openai / openai/codex

[macOS] ChatGPT 26.818.41705 still enters unbounded SkyComputerUseService spawn storm, freezes Mac, and crashes

Open
#40,153 8 comments 2 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

app bug computer-use performance
Dominant language
Rust
Stars
125k
Forks
19.4k
PR merge metrics
PR metrics pending

Description

What version of the Codex App are you using (From “About Codex” dialog)?

[macOS] ChatGPT 26.818.41705

What subscription do you have?

ChatGPT Pro

What platform is your computer?

macOS 26.5.2 (Build 25F84), Apple Silicon arm64 — Darwin 25.5.0

What issue are you seeing?

a/.hermes/workspaces/operator/chatgpt-codex-spawn-storm-report.md → b/.hermes/workspaces/operator/chatgpt-codex-spawn-storm-report.md
@@ -0,0 +1,125 @@
+# [macOS] ChatGPT 26.818.41705 still enters unbounded SkyComputerUseService spawn storm, freezes Mac, and crashes
+
+## Summary
+
+The unified ChatGPT/Codex desktop app on macOS repeatedly creates SkyComputerUseService helpers without an apparent bound. The helper population rapidly consumes several gigabytes of RAM, drives system load extremely high, makes the Mac unresponsive, and ends with a ChatGPT crash.
+
+This reproduces while the app is idle; Computer Use does not need to be invoked manually.
+
+The behavior appears to match open reports such as #38760, #38841, and #39151, but it still reproduces on the newer desktop build 26.818.41705 (build 6971).
+
+## Environment
+
+- ChatGPT desktop app: 26.818.41705 (build 6971)
+- Crashing predecessor observed immediately beforehand: 26.818.41509 (build 6962)
+- Bundle identifier: com.openai.codex
+- macOS: 26.5.2 (25F84)
+- Architecture: Apple Silicon / ARM64
+- Hardware model: Mac16,10
+- Memory: 16 GiB
+- Codex Framework: 151.0.7922.170
+
+## Steps to reproduce
+
+1. Fully terminate ChatGPT and all existing Computer Use helper processes.
+2. Launch /Applications/ChatGPT.app normally.
+3. Do not invoke Computer Use; leave the app open.
+4. Observe the process population and resource use.
+5. SkyComputerUseService instances begin accumulating within seconds.
+6. The UI becomes increasingly unresponsive and ChatGPT eventually crashes.
+
+## Observed results
+
+### First live observation
+
+- SkyComputerUseService: 81 processes
+- Combined helper RSS: approximately 4.5 GiB
+- System load average: approximately { 24.08 26.03 29.96 }
+- Multiple additional Codex renderer, security, Node, and Crashpad processes were present.
+
+### Clean restart verification
+
+After fully terminating the ChatGPT/Computer Use process tree and reopening the app:
+
+- Within approximately 8 seconds: 13 SkyComputerUseService processes
+- Combined helper RSS: approximately 1.0 GiB
+- Combined helper CPU: approximately 146%
+- Main ChatGPT RSS: approximately 800 MiB
+- The process population continued increasing, so the app was terminated to protect the host.
+
+### Subsequent reproduction
+
+On another launch:
+
+- SkyComputerUseService: 110 processes
+- Combined helper RSS: approximately 7.5 GiB
+- Codex renderers: approximately 2.2 GiB combined RSS
+- Main ChatGPT process: approximately 880 MiB RSS
+- Load snapshot: { 49.82 94.13 69.94 }
+- Peak recent load observed during cleanup: { 95.70 101.35 72.89 }
+- The system had begun swapping heavily and the entire Mac was becoming unresponsive.
+
+Force-terminating the ChatGPT/Computer Use process tree immediately released substantial memory.
+
+## Crash evidence
+
+Two macOS diagnostic reports were generated within minutes:
+
+- ChatGPT-2026-08-22-171729.ips
+- ChatGPT-2026-08-22-171923.ips
+
+Common or relevant fields:
+
+- App version: 26.818.41509
+- Build: 6962
+- bug_type: 309
+- Faulting/triggered thread name: computer-use
+- Work queue limit recorded: Dispatch Thread Soft Limit: 64
+
… omitted 47 diff line(s) across 1 additional file(s)/section(s)

What steps can reproduce the bug?
  1. Fully quit ChatGPT and terminate any existing SkyComputerUseService helper processes.
  2. Launch /Applications/ChatGPT.app normally.
  3. Leave the app open without invoking Computer Use or starting a Computer Use task.
  4. Open Activity Monitor and search for SkyComputerUseService.
  5. Observe that new SkyComputerUseService processes appear continuously within seconds.
  6. Continue observing as the helper count, memory consumption, and system load rise rapidly.
  7. The ChatGPT UI becomes unresponsive and eventually crashes.

On my Mac, a clean restart produced 13 SkyComputerUseService processes using about 1 GiB within approximately eight seconds. A subsequent occurrence reached 110 helpers using about 7.5 GiB, with system load reaching approximately 95–101.

What is the expected behavior?

ChatGPT should maintain a single bounded Computer Use service lifecycle and remain stable while idle.

If the Computer Use helper cannot initialize or complete its bootstrap handshake, the app should use a retry limit, exponential backoff, process cap, and circuit breaker. Failed helpers should be terminated and reaped.

Leaving ChatGPT idle must not continuously create helper processes or worker threads, exhaust memory and process limits, freeze the UI or operating system, or crash the application.

Additional information

No response

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the reproduction details in .hermes/workspaces/operator/chatgpt-codex-spawn-storm-report.md and reproduce the idle launch while monitoring SkyComputerUseService in Activity Monitor. Trace the Computer Use startup path responsible for the helper population; done means idle ChatGPT keeps a bounded helper lifecycle without runaway memory, load, freezes, or crashes.

Written by the indexing model from the issue text.

Assessment

Tech stack
macos
Domain
desktop, performance
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Needs clarification
Newbie friendliness
28/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.