Most writing about AI "skills" is engineering writing. Reusable prompts, agent tools, coding patterns. Useful, but it stops at the engineering boundary. The more interesting thing happened when ours crossed it.

We built a shared catalogue of reusable skills. Each one encodes a decision made once, so it never has to be made again. Engineers built them, agents used them, demand grew for more. That much was expected. Then risk, compliance, and content teams started contributing.