Hugging Face has released Hugging Face Skills, a standardized framework for AI/ML tasks that works across major coding agent tools. The repository provides self-contained skill definitions for dataset creation, model training, evaluation, and other machine learning workflows that integrate with OpenAI Codex, Anthropic Claude Code, Google Gemini CLI, and Cursor.

Skills are organized as folders containing a `SKILL.md` file with YAML frontmatter, instructions, and supporting scripts. The framework follows Anthropic's skill format while maintaining compatibility with other agent tools through alternative file structures like `AGENTS.md` for Codex and `gemini-extension.json` for Gemini CLI. With 4.9k stars on GitHub, the project demonstrates strong community adoption.

The repository includes nine specialized skills covering Hugging Face Hub operations, dataset management, model evaluation, compute job execution, model training with TRL, paper publishing, and tool building. Each skill provides detailed guidance for coding agents to execute specific ML tasks, from estimating GPU memory requirements to publishing research papers on Hugging Face Hub. The framework enables developers to extend their coding agents with Hugging Face-specific capabilities while maintaining interoperability across different agent ecosystems.