Hi Cursor team, I’m trying to set up a workflow where Claude writes a plan, then automatically spins up a separate GPT-5.4 reviewer subagent inside Cursor to review that plan. They do back-and-forth and claude finalizes the plan.
I’ve seen subagents and custom model selection mentioned, but I’m not sure whether this is actually supported end-to-end yet. Specifically:
Can a Claude session directly trigger a subagent with a different model?
If so, what’s the correct way to configure a reviewer subagent to use GPT-5.4?
How can I accomplish this?
My goal is a simple plan-review loop:
Claude drafts the plan.
GPT-5.4 reviews it.
Claude revises based on the review.
Would appreciate any docs, examples, or confirmation on whether this is possible to do
Hey, this workflow is possible, but you’ll need to set up the orchestration via a prompt or a rule. There’s no built-in automatic plan review loop yet.
How to set it up:
Create a subagent file at .cursor/agents/reviewer.md:
---
name: reviewer
description: Reviews plans for correctness and completeness
model: gpt-5.4
---
You are a plan reviewer. Analyze the plan for gaps, risks, and improvements. Return structured feedback.
In the main Claude session, ask it to call this subagent via the Task tool after it creates a plan. For example, add a Cursor rule in .cursor/rules/plan-review.mdc so you don’t have to repeat the instruction every time.
Key points:
The Task tool supports an optional model parameter, so the parent agent on Claude can spawn a GPT-5.4 reviewer.
The review then edit loop is just multiple Task calls in a row, driven by your prompt or rule.
On the Free plan, the set of available models for subagents may be limited. If GPT-5.4 doesn’t work, that’s expected. Pro and Ultra usually have a wider selection.