Comment by nickorlow

This does seem like an interesting idea. I'd wonder what the evaluation looks like?

Does it just look at code quality? Or does it also incorporate some amount of apparent skill shown w/ prompting/managing an agent?