one workflow that improved my experience with Cursor is breaking tasks into smaller prompts.
instead of asking:
“rewrite this whole module”
i usually ask:
• explain this function
• suggest improvements
• refactor only this part
this makes the results much more accurate and easier to review.
Has anyone else found good prompting patterns when working with Cursor?
beside that, when creating a plan, i tend to say (write) that it should break it into phases then instead of hitting build button i switch to agent mode manually then instruct it to go phase by phase (where each phase can have multiple tasks). i’ve noticed some improvements on how it does stuff (depending on model of course) but it also make easier to review what it did (and if necessary to fix/redo/adapt/etc), on phase level, instead of waiting to finish whole.. not sure if this helps anybody or if it’s actual way to go, but it did helped me 