Kelle Cruz

Passionate about making Astronomy better, more efficient, and community focused. Associate Professor at CUNY Hunter College, Research Associate at the American Museum of Natural History. Member of the Astropy Project. Developer of the AstroDB Toolkit.


Session

11-02
16:15
15min
Teaching the Machines Our Tools: Why we need to be building “Skills”, not just writing docs.
Kelle Cruz

“Skills” are a relatively new innovation that teach AI assistants (like Claude or ChatGPT) how to correctly and safely use a specific piece of software, tool, or workflow. Skills are essentially markdown files that provide structured, machine- and human-readable instructions which are used like modular add-ons to an AI assistant. Current AI assistants’ competence with any given tool is largely accidental, shaped by what happened to appear in the training data rather than anything we deliberately taught it. In this talk, I will argue that building Skills should become a standard part of releasing astronomical software and tools, alongside documentation, tests, and tutorials. We can use Skills to encode the tacit knowledge that we normally pass on through documentation, FAQs, and help desks. Skills also provide a mechanism for us to put guardrails on AI assistants and encourage best practices. In this talk, I'll show what a Skill looks like in practice, share examples and use cases, point to Skill-building tools, and discuss ways we can potentially coordinate our Skill development efforts.

AI as tool for software engineering
Banquet Hall