My workflow uses serveral different agents, I mostly use Claude Code and Cursor but for very different patterns.
Claude Code I can use to plan, create a detailed PRD and have it go through and execute the plan, it will continue to work until it has done everything, it can (mostly) be prompted to continually iterate through (Implement, Test, Correct, Verify) and complete it.
Cursor however has now become more of a surgical tool, especially since Cheetah / Composer where I can give it tasks that I see Claude thrashing around with and it will correct them at which point I can go back to Claude to continue the iteration.
My question is why can’t cursor go through and iterate on these in the same way? It always seems eager to end the flow and the tool timeouts seem to support that this isnt the goal either?
Are there some plans in place or some workflows that exist that allow Cursor to work this way?