Question: I already have a codebook I want the AI to use. How to work this codebook into the prompt?

Do you want to:

1. stick only to the codebook
2. stick to codebook mostly but code other stuff too which does not fit into the codebook. If you do this, make it add an additional tag so you can easily find and exclude those if you want, e.g. [new]
3. A good compromise, if your codebook is hierarchical, say: you have to stick to level 1 (or level 2) of the hierarchy but you can improvise the last part of the label.

Try that with one iteration and see what you need to do for one or more subsequent iterations in terms of insisting on it really following that rule as well as following other rules like making more connected stories.

Question: Also, am I right in thinking for the codebook I should include pos and neg sentiments for different labels? (e.g. if I have 'increased income' I should include 'decreased income'?

Yes! If you expect both, then give both. But don't get confused with the world of magnets in soft recoding, where a decrease in blah blah might get attracted to the magnet increase in blah blah , so in that case getting the AI to code sentiment as well is a good way to distinguish between the two. But that problem should not appear in the same way with an explicit codebook. If you are going to offer explicit codes, including pairs like that, you probably don't need to get it to code sentiment at all unless you want that for other reasons.