Subagent delegation not working in Cursor IDE 3

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

As of the latest update to Cursor iDE 3, my orchestrator setup is not working anymore.
Before the update, the delegation worked perfectly, having a central agent that delegated work to subagents, while working through a 12-step custom pipeline, built for adding or debugging complex features.

Steps to Reproduce

Before:
/orchestrate command would start the /react-manager agent, that would delegate the 12-stage pipeline work to specialized subagents, as you can see in the attached document

Now:
/orchestrate seems to start the /react-manager, how ever, the delgation is interrupted after planning, and all the code is executed by the main agent

Expected Behavior

The /orchestrate command should start the /react-manager agent, that delegates the pipeline work to subagents

Operating System

Windows 10/11

Version Information

Cursor IDE 3.2.21

Does this stop you from using Cursor

No - Cursor works, but with this issue

Hey, thanks for the detailed report. This is a known limitation with nested subagents in the current build. It’s not a regression specifically from 3.2.21, it’s more like something that’s been there for a while and is showing up more clearly now.

What’s happening: the subagent system prompt includes a reminder telling it not to spawn additional subagents unless that’s explicitly requested. On top of that, the ability for a subagent to spawn other subagents (deep nesting like manager → subagent → more subagents) is currently limited. So your /react-manager ends up doing all the work itself instead of delegating, because the model acting as a subagent refuses to spawn further subagents.

Similar threads with the same symptom:

Workaround for now is a flat structure: have the main agent delegate directly to specialist subagents (one level of nesting), without a middle orchestrator that delegates further. So instead of main -> /react-manager -> [specialists], do main -> [specialists] directly, and split the 12-step pipeline into a series of calls from the main agent.

The issue team is aware, but I can’t share an ETA for a fix yet. If I get an update, I’ll post it in the thread.

I see, thanks for the information, i’ll rewrite my orchestrator then, and bypass the middle man, and redefine the agent as a rule. Let me know when and if this gets resolved. Have a great day.