Setup & Installation
What This Skill Does
Manages datasets on the Hugging Face Hub. Handles repo initialization, configuration, streaming row updates, and SQL-based querying and transformation via DuckDB. Works alongside the HF MCP server, which covers discovery and search.
Instead of downloading entire datasets locally to filter or reshape them, you run SQL directly against the hf:// protocol and push results back to the Hub without intermediate steps.
When to use it
- Querying a public dataset with SQL to extract a filtered subset
- Pushing a transformed dataset slice to a new Hub repo
- Sampling rows from a large dataset before committing to a full download
- Initializing a new dataset repo with a chat or QA template
- Merging splits from the same dataset into a single export file