My cursor IDE malfunctions when using the auto model to call subagents. It prompts that the task does not exist.
Steps to Reproduce
I chose the auto mode and then directly called the subagents (one of my bug-fixing subagents - /bug-hunter).
like：/bug-hunter
review：You entered /bug-hunter, indicating that you want to use the bug-hunter sub-agent to do a bug check. I don’t have any available Task tools on my side, so I can’t transfer the task to the ‘bug-hunter’ sub-agent. The invocation of sub-agents is generally handled internally by Cursor’s Agent process, and I don’t have access to this invocation entry in the current session.
Expected Behavior
Here, I should directly call my sub-agent for bug reviews. I’m even on the Ultra plan, but this problem still occurs.
The team is aware and is already looking into it. There’s no ETA yet, but your report helps us prioritize.
A couple things to try:
Start a new chat and try calling the subagent again
Try setting a specific model instead of “auto” to check if this is only happening in auto mode
Quick question: was Privacy Mode enabled when you got that Request ID? If yes, please temporarily turn it off, reproduce the issue in a new chat, and grab a new Request ID. That’ll help the team dig deeper into the logs.
One more question: do the built-in subagents (explore, bash, browser) work for you, or is the issue only with custom subagents?
Hello, a few days ago (about five days ago) when I was using it, I was using sonnet4.5, and it could run and even create subagents on its own. At that time, privacy mode was enabled.
It seems that the built-in subagents cannot be used in the current auto mode either, ID: 2ddd393a-2153-4725-bce6-34a3b00d19ec.
But I don’t have any planned balance for this month, so I can’t test whether sonnet4.5 can use subagents at the moment, sorry.
Thanks for the extra details. The fact that built-in sub-agents also don’t work in Auto mode is important, I’ll pass that on to the team.
About it working with Sonnet 4.5 five days ago, that’s a useful insight. The issue might be related to how Auto mode picks the model and wires up the Task tool.
If you get a chance to test with a specific model (not Auto), it’d be great to hear the result, but it’s not critical.
I was looking into this problem more and selecting different models. The Task tool is not loaded when I select “Composer 1” as a model. This is probably why it doesn’t work in “Auto” mode either
Thanks for the extra diagnostics, it really helps.
You’re right: Composer 1 doesn’t load the Task tool, so subagents don’t work. When you pick Auto mode, the system selects Composer 1, and that’s why you’re seeing the issue.
Temporary workaround: pick a specific model instead of Auto (Sonnet 4.5, GPT-5.2, Opus 4.5, etc.) and subagents should work.
But there is another issue- I documented my sub-agent to inherit the model when delegated. However, the sub-agent model defaults to Composer 1 no matter what I’ve specified the model on.
Hi @Colin , I just checked in 2.5.17 and this issue is still not fixed. When the agent window is using Composer or Auto, it does not have access to the Task tool and cannot trigger subagents. When the agent is set to other models (e.g., Sonnet), it works correctly. See screen shots below.
This is a cost issue, since requiring the agent to be in Sonnet burns up the API budget on my plan.
), which is around the corner!
