Hugging Face Skills bridges the gap between AI coding agents and the HuggingFace ML ecosystem by providing standardized skill definitions that any ACP-compatible agent can invoke. The collection includes 13 skills covering the full ML workflow: dataset creation and curation, model fine-tuning using TRL for language models and vision transformers, evaluation and benchmarking, and deployment to HuggingFace Spaces and Inference Endpoints. Installation requires just a single npx skills add command.
Each skill is defined as an ACP-compliant specification that describes the capability, its parameters, expected inputs and outputs, and execution requirements. When a coding agent encounters an ML task like fine-tuning a model on custom data, it can discover and invoke the appropriate HuggingFace skill without the developer needing to write boilerplate training code. The agent handles configuration, resource allocation, and execution monitoring through the skill interface.
With approximately 9,500 GitHub stars and Apache 2.0 license, HuggingFace Skills represents the official entry point for the ACP agent skill ecosystem from one of the most important companies in the ML infrastructure space. It was featured at the AI Agent Conference 2026 and has been trending on GitHub, signaling broad developer interest in the pattern of extending coding agents with specialized ML capabilities.