Hi Everyone,
Been trying to solve the multi-agent collaboration problem in Cursor.
I wrote a small plug-in that is a “light-weight” agent collaboration (comes in super handy in plan/debug modes).
Inspired the Zen mcp /consensus mode; however true cursor-native (no extra providers API keys required)
The lead model (whatever you’re chatting with — Opus, GPT, Gemini, etc.) drafts a plan or a debugging hypothesis, hands it to two or more read-only “planning buddy” subagents pinned to different models, collects their critiques, and synthesizes a single refined output that incorporates every perspective. You review the consensus output and approve it before any code is written or any fix is applied.
The diversity of perspective comes from the model swap, not from differently-prompted personas.
If anyone is interested to try it out