hf-cli

The Hugging Face Hub CLI skill wraps the hf command-line tool for interacting with the Hugging Face Hub, covering models, datasets, spaces, buckets, repositories, papers, jobs, and collections. It is the successor to the deprecated huggingface-cli, and applies broadly whenever a user mentions Hugging Face or needs to download, upload, or manage AI and ML assets, or wants cloud storage for training checkpoints, data pipelines, or agent traces. Installation is a single script: curl -LsSf https://hf.co/cli/install.sh | bash -s.

The tool groups functionality into top-level commands and subcommand namespaces. Core file operations include hf download and hf upload (both accepting --type model, dataset, or space, include/exclude filters, and revisions), hf cp for copying between local paths, repos, and buckets, and hf sync for reconciling a local directory with a bucket. Authentication lives under hf auth (login, logout, switch, token, whoami, list). Storage is managed under hf buckets (create, delete, info, list, move, remove, cp, sync), and the local download cache under hf cache (list, prune, rm, verify). Additional namespaces cover hf collections and hf datasets, the latter exposing dataset cards, info, and leaderboards for comparing model benchmark scores.

Most commands accept a --format option (auto, human, agent, json, or quiet) so output can be tuned for human reading or machine parsing, which is useful when the CLI is driven by an agent. The skill notes it was generated from huggingface_hub v1.22.0 and can be regenerated with hf skills add --force. It encourages reaching for hf --help to discover the full command surface rather than memorizing every flag.