Feature request for product/service
Cursor IDE
Describe the request
I would like to propose a feature that improves handling complex problems and large code modifications through a Multi-Model Review Workflow.
The idea is to allow multiple LLMs (from different providers or even the same model across multiple iterations) to collaborate in defined roles:
Planner (creates the plan)
Implementer (applies the change)
Reviewer (validates and approves or requests fixes)
Optional Security Reviewer
One model proposes a solution, another reviews it, identifies gaps or risks, and sends feedback for refinement before final approval.
Practical Use Cases
Question:
“Is the application secure?”
One model analyzes the code, another reviews it from a security perspective and proposes improvements.
Question:
“How can we improve performance or quality?”
One model suggests architectural or algorithmic improvements, another evaluates side effects and risks.
During large refactors across multiple files, a review loop could be triggered before finalizing changes.
Two Possible Activation Approaches
Option 1: Complex Mode inside the Agent menu
Add a new option alongside Ask / Plan:
Ask
Plan
Complex Mode
When selected:
Multi-model review loop is activated
User selects models
Defines number of rounds
Defines stopping conditions (fixed rounds, consensus, no critical issues)
This keeps the UX simple and integrated within the current Cursor interface.
Option 2: Workflow Designer
Introduce a lightweight workflow builder (inspired by tools like n8n but simplified), where users can:
Add model nodes
Assign roles
Define routing logic (review → back to implementer → approve)
Configure stopping conditions
Presets could be provided:
Safe Refactor
Security Review
Performance Optimization
Architecture Review
Benefits
Reduced hallucinations
Higher confidence in large changes
Lower regression risks
Makes Cursor feel like an AI team inside the IDE
Thank you for your work, and I would be happy to provide more detailed workflow examples or help refine this concept further..