The Hugging Face Skills repository provides a collection of standardized skill definitions that enable AI coding agents to interact with the Hugging Face ecosystem. Skills are self-contained folders that package instructions, scripts, and resources together, allowing agents like OpenAI's Codex, Anthropic's Claude Code, Google DeepMind's Gemini CLI, and Cursor to perform specific AI and ML tasks. Each skill includes a SKILL.md file with YAML frontmatter containing the skill's name and description, followed by guidance that agents follow when the skill is active.
The repository contains 20 available skills covering a wide range of Hugging Face-related functionality. The hf-cli skill serves as the recommended starting point, teaching agents every hf command for searching models, managing datasets and buckets, launching Spaces, and running jobs. Additional skills enable agents to estimate memory requirements for model weights with hf-mem, discover optimal models through leaderboards with huggingface-best, and manage evaluation results in model cards via huggingface-community-evals. The huggingface-datasets skill allows exploration and querying of Hugging Face datasets through the Dataset Viewer REST API without Python dependencies, while huggingface-gradio enables building web UIs and demos.
For machine learning workflows, the repository includes specialized training skills. The huggingface-llm-trainer skill covers training and fine-tuning language models using TRL on Hugging Face Jobs infrastructure, supporting SFT, DPO, GRPO, and reward modeling methods. The huggingface-vision-trainer skill handles object detection and image classification model training with support for COCO dataset format and various model architectures. The train-sentence-transformers skill provides fine-tuning capabilities across bi-encoder, cross-encoder, and sparse encoder architectures. Additional skills like huggingface-local-models enable running models locally with llama.cpp and GGUF formats, while huggingface-spaces facilitates building and deploying applications on Hugging Face Spaces with support for ZeroGPU hardware.
The skills follow the standardized Agent Skills format from agentskills.io, making them interoperable across multiple agent platforms. Skills are available through the Cursor Marketplace, OpenAI's Codex Plugins Directory, and can be installed directly into Claude Code, Codex, Gemini CLI, and Cursor through various installation methods. For agents that don't support skills natively, the repository provides an agentsmd/AGENTS.md fallback file.
GitGenius activity data shows the repository maintains active community engagement with a median issue and pull request response latency of 44.9 hours across tracked items. The most active contributors include evalstate with 9 events, burtenshaw with 5 events, and Muitamax with 4 events. The repository shares overlapping contributors with related projects including langfuse/langfuse, jeffallan/claude-skills, and hpcaitech/colossalai, indicating integration within a broader ecosystem of agent and AI tooling projects.
The repository is classified across multiple domains including AI Agents, Agent Skills, Tool Use, Function Calling, Task Automation, AI Capabilities, Information Retrieval, Action Execution, API Interaction, and Modular AI. This broad classification reflects the repository's comprehensive approach to enabling agents to leverage the full Hugging Face platform through modular, reusable skill definitions that can be combined to automate complex ML workflows and Hub interactions.
