Here's a simple framework I use for improving the performance of your fav agent on (semi-)complex tasks.