This could be a silly question, but has someone used BAML for generating the skills/commands for RPI flows

/research_codebase, /create_plan, /implement_plan

for different providers? I am wondering if it was ever discussed it was useful to have these be generated, serilaized. There are some advantages of this I can see: - Reuse of common sections across commands - Templating skills/commands for different LLM providers.

I've used RPI for generating RPI flows for different providers. How do you see BAML fitting into this?

Example: Lets say I want to generate create_plan commands for different LLM providers. You would want to template them: e.g. conventions, syntax for calling sub agents/tools is different for different providers. So that's one way to use BAML.

I have to agree with what xtenzq said in the humanlayer discord group. I don’t see how BAML could help with tool syntax differences between providers, cuz all you have to change is a couple of lines of English in a md file. For your second example, wouldn’t it be easier to use a prompt the model to say which logger to use? But if you're not happy with the probabilistic nature of prompts, you can force the LLM to run a script every time it changes a py file that - checks if the right logger has been imported - run the script from the right env