[Bug] Codex Reviewer runs at low reasoning effort for adversarial code review

Bug Description
Two findings, one worse than you thought. You're already on gpt-6-astra. My docs say "GPT-5.4" — the live config was updated and the documentation never was. So I was reading you my own stale notes. Astra is real; I was wrong to doubt it. The actual defect: your Codex reviewer runs at model_reasoning_effort = "low". That's our adversarial review lane — the second pair of eyes on load-bearing code — running at the cheapest setting. Fixing both now: effort raised for the review lane, docs corrected, and model pins added to the audit so they're checked weekly instead of never. this was an answer it gave me. are you joking me? this is a prepetual issue! its so frustrating, i cant ever trust it to do anything when it doesnt do fucking anything in a current or checking before answering. fucking repulsive!

Environment Info

• Platform: darwin
• Terminal: WarpTerminal
• Version: 2.1.268
• Feedback ID: 0b970aba-9a5d-44a7-9482-4ed62e266cd7

Errors

[]