I’ve been working on something for a while now. You can see the entire thing at GitHub - TaylorHuston/ai-toolkit: A lightweight, fully customizable framework to enable Spec Driven Development with Claude Code. , but the TL;DR is it’s a Claude Code plugin that is centered around the idea of specific, multi agent workflows. Claude Code has subagents. Which are really just a custom prompt and isolated context window. But this allows me to have my ‘backend-specialist’ agent write some code and then immediately have my ‘code-reviewer’ agent review and score said code. If it gives a score of less than 90, my workflow kicks it back to the ‘backend-specialist’ agent to make improvements based on the ‘code-reviewer’ feedback. This loops as long as needed. I’ve found having multiple subagents review each other’s work dramatically reduces errors and the premature “this task is done!” statements that I was getting when I was just using the main agent to work through a task.
I was hoping to be able to do something similar in Cursor 2.0, but after playing around with it a bit, that doesn’t seem to be possible? I can do parallel agents, but I don’t see a way to define specialist agents and then define a workflow for them to work together, hand off tasks, etc. I asked Composer 1 if it was possible and it suggested just using cursor rules, but I don’t think that will actually accomplish what I am after.