Hugging Face Skills 🧩 Skill Free

What it does

Hugging Face Skills is Hugging Face's official set of Agent Skills for machine-learning work on the Hub. The entry point is hf-cli, a skill that teaches the agent the hf command line. With it, the agent can search models, manage datasets and buckets, launch Spaces and run jobs. From there, more specialised workflow skills can be added when you need them.

What is inside

At the time of writing (September 2026) the repository contains about 25 skills, including:

• Training: fine-tuning language and vision models with TRL or Unsloth on Hugging Face Jobs, training sentence-transformers, and tracking experiments with Trackio
• Evaluation: running community evaluations locally with inspect-ai and lighteval
• Data and models: Dataset Viewer API workflows, paper lookup and publishing, finding the best model for a task, choosing local GGUF models, and estimating memory needs
• Apps and deployment: building Gradio apps, Spaces and ZeroGPU demos, using Transformers.js, and a set of skills for deploying models to Amazon SageMaker

Works with

The README documents support for Claude Code, Codex, Gemini CLI and Cursor. The hf-cli skill is also listed in the Cursor Marketplace and the Codex plugins directory. For agents without skill support, there is an AGENTS.md fallback.

How to install

For Claude Code:

/plugin marketplace add huggingface/skills
/plugin install hf-cli@huggingface/skills
hf skills add <skill-name>

For Gemini CLI:

gemini extensions install https://github.com/huggingface/skills.git --consent

For Codex, copy the skills you want into an .agents/skills folder.

Maintenance and safety

The repository is Apache 2.0 licensed and actively updated. Many skills run scripts and call Hugging Face services. Training skills can start paid Jobs on cloud GPUs, and the SageMaker skills create AWS resources, so check costs and credentials before letting the agent run them. The Cursor plugin also configures the Hugging Face MCP server.

Who should use it

Developers and researchers who train, evaluate or ship models on the Hugging Face Hub. For app work that does not touch ML, only hf-cli is likely to be useful.