[VS Code extension] GPT-5.4 sometimes outputs unrelated lottery text and leaked internal-looking fragments

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

vscode codex extension

What subscription do you have?

plus

What platform is your computer?

Ubuntu

What issue are you seeing?

Codex (GPT-5.4) occasionally outputs obviously unrelated garbage text during a normal conversation.

Instead of answering the prompt, it suddenly inserts random phrases such as:

• "中彩票"
• "天天中彩票"
• "彩票争霸"
• mixed fragments like "json code", "analysis to=web", "commentary to=..."

This looks like output contamination / prompt leakage / internal tool text leaking into the final answer.

The issue is intermittent, but it has happened multiple times. When it occurs, the response becomes unusable and unrelated to the actual request.

I’ve attached a screenshot showing an example.

<img width="382" height="155" alt="Image" src="https://github.com/user-attachments/assets/76464797-cdc9-4bf4-8300-502477dc09ff" />

What steps can reproduce the bug?

long conversation

What is the expected behavior?

_No response_

Additional information

_No response_