[scottyeager]

I'm not the OP, but I did build a tool that I use in the same way: https://github.com/scottyeager/Pal

Actually for many cases the LLM already knows enough. For more obscure cases, piping in a --help output is also sometimes enough.