Max CPU in Node child process of cursor-agent CLI

Where does the bug appear (feature/product)?

Cursor CLI

Describe the Bug

macOS - High CPU usage >97% in Node child process remaining running under my Terminal process after exiting Cursor CLI.

Steps to Reproduce

Run cursor-agent, do some work, exit. I think I remember at some point it stopped working - no responses, but /exit was graceful. May just be one-off, unlucky.

Expected Behavior

All subprocesses to be closed.

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Cursor CLI - 2025.09.18-7ae6800

For AI issues: which model did you use?

Auto

Additional Information

Warned by battery widget in macOS of Terminal using high power.

In Activity Monitor, the Node process is clearly the culprit. Re starting cursor-agent and exiting does not help (orphaned, I suspect). Can see it is Cursor CLI from info in Open Files and Ports.

Process has burned through all my battery and USB power from screen unable to sustain the machine.

The most frustrating thing is that the Node process starts back up after restarting Terminal! Gah!

IMPORTANT

Problem found to be tabtab running from my zshrc file.

The open files were definitely Cursor CLI files which makes me wonder whether Cursor opens sub-terminals in which to run commands, and perhaps those terminals execute zshrc which runs the bad Node process.

My point being that some problems that appear with Cursor could actually be related to the end user’s terminal config.

CLOSE? I can’t figure out how to close this issue. Sorry.

Hey, you’re right that Cursor may spawn sub‑terminals that execute your shell config, that’s a common source of confusion when debugging process issues. Thanks for sharing the solution, it’ll help other users who run into the same situation.

Thank you.

I was thinking this morning that LLMs can solve so many problems because we posted them to the web, and then we posted the solutions. I wonder whether LLMs could use the web like a shared memory, in the way that humans have, to post high value learnings and QA.

Not sure how it might work. It could just be a public place where tools post problem-solutions. Of course the output would need approval, because it might contain company secrets/IP etc.

Food for thought.