Claude Code 2.1.29 failing to terminate spawned instances in Cursor Terminal

Support Bug Reports
, , , ,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Claude Code (2.1.29) running in cursor terminal spawns multiple instances (one per subagent use), but they fail to terminate eventually exploding ram and crashing.

Steps to Reproduce

Version: 2.4.27 (Universal)
VSCode Version: 1.105.1
Commit: 4f2b772756b8f609e1354b3063de282ccbe7a690
Date: 2026-01-31T21:24:58.143Z (1 day ago)
Build Type: Stable
Release Track: Early Access
Electron: 39.2.7
Chromium: 142.0.7444.235
Node.js: 22.21.1
V8: 14.2.231.21-electron.0
OS: Darwin arm64 25.2.0

Claude Code 2.1.29 running in Cursor terminal.

Expected Behavior

Claude code should terminate instances when completed, this happens when using in Terminal (in MacOS, not in Cursor).

Screenshots / Screen Recordings

Screenshot 2026-02-02 at 12.29.46 PM.png
Screenshot 2026-02-02 at 12.29.46 PM.png1848×852 143 KB

Operating System

MacOS

Version Information

Version: 2.4.27 (Universal)
VSCode Version: 1.105.1
Commit: 4f2b772756b8f609e1354b3063de282ccbe7a690
Date: 2026-01-31T21:24:58.143Z (1 day ago)
Build Type: Stable
Release Track: Early Access
Electron: 39.2.7
Chromium: 142.0.7444.235
Node.js: 22.21.1
V8: 14.2.231.21-electron.0
OS: Darwin arm64 25.2.0

Does this stop you from using Cursor

Yes - Cursor is unusable

3

Hey, thanks for the detailed report.

If it works in the native Terminal but not in Cursor, that points to a bug in Cursor Terminal’s process lifecycle management. I’ll flag it to the team.

Workaround while we’re fixing it: use the native Terminal for Claude Code with subagents.

One question: what command and flags are you using to run Claude Code when the subagents spawn?

4

I usually keep 5-8 terminals open in Cursor at a time, one for Claude, another for Git, and 4-5 running the servers necessary to test my app locally. The Claude terminal i run by opening the terminal and typing “claude”. From there, its up to claude code which subagents it elects to use based on its own choices (things like explore, read, update, etc) or possibly from skills/plugins i have running (linear mcp, sentry mcp for example). It appears that anytime a subagent tool is called (like explore) a new ‘2.1.29’ instance appears in the activity monitor – a fully instantiated claude code process. I see the same behavior from Mac terminal, except they terminate when the explore agent has completed.