Image generation tasks cause sustained upload saturation, network slowdown, and connection drops (Broken pipe / WebSocket fallback)

What version of Codex CLI is running?

codex-cli 0.125.0

What subscription do you have?

ChatGPT Plus

Which model were you using?

gpt-5.5 ,gpt-5.4

What platform is your computer?

Linux 6.17.0-22-generic x86_64 x86_64

What terminal emulator and version are you using (if applicable)?

Linux terminal

What issue are you seeing?

When running image generation or image-related tasks via Codex, the network becomes extremely slow and unstable.

Observed behavior:

• Upload traffic continues indefinitely and does not complete
• Upstream bandwidth becomes saturated, affecting all other network activity
• WebSocket connection frequently drops and falls back to HTTPS
• Eventually results in errors such as:

"Falling back from WebSockets to HTTPS transport. stream disconnected before completion: failed to send websocket request: IO error: Broken pipe (os error 32)"

• The session is often interrupted before task completion

System observation:

• Monitoring tools show continuous upstream traffic during the issue
• Immediately after stopping the Codex task, network usage returns to normal

This issue occurs frequently when working with image generation.

What steps can reproduce the bug?

1. Run a Codex task involving:

• image generation
• image analysis
• or referencing images in prompts

1. Observe network behavior:

• upstream traffic continuously increases
• network becomes unresponsive
• WebSocket disconnect occurs

1. Eventually, the session fails with a transport error (Broken pipe)

What is the expected behavior?

• Network usage should remain bounded and not saturate upstream bandwidth
• Large directories should not be scanned or uploaded unless explicitly required
• Image uploads or analysis should be controlled and limited
• The connection (WebSocket) should remain stable during task execution
• Tasks should complete without causing global network degradation

Additional information

Likely causes (hypothesis):

• Recursive scanning of large directories (models, outputs, node_modules)
• Automatic upload or synchronization of large files
• Excessive handling of image data without size or scope limits

Impact:

• Makes image-related workflows nearly unusable
• Causes repeated task interruption
• Severely degrades overall system usability

Suggested improvements:

• Ability to exclude directories from scanning (e.g., models/, output/)
• Limit upload size and bandwidth usage
• Explicit control over image upload/analysis behavior
• Better visibility into what files are being processed or uploaded

This issue is highly reproducible in environments with large AI/ML assets.