Hugging Face Skills: ML Workflows for Claude Code
Hugging Face's official skills teach Claude the hf CLI, datasets, model training, evaluation, Gradio and SageMaker deployment. Install steps, uses and caveats.
By Claudekit Team Published 1 min read
- Category
- Machine Learning
- Source
- github.com
- Repository
- github.com
- GitHub stars
- 11,204
- License
- Apache-2.0
Install
/plugin install hf-cli@huggingface/skillsHugging Face Skills are official agent skills for AI and ML tasks on the Hugging Face Hub: downloading and uploading models and datasets, building datasets, training and evaluating models, building Gradio demos, and deploying to Amazon SageMaker. They follow the Agent Skills standard and work in Claude Code, Codex, Gemini CLI and Cursor.
What it covers
The repo held 25 skills when we checked today, including hf-cli, hf-mem (estimate memory for Safetensors or GGUF weights), huggingface-best (find the best model for a task from benchmarks), huggingface-datasets (Dataset Viewer API workflows), huggingface-community-evals (inspect-ai and lighteval), huggingface-gradio, and a set of hf-cloud-* skills for SageMaker deployment planning, IAM preflight, serving images and production defaults.
Install
From the README:
/plugin marketplace add huggingface/skills
/plugin install hf-cli@huggingface/skillsThen add more skills with the CLI:
hf skills add <skill-name>A huggingface-skills plugin is also listed in Anthropic’s official marketplace.
Example prompts
Find the best open model under 8B parameters for Japanese summarization
How much GPU memory do I need to load this GGUF model?
Explore the splits of this dataset and download only the validation parquet
Deploy this model to a SageMaker endpoint with autoscaling and alarmsTips and limits
- Uploads, jobs and SageMaker deploys cost money or affect public repos; confirm before running.
- Log in with the
hfCLI and keep tokens out of prompts and committed files. - Benchmarks reflect published scores, not your data; evaluate on your own set.
Related: Scientific Agent Skills, AWS MCP servers, skills guide.
Frequently asked questions
Which Hugging Face skill should I install first?
hf-cli. Hugging Face recommends it as the bootstrap skill; it teaches the agent the hf CLI for models, datasets, Spaces, buckets, jobs and more.
How do I add the other skills?
After hf-cli is installed, run hf skills add followed by the skill name to pull additional workflow skills.
Does it work without skills support?
Yes. The repo provides an agentsmd/AGENTS.md fallback bundle for agents that do not support skills.