I'd suggest starting with a list of common biases and producing a question (or a few?) for each. The questions could test the biases and you could have an explanation of why the biased reasoning is bad, with examples.

It would also be useful to group the biases together in natural clusters, if possible.