01
What is available?
Check Viewer support, then resolve the exact configuration and split before requesting data.
Inspect first. Download with purpose.
Turn a Hugging Face repository ID into a clear profile: its splits, columns, sample rows, statistics, and available Parquet files.
A real dataset, inspected
plain_textTraining label distribution
Observed through the Dataset Viewer API.
01
Check Viewer support, then resolve the exact configuration and split before requesting data.
02
Preview fields and fetch bounded slices with a zero-based offset and an explicit length.
03
Search string columns or apply supported predicates to narrow the rows you inspect.
04
Read row totals, byte sizes, and available column statistics before choosing a workflow.
05
List actual Parquet URLs for your preferred analysis tool. Listing a file does not download it.
06
The original Skill also covers optional dataset uploads. Account access and authorization are required.
A good fit
Start with a repository ID and a concrete question. Receive API-backed metadata and bounded row samples you can examine or reuse.
Know the boundary
This package does not include datasets, train models, guarantee data quality, or make gated content public. Viewer support varies by dataset.
Before you start
AssistantA Skill-capable assistant that can issue HTTPS requests, or use curl/Python.
AccessInternet is required. The tested public reads needed no additional Hugging Face token.
Optional uploadsUse the Hub UI, @huggingface/hub, or hf with your own account permissions.
Data rightsThe Skill is Apache-2.0. Each dataset retains its own license and terms.
Validation covers direct public HTTP requests. Five search/filter requests returned HTTP 500; authenticated uploads and agent end-to-end execution were not tested.