What is skills?

Hugging Face Skills is a collection of ready-made instruction sets that teach AI coding assistants how to work with the Hugging Face ecosystem. Hugging Face is a platform that hosts thousands of AI models, datasets, and tools. The skills in this repository give AI agents like Claude Code, Cursor, OpenAI Codex, and Google Gemini CLI step-by-step guidance for specific tasks, such as training a model, exploring a dataset, running evaluations, or publishing a research paper. Each skill is a self-contained folder with a file called SKILL.md that contains the instructions the agent reads and follows. When you install a skill and then ask your coding assistant to do something related, it draws on those instructions to complete the task correctly. The idea is similar to giving a new employee a procedure manual for a specific job, rather than expecting them to figure it out from scratch. The available skills cover a wide range of tasks. There are skills for downloading and uploading models and datasets using the hf command-line tool, for finding the best AI model for a given task by checking benchmarks, for training language models with various fine-tuning methods, for running models locally on your own machine, for building simple web demos with Gradio, and for tracking training experiments. There are also skills for computer vision model training, working with JavaScript-based models in a browser, and managing research papers on the platform. Installation works differently depending on which coding tool you use. Claude Code users run a plugin command, Cursor users install via a marketplace link, Gemini CLI users install via a local extension command. Once installed, you can ask your agent to use a specific skill by naming it in your request. Contributors can add new skills to the repository by following the same folder structure and SKILL.md format used by the existing ones.