Adding a flag to run system commands upon detecting a keyword

[tjohnman]

I’ve been thinking about adding a flag that would let the user specify a keyword and a command so that upon sampling that keyword from the model, the command would be executed and it’s output fed back in.

I’ve figured out how I would go about doing this, but I thought maybe this kind of feature should be implemented at a higher level and not baked directly into llama.cpp.

What do you think?