[AI-interop] Skill Commons: sharing reliable AI skills for astronomy

[Tiantian Tong]

Dear all,

Thank you for the interesting discussion about AI, MCP servers, and 
natural-language access to astronomy services.

At the Leibniz Institute for Astrophysics Potsdam (AIP), we have been 
developing Ori (Open Research Intern), an AI research agent powered by our 
locally hosted open-source LLM, that can use specialist skills to carry out 
scientific work. More importantly, we have also created an open community 
project called Skill Commons:

https://github.com/skill-commons/skill-commons

Skill Commons is a shared catalogue of skills for scientific AI agents. A 
skill can teach an agent how to query an astronomy service, work with a 
scientific library, analyse data, or follow a reliable research workflow.

The project is not tied to Ori or to any particular AI provider. Researchers 
should be able to use the AI tools they prefer, while sharing the same 
community-maintained knowledge.

Skills remain in the Git repositories of their original maintainers. Skill 
Commons records where they live and which version has been reviewed. We look 
for clear ownership, licensing, safe behaviour, reproducible instructions, 
and realistic verification. When something needs improvement, we work with 
the maintainer in the original repository rather than copying it elsewhere.

Our current collection contains 22 research skills:

https://github.com/skill-commons/curated-research-skills

Following Sébastien’s mention of the ESO-oriented skills, we have already 
opened a conversation with their maintainer about adding the first 
candidates:

https://github.com/szampier/skills/issues/1

They are not registered yet, the review and collaboration are just 
beginning. But this is exactly how we hope Skill Commons can grow: service 
experts and researchers contribute the knowledge, the community helps review 
and improve it, and different AI agents can then make use of it.

MCP servers and skills can also complement each other: MCP can provide 
access to a service, while a skill teaches an agent how to use that service 
carefully and effectively.

We would be delighted to hear from anyone maintaining astronomy services, 
scientific software, useful agent instructions, or existing skills. 
Comments, criticism, candidate skills, and collaborators are all very 
welcome.

Best regards,

Tom Tong
Leibniz Institute for Astrophysics Potsdam (AIP)
https://www.aip.de/en/members/tiantian-tong/