Release optimized skills and benchmark datasets on Hugging Face

Hi @Yif29 🤗

Niels here from the open-source team at Hugging Face. I discovered your work through Hugging Face's daily papers as yours got featured: https://huggingface.co/papers/2605.23904.
The paper page lets people discuss about your paper and lets them find artifacts about it (your models, datasets or demo for instance), you can also claim the paper as yours which will show up on your public profile at HF, add Github and project page URLs.

I saw you've released the code for the SkillOpt optimizer on GitHub! It'd be great to also host the optimized skill artifacts (the best_skill.md files) and the benchmark datasets (like SpreadsheetBench, LiveMathematicianBench, and OfficeQA) on the 🤗 hub to improve their discoverability and visibility.

As you mentioned in the paper, these optimized, reusable skills could become a new adaptation layer for future agents. Hosting them on Hugging Face would make it much easier for others to download and plug them into their own agentic loops.

Uploading optimized skills

You could host these as model repositories (using the text-generation tag) or as a collection of configuration artifacts. Hosting on Hugging Face will give you more visibility and enable better discoverability through metadata tags. See here for a guide: https://huggingface.co/docs/hub/models-uploading.

Uploading datasets

It would also be awesome to make the benchmark datasets available on 🤗, so that people can easily evaluate their agents using:

from datasets import load_dataset
dataset = load_dataset("microsoft/SpreadsheetBench")

Besides that, there's the dataset viewer which allows people to quickly explore the task items in the browser.

Let me know if you're interested or need any help regarding this!

Cheers,

Niels
ML Engineer @ HF 🤗

Start by reviewing the SkillOpt repository and the existing best_skill.md artifacts, then inspect SpreadsheetBench, LiveMathematicianBench, and OfficeQA. Use the linked Hugging Face model-uploading and dataset-viewer guidance to determine the required metadata and repository structure. Done means the optimized skills and benchmark datasets are published on Hugging Face and can be discovered or loaded by users.