Things like 1 input, 1 output with 1 bit of memory and looking at them from a decision theory point of view. Asking questions like "Would we view it as having a goal/decision theory?"

I tend to let other people draw those lines up. It just seems like defining words and doesn't tend to spark my interest.

If not what is the minimal agent that we would and does it make any trade offs for having a decision theory module in terms of the complexity of the function it can represent.

I would be interested to see where you went with your answer to that one.