Build Hugging Face API CLI Scripts

This guide turns repeated Hugging Face Hub tasks into reusable command-line scripts. Utilities can call the API directly, use the hf CLI, or read model and dataset cards from repositories. Shell scripts are preferred for simple operations; Python or TSX is appropriate when complexity or user requirements justify it. The design favors small interfaces that support chaining, piping, and intermediate processing.

Every script must accept --help and explain its inputs and outputs. Inspect the shape of representative API results before committing to a design, test non-destructive utilities before handover, and provide usage examples. Authentication uses HF_TOKEN in an Authorization header for the appropriate access and rate limits. Handle authentication without exposing the token in shared examples or output.

Use this guide for composable Hub automation rather than repeated ad hoc requests. Its examples describe model searches, paper-related discovery, card frontmatter extraction, metadata enrichment, and sorting or filtering JSON results. Baseline utilities return raw JSON; an enrichment utility reads model IDs from standard input and emits one JSON object per line for streaming workflows.

The catalog copy does not bundle the referenced helper scripts. These include model-paper discovery, card-frontmatter extraction, baseline API clients in shell, Python, and TSX, and an enrichment script. Dependent steps require the original supporting material or an existing installation. Resolve paths relative to that installation, inspect the files, and check licensing and execution prerequisites. Do not fabricate missing helpers, substitute an unrelated same-named file, or claim an example pipeline ran when its dependencies are unavailable. Fully documented independent steps can still proceed within the authorized scope.

Inputs include the requested Hub operation, search parameters or model IDs, relevant authorization, and a clear output format. Outputs are reusable scripts, their help text, tested examples, and structured results suitable for downstream processing. Clarify missing permissions, required inputs, or success criteria instead of guessing them.

Query only relevant parts of the large OpenAPI document, for example:

curl -s "https://huggingface.co/.well-known/openapi.json" | jq '.paths["/api/models"]'

It lists endpoint families for models, datasets, Spaces, collections, daily papers, notifications, settings, identity, and trending content. Constrain live example responses to small but representative result sets. The hf CLI provides authentication, caching, downloads, endpoints, jobs, repository operations, and uploads; inspect the installed tool's help for the operation being automated. The guide's command listing describes its documented interface, not proof that a local installation or every referenced support file is present. Environment-specific validation and testing remain necessary before handing over the utility.