Hugging Face Skills: an agent that knows the Hub and can train models
Working on Hugging Face means dozens of hf commands, the datasets API, training scripts and Space settings that an agent only half remembers. The set holds about 26 skills: the core hf-cli skill is generated from the CLI you have installed so it tracks your version, and the rest are added on demand with hf skills add.
hf-cli teaches every hf command: model search, uploads and downloads, buckets, repos, jobs and launching Spaces.
huggingface-llm-trainer and trl-training cover SFT, DPO, GRPO and reward modelling, plus GGUF conversion for local use.
huggingface-datasets uses the Dataset Viewer API for splits, paginated rows, text search, filters and parquet links.
huggingface-gradio, huggingface-spaces and huggingface-zerogpu help build a demo, deploy it to Spaces and fix a broken build.
The hf-cloud family prepares SageMaker deploys: AWS profile discovery, IAM role checks, serving image choice and an autoscaled endpoint.
When to use it
You want to fine-tune a small language model on your own dataset without hand-writing a TRL config or guessing at hardware.
You plan to run a model locally with llama.cpp and need to know which GGUF quant fits your laptop's memory.
You trained a LoRA and want a Gradio demo for it live on Hugging Face Spaces quickly.
Ask your agent
Find the best open model on the Hub for Spanish speech recognition and compare it with two runners-up on benchmarks
Fine-tune Qwen on my Q&A dataset with SFT using Hugging Face Jobs
How much memory does this model need in GGUF Q4, and will it run on a 16 GB Mac?
Questions
How do I install Hugging Face Skills in Claude Code?
Add the marketplace with /plugin marketplace add huggingface/skills, then run /plugin install hf-cli@huggingface/skills. Further skills come in with hf skills add followed by the skill name.
Why does the marketplace only offer hf-cli?
That is deliberate: the plugin covers core Hub operations, and the full skill list is reachable through hf skills list and hf skills add.
Can I use these skills in Codex or Gemini CLI?
Yes. For Codex, copy skill folders into .agents/skills; for Gemini CLI, the repo ships gemini-extension.json and installs with gemini extensions install.
