I’ve been playing with it a bit today, and am certainly finding it useful.
I had one agent with a system prompt that had grown quite a bit through various iterations, and now have that broken down into 7 skills, each in their own topics.
I am interested in seeing some of the improvements you’re hinting at in “limitations”.
In particular, I currently am listing all the skills I want the agent to load in the agent’s system prompt. I’d prefer to have it discover them automatically (or have one more layer of indirection where I point it to a single skill that itself is a topic that links to other skills)
I tried that “indirection” approach initially, but I don’t think it works quite yet to get things injected as custom system messages, so I took a step back with that for now.
But it’d be nice because then I can have a more collaborative workflow for ongoing skills development, without having to edit the system prompt.