I'm not sure how useful this will be, but it seems like more complex plans or do_nothing stages may benefit from the AI generating criticism of its own plan, and it doesn't seem to harm performance judging from a few tests (though feel free to check yourselves, of course!).
Hey @swapneils! Thanks for the PR! We actually did have criticisms at first, but decided to remove it to keep things simpler. So let's hold off on this until we're sure it gives noticeably better results.
I'm not sure how useful this will be, but it seems like more complex plans or do_nothing stages may benefit from the AI generating criticism of its own plan, and it doesn't seem to harm performance judging from a few tests (though feel free to check yourselves, of course!).