schema.json; any target YAML field not defined in the schemas is rejected.
  3. Ruleset Management: It includes a versioned, machine-readable ruleset located in the rules/ directory.
  4. Deterministic Tools: The plugin exposes four deterministic tools for processing scene data:
    • get_authoring_rules
    • analyze_point_input
    • validate_authoring_model
    • validate_tour_robot_scene

Typical Usage

Developers typically use the four deterministic tools above to process scene data. By default, the plugin outputs a generated scene as a <scene_id>/ directory containing:
* map_config.yaml
* scene_config.yaml
* explain_config.yaml
* reception_pack.yaml

Development and Testing

For development checks, you can use the following commands:

pnpm test
pnpm run smoke
pnpm pack --dry-run

Notes and Constraints

  1. Import Threshold: Every generated scene must complete the validation defined in references/validation-sop.md. Only scenes with is_publishable: true can be imported; scenes with is_valid: true but with warnings cannot be imported.
  2. Dynamic PAD Reception: Newly generated scenes enable dynamic PAD reception by default. If user instructions are missing, this feature must not be silently disabled.

Runtime Dependencies

The plugin runs on Node.js and has specific version requirements for upstream dependencies:
* Node.js version must be >= 20.
* The following DSH ecosystem components are required (version requirement >=):
* @deepseek-ai/cordis
* @deepseek-ai/dsh-skill
* @deepseek-ai/dsh-tools

Conclusion

By using codified rules and deterministic tools, the plugin addresses standardization issues in scene generation. For more details, see the plugin catalog page or the GitHub repository.