I found that the Claude Code support sub-agent which could manage the different context such as code-review, architechure. So all the agents work like a agent team.
But now I use them with different prompt in the main context, that is totally different.
So will Cursor support this? thank you.
Codex community is also working on this feature in PR 3655.. I think it will definitely be released within a month.. It would be great to have some hype on this for Cursor
I think native subagent support is a very desirable feature to have in Cursor.
In fact, I just bought a Claude Pro subscription to have Claude Code and the subagents feature. I still prefer the Cursor workflow, but this feature is powerful.
I bump this request subagents are hugew!
It’s strange that this isn’t gaining more traction. Honestly I think it’s all about subagents now. fanning out work to specialized agents that only ask questions when needed. You can like simulate agent type .md prompts via the /commands in cursor, but it’s not exactly the same. especially it’s not starting work in parallel or allows taking influence on the coordinator/planner agent that would control the required subagents. Definitely a BIG MISS for cursor and makes me use VSCode with copilot or Claude Code a lot more again.
Maybe this will help speed things up
We’d also love subagents in Cursor!
The whole .cursor structure should be expanded.. agents and the ability to pass a task between them, would be awesome!!! Especially when those agents can have their tools locked like the old custom mode agents
I would find this very useful.
Hey about sub-agents I did a small test on using cursor-cli to enable cursor editor to dispatch parallel sub-agents and it seems to work, only thing I did was add this to my user rules in the cursor settings:
”WHEN I ASK FOR THEM, you can call sub-agents by running the cursor CLI like this:
cursor-agent -p "PROMPT HERE" --output-format text
Keep the sub agents prompts short for now.
IMPORTANT: if you need to use sub-agents for modifications you need to add the --force parameter.”
It triggers the cursor-agent cli in non-interactive mode, and once done it just returns the result as text, so the main agent in the IDE can parse the results. Since Claude Code also support custom prompts for the agents I replicated that using commands, so for example you can have a /testing command, and then ask cursor to trigger the agent using that command, and it will be equivalent to having a specific sub-agent. You can even ask it to use specific models as well if you add that to the instructions
It s available in cursor IDE as well. A configuration is needed:
Go to cursor seettings → Beta → change update access to Nightily.
Restart cursor. After restart a new update request will popup. Update it.
After this update, go to cursor settings → now the rules and commands menu will change to: rules, subagents, commands. access it →scroll down to see subagents menu.
Can enterprise users test this?
