Build Hugging Face API CLI Scripts

This skill turns an agent into a builder of reusable command-line scripts and utilities around the Hugging Face API and the hf CLI tool, favoring composable, pipeable interfaces over one-off API calls. It documents the main high-level API endpoints (/api/datasets, /api/models, /api/spaces, /api/collections, /api/daily_papers, /api/notifications, /api/settings, /api/whoami-v2, /api/trending, /oauth/userinfo), the full OpenAPI spec at https://huggingface.co/.well-known/openapi.json (160 endpoints, too large to read directly, so it's queried with jq instead), and the hf CLI's own command set (auth, cache, download, endpoints, env, jobs, repo, repo-files, upload, upload-large-folder, version), noting that hf has replaced the deprecated huggingface_hub CLI.

Use it when you need reusable CLI scripts around the Hugging Face API or the hf tool, want shell-friendly utilities that support chaining and piping, or are automating repeated Hub tasks and need a composable interface instead of ad hoc API calls.

Every script must accept --help describing its inputs and outputs, and non-destructive scripts should be tested before being handed to the user; shell is preferred unless complexity calls for Python or TSX, and authentication goes through the HF_TOKEN environment variable as an Authorization header, for example curl -H "Authorization: Bearer ${HF_TOKEN}" https://huggingface.co/api/. Example scripts shown include hf_model_papers_auth.sh (chains trending, model metadata, and model card parsing with fallbacks, using HF_TOKEN automatically), find_models_by_paper.sh (authenticated search with a retry path when arXiv-prefixed searches are too narrow), hf_model_card_frontmatter.sh (uses the hf CLI to download model cards and emit NDJSON of license, pipeline tag, tags, and gated-prompt flag), three ultra-simple baseline scripts in bash, Python, and TSX, and hf_enrich_models.sh, which reads model IDs from stdin and streams one JSON object per line. A composability example chains them:

baseline_hf_api.sh 25 | jq -r '.[].id' | hf_enrich_models.sh | jq -s 'sort_by(.downloads) | reverse | .[:10]'

Talks to the Hugging Face Hub API directly, documented via its OpenAPI schema, and to the hf CLI for repo, download, upload, and Inference Endpoint management; jq is used throughout for querying both the OpenAPI spec and API responses. A related example pipes model IDs through hf_model_card_frontmatter.sh to summarize license and gated-prompt status: printf '%s\n' openai/gpt-oss-120b meta-llama/Meta-Llama-3.1-8B | hf_model_card_frontmatter.sh | jq -s 'map({id, license, has_extra_gated_prompt})'.

Developers automating Hugging Face Hub workflows, such as querying model or dataset metadata, checking licenses and gated status, or managing repos and uploads, who want a library of small, composable, well-documented scripts rather than writing one-off curl calls each time.