I setup multiple layers of subagents in a way which
subagent (orchestrator) → subagents → finer subagents.
The way I typically invoke it was through the chat window on IDE and simply say
run /orchetrator-subagent with …
I started to explore Cursor CLI as it’s less resource intensive but
Steps to Reproduce
Call agents that spawn subagents. works in IDE, fails on CLI.
Getting an error “It immediately failed because that sequencer requires spawning additional internal subagents, and this session does not expose that subagent-spawning interface.”
Basic subagent support in the CLI was added recently, but nested subagents, where a subagent spawns other subagents, still don’t work in the CLI. The subagent CLI session doesn’t get the interface needed to start child subagents, so you hit that error.
I’ve passed this to the team. No ETA yet, but your report helps us prioritize it.
Workarounds for now:
For nested setups, please use the IDE like you did before.
Or try restructuring into a flat setup where the orchestrator dispatches all tasks directly via the Task tool, without a middle layer.
Also just noticed this problem, which breaks CLI agents compared to local ones.
For instance, having CLI agents that do auto-implementation and then invoke a reviewer sub-agent, who then invokes multiple sub-reviewers, effectively breaks here.
The same goes for agent teams that do work.
This puts Cursor very far behind things like Claude Code CLI.
afaik claude cli wasn’t allowed to have nested sub-agents as well, the only way is bypass through plugin. is that what you’ve been doing as well? I’d switch over if there’s an official way to do it in claude.
You know that’s actually a really good call out. I use Claude Code in the workspace at work more often than I use it at home, and invoking sub-agents works in our workspace monorepo because we have a plug-in that enables it, by the looks of it. After reading the Claude Code docs, Claude Code does not actually support sub-agents invoking sub-agents natively.