huggingface/skills ? reverse-engineered prompt

Hey, can you build me a set of tools or 'skills' that make a coding agent an expert on the Hugging Face ecosystem? The main idea is to give the agent pre-packaged abilities for common machine learning tasks. For example, it should be able to train or fine-tune LLMs and vision models using Hugging Face's infrastructure, and it should know how to manage those jobs. I'd also want it to handle data, like searching and querying datasets on the Hub without a ton of manual code. It should also be able to build quick Gradio web demos for my models, manage evaluation results, and even help publish research papers directly to the Hub. These skills should be easy to install and compatible with different agents like Cursor, Claude, or Gemini, so I can just reference them in my prompts.