I have been ramping up on ‘AI Engineering’ for a few weeks and it’s been a blast.  But there are aspects in how AI gets applied that has me scratching my head.  Most of the applications I see involve chatting with an application and it is generating a response.  Great for things like a support helpdesk I suppose but how does that help me manage inventory for a supply chain or schedule patients for surgery? 

Then I see people talking about storing prompts and treating them like code.  Why? Once I’ve run a prompt and it has generated my stuff, why would I run it again? 

Then Github’s Spec-kit showed up, and I saw their prompt files.  This has brought up a bunch of stuff, but it answers how a prompt can be reused:

• We’re attaching files that change to the context and the prompt is using those files to generate its response.  Say I provide July’s orders to the machine, and it generates its response from that, then I provide August’s orders I’ll get a different response.  Makes sense.
• The prompt is a template, and we can replace fields in the template with new values that are provided.
• Like the templating reuse, we have the system prompt, which is provided ahead of the user’s prompt, we could have our reused prompt also attach itself ahead of the user’s prompt.  For even more fun we could have 2 prompts, one gets attached before the user prompt and the other attaches at the end.

Now my interest is finding out how spec kit adds its slash commands to copilot so I can create my own.