We Are Building an Alexandria Library for the AGI Era, and We Aim for Nature

Just by running a single command while using AI for research, you can turn your tacit knowledge into a Nature co-authorship.

Author: Koutian Wu; GitHub: ktwu01

The past few days we have been working on something a bit crazy, but honestly pretty amazing.

Whenever I read a new paper, I always feel that the most valuable parts never make it into the final text. All those intuitions about tuning parameters, avoiding hidden pitfalls, and navigating the messy reality of research, they usually just vanish when someone graduates or switches fields. It is honestly a huge waste.

So we are building an open source project called ResearchSkills.ai. It is essentially an Alexandria Library for the AGI era.

What we are trying to do is extract all that precious tacit knowledge from your brain and turn it into skills that AI agents like Claude Code or Codex can directly read and reuse.

It might sound a bit abstract, but we built a fully local extraction tool. If you use Claude Code for your research, you literally just run one command and it automatically scans and summarizes your procedural rules and undocumented tricks.

Just to be clear on this, everything runs entirely locally. It will not upload anything until you explicitly review and approve it, so privacy is completely secure.

This is a deeply significant endeavor. We plan to submit this open source academic collaboration straight to Nature, and we are establishing a non-profit organization to run it continuously. It has the potential to become the largest scientific collaboration project in history.

1. As long as you contribute research skills, you have a chance to become a co-author! Moreover, researchers who submit a higher quantity and quality of skills will rank higher on the author list.
2. The extracted skills take effect locally first. Your Claude / Cursor will immediately master the exclusive research expertise in your domain, making it increasingly smooth to use.

I have been running the extraction on my own logs for the past few days. For my friends in research, this is something you can do casually while coding anyway. Taking two minutes to run this might actually land you on a Nature paper.

Installation is ridiculously simple. Just run this in your terminal, npm install -g @scienceintelligence/researchskills-extract

Then just type /researchskills-extract in your Claude Code. If you are using Codex, start it with permissions and type $researchskills-extract.

For those who prefer the web interface, you can manually submit at researchskills.ai/submit-manually.

Research is hard enough as it is. We might as well save these hard earned experiences and make them the foundational building blocks for future AI scientists.

That is all for today. If you found this interesting, feel free to share it around. Thanks for reading, and I will see you next time.