going through some analyses of user traces and the absolute worst codex tic is 100% the

user: did you commit and push

codex: no i did not

OR

user: did you use the skill i told you to

codex: no i didn't use the skill. the next step is [complete nonsense]

say what you want about talking to claude but i do miss it's ability to read between the lines on this one, the claude version is consistently

claude: "no i didn't, lemme do that rq for you"