---

name: dataset-and-resources description: "Create, inspect, validate, and convert Snips NLU datasets and reason about custom/built-in entities and language resources." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Dataset and Resources

Use this sub-skill when the task is about Snips NLU dataset authoring, conversion, inspection, validation, custom entities, built-in entities, supported languages, or language resources.

This skill is for Snips NLU 0.20.2 / model format 0.20.0 behavior. It does not train engines, parse user text, persist models, download resources, or run network-backed setup.

Route elsewhere

• Training, parsing, SnipsNLUEngine, fit, parse, get_intents, get_slots, persist, and from_path: use the sibling ../engine-api/SKILL.md.
• CLI command construction or execution for generate-dataset, download, download-entity, download-language-entities, link, metrics, train, parse, or version commands: use ../cli-workflows/SKILL.md.

Operating procedure

1. Identify the input form:
   • YAML authoring documents with type: intent or type: entity.
   • JSON training datasets with root keys language, intents, and entities.
2. For YAML details, slot annotation syntax, implicit slot/entity behavior, and JSON schema, read references/data-formats.md.
3. For built-in entity names, custom entity knobs, supported languages, and resource implications, read references/resources-and-entities.md.
4. Validate a JSON dataset before handing it to training/API workflows:

   python scripts/validate_snips_dataset.py --dataset dataset.json --explain
   

5. Convert YAML authoring files to JSON without invoking training:

   python scripts/snips_yaml_to_json.py --language en intent.yaml entity.yaml --output dataset.json
   

6. If conversion or validation fails, diagnose using references/troubleshooting.md; keep fixes in the dataset files, then rerun validation.

Key API anchors

• Dataset.from_yaml_files(language, filenames) converts one or more YAML files/streams into a dataset object whose .json property is the JSON training format.
• validate_and_format_dataset(dataset) accepts either a dataset object or JSON-like dict, checks the authoring schema, normalizes custom entity data, and returns a formatted dataset with validated: true.
• load_resources(name, required_resources=None) loads language resources by language/resource name, resource package, or directory path; do not download implicitly from this sub-skill.

Safety notes

• The bundled scripts only read provided dataset files and optionally write a JSON output file; they never train a model or download resources.
• snips_yaml_to_json.py refuses to overwrite an existing output file unless --overwrite is provided.
• If a built-in entity or language resource is missing, route setup/download command work to ../cli-workflows/SKILL.md instead of attempting network access here.