For anyone interested, I've put my testing framework up on Github. It allows you to run multi-turn dialogues easily, with lots of configurable output options, including writing results.
It will also, helpfully, take care of getting access and refresh tokens for you. When you first run it (or run the #init method), a browser window will pop up and take you through the initial OAuth flow to get and store the tokens.
Scroll to the end of the Readme to see how to use it with some simple RSpec tests.