huggingface/ml-intern ? reverse-engineered prompt

I want to build an autonomous AI agent that acts like a machine learning intern. The idea is that I can give it a high-level ML task, and it can figure out how to research, code, and execute it on its own. For example, I should be able to tell it "fine-tune a Llama model on a dataset I have" and have it work independently.

The agent should be built in Python and integrate deeply with the Hugging Face ecosystem. It'll need tools to search for documentation, read research papers, and find datasets and models on Hugging Face. It should also be able to search GitHub for code examples and have a local sandbox to run the code it generates.

The main way to use it should be a command-line tool. I'd like it to have an interactive chat mode, but also a simple one-shot mode where I can give it an instruction and let it run. It will use models from services like Anthropic, so it should be easy to configure API keys for that and Hugging Face.