I had a similar experience, but I have a different conclusion. I used GitHub Copilot (with Claude Sonnet/Opus) until they made their horrific usage model change. I used a PRD skill and the plan feature was great. It asked me good questions which I didn't think about during my initial prompt. Then I switched to Claude Code. The model's capabilities felt impressive. It also asked me a few questions (but way less and only once/twice) in plan. But when reviewing the code, I found weird architectural/data flow decisions which just didn't make sense and it didn't really disclose them in beforehand.

My initial thought was to improve architecture documentation, so the model can read and update it and stops bolting on new features without consideration for the whole project. It did not help.

I'm now testing/comparing Codex and it found my old PRD skill from GitHub CoPilot. When I applied that to Claude Code, I now get similar good results. So my conclusion is: Yes, Opus 5 is bold by default, but you can tell it to be more unsure and get good results too.