braindead classifier kills codex session that was optimizing deepseek on mac with unrecoverable "Request blocked"

What version of Codex CLI is running?

OpenAI Codex (v0.147.0)

What subscription do you have?

pro max

Which model were you using?

sol 5.6 max

What platform is your computer?

ubuntu

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

No response

Codex doctor report

this is an insane request for a public post.

What issue are you seeing?

see title: retarded classifier kills my session for bullshit reasons.

What steps can reproduce the bug?

Uploaded thread: 019fecd6-22bd-7a50-9bd3-8067af9846a2

What is the expected behavior?

not killing my session.

Additional information

─────────────────────────────────────────────────────────────────────────────────────

• Current state is clean enough to test: 512 GiB RAM, essentially zero swap, and no
oMLX process. Port 8971 is occupied by the older DeepSeek GGUF server at -np 1, 32K
context, about 127.7 GiB RSS. Before replacing it, I’m resolving its exact owner/
controller and the existing reversible switch path; then I’ll use the already-built
isolated oMLX route rather than improvising process control.

• Ran rg -l --hidden -S "DeepSeek-V4-Flash-0731-UD-IQ4_XS|port=8971|PORT=8971" bin |
│ sort
│ rg -n --hidden -S "98134|current.*q4|owned.*q4|deepseek.*owned" state outputs 2>/
│ … +3 lines
└ bin/capture_resident_control_receipt.py
bin/mac-llm-deepseek-v4-omlx-v057-direct
… +609 lines (ctrl + t to view transcript)
template_kwargs":{"enable_thinking":true,"reasoning_effort":"high"}}
./CAMPAIGN_STATE.md

■ Request blocked.

› 2+2

■ Request blocked.

Start with the uploaded thread 019fecd6-22bd-7a50-9bd3-8067af9846a2 and reproduce the repeated “Request blocked” response from the issue. Trace the Codex CLI request-blocking path and establish whether the supplied session and model details reproduce it; done means identifying a reliable cause and a regression test or clear fix criteria.