正文 Markdown
It's interesting how the same model (Opus 4.8) can produce very different output whether you're using it via Claude Code or regular Claude. I find the latter still better for writing tasks, probably because it has a default prompt that's not optimized for coding.