Not sure how to phrase it, but in the Codex desktop app for macOS, some messages are hidden by default, creating this confusing flow like the one below:

The question alone without any context, only until I uncollapse the dialog above:

this has been happening a lot. I understand that the model’s reasoning might be hidden by default, but I don’t think this is the case.

Either that or it’s because I’m using the obra/superpowers/brainstorming skill, which prompts for many questions and design specs before proceeding to execute the implementation, not sure if anyone else is experiencing a similar situation.

I use Codex in VS Code with the Codex extension.

I also see many of these collapsible sections. From what I can tell, they often represent the details of a subgoal while it is being processed. Once that subgoal is completed, the section may be reduced to a simpler summary line in the session.

Personally, I like this presentation. It makes it easier to look back through a goal and its subgoals without having to reread all of the intermediate details, while still leaving those details available when I need them.

I program in Prolog and have watched goals and subgoals being processed in a similar way for years, so to me this is largely business as usual. The main difference is that, instead of seeing something resembling Prolog terms or a proof trace, Codex is presenting the process as human-readable text.