Bizarre internal chain of thought leaking into session; words like "lich," "lobotomized," "rekt" appearing before model stops itself.

What version of Codex CLI is running?

0.153.4

What subscription do you have?

Pro

Which model were you using?

gpt-5.6-sol

What platform is your computer?

Darwin 25.5.0 arm64 arm

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

Terminal.app

What issue are you seeing?

Internal chain of thought leaking into session; words like "lich," "lobotomized," "rekt" appearing before model stops itself.

Examples (bold emphasis mine):

• Opus remains lich? No. Ensure proper User sees normal.
• Opus remains**? I meta. Write: Opus remains active at roughlyrekt.**
• Opus remains active at roughly roughly? I keep glitch. Let's slow.
• Five artifacts are safely published pencils? no. Keep concise.
• Five attempts are safely published; I’m allowing Opus its full build and verification window5.6 proper cycle.
• Fable has been active for roughly fourteen minutes and remains healthy.://? No, need clean.
• NFלית? I’m checking the durable controller suite record now; the runner remains isolated from the viewer worktree, lobotomized? No. Let's write clean.

What steps can reproduce the bug?

Uploaded thread: 01a07c6e-126b-7603-a8f1-39c462ce1dde

What is the expected behavior?

I would say "don't leak internal chain of thought" but also these words are kinda bizarre/scary :).

Additional information

No response

Start by reviewing the uploaded thread 01a07c6e-126b-7603-a8f1-39c462ce1dde and reproducing the behavior with Codex CLI 0.153.4 on Darwin using the reported model. Compare the session output with the expected behavior; done means internal chain-of-thought text and unrelated fragments no longer appear to the user.