Interesting post, but that much scaffolding around the tests is a bit of a red flag to me.
I personally try to move as much logic as possible into easily testable, independent methods. This leaves things like Rake tasks as very small, "obviously correct" shells around the actual logic.
I highly recommend checking out the talk Boundaries by Gary Bernhardt[0], which explains this idea far better than I can.