Wow, I tried GPT 5.5 Extra High for the first time. Used almost all my 5 hour window in 35 mins. Thinking it might be time to upgrade soon. Honestly the $20 plan is absurdly generous but I want more.

I never even mentioned that the dev server was running in tmux. Was not expecting Codex to be so helpful

Finally figured out why @mattpocockuk's skills are always in my model context. Codex doesn't honour the disable-model-invocation: true flag. Instead Codex uses openai.yaml with policy: allow_implicit_invocation: false See: developers.openai.com/codex/…