I'm working on something a bit like what you described. So far it has a command suggestion mode and also a generic question mode. Inbound pipe support is a TODO

I'm working on something a bit like what you described. So far it has a command suggestion mode and also a generic question mode. Inbound pipe support is a TODO, but command substitution already works fine for many use cases like passing file contents to the LLM. I'm pretty wary of adding any automatic execution of AI generated commands, though using some isolation scheme like containers is an interesting possibility. https://github.com/scottyeager/Pal https://github.com/scottyeager/Pal