Hugging Face Skills provide standardized Agent Context Protocol (ACP) definitions for AI/ML tasks, enabling interoperability across major coding agents like OpenAI Codex, Anthropic Claude Code, and Google Gemini CLI. They streamline complex ML workflows by equipping agents with task-specific instructions, scripts, and resources, enhancing developer productivity.

How It Works

Skills are self-contained directories packaging instructions (via SKILL.md, AGENTS.md, or gemini-extension.json), scripts, and resources for specific AI/ML use cases. This modular approach allows coding agents to dynamically load and execute predefined workflows for tasks such as dataset creation, model training, and evaluation, abstracting away underlying complexities.

Quick Start & Requirements

• Claude Code: Register repo as plugin (/plugin marketplace add huggingface/skills), then install skills (/plugin install <skill-folder>@huggingface-skills).
• Codex: Skills identified via AGENTS.md. Verify loading with codex --ask-for-approval never "Summarize the current instructions.". See Codex AGENTS guide.
• Gemini CLI: Install locally (gemini extensions install . --consent) or via GitHub URL (gemini extensions install https://github.com/huggingface/skills.git --consent). See Gemini CLI extensions docs.
• Dependencies: Requires compatible AI coding agent tools. Integrations with Cursor, Windsurf, and Continue are planned.

Highlighted Details

• hf_dataset_creator/: Facilitates structured training dataset creation with prompts, templates, and scripts.
• hf_model_evaluation/: Orchestrates model evaluation jobs, generates reports, and maps metrics.
• hf-llm-trainer/: Comprehensive skill for LLM training, including helper scripts like train_sft_example.py, convert_to_gguf.py, and cost estimators.
• hf-paper-publisher/: Tools for managing research papers on Hugging Face Hub, including arXiv indexing and article generation.

Maintenance & Community

The provided README does not detail specific contributors, sponsorships, or community channels.

Licensing & Compatibility

The specific license type is not explicitly stated in the README. The project emphasizes broad compatibility with multiple AI agent frameworks.

Limitations & Caveats

The term "Skills" originates from Anthropic's Claude AI, although the repository aims for cross-platform compatibility. Each skill requires explicit installation or registration within the target coding agent environment.