>>105616078
I assume it wouldn't work well regardless but that's the point of a complex prompt to benchmark something. It seems like currently you need to either keep the prompt to mostly 1 action or explain everything in absolute detail and hope it will understand it better.