I think Claude Code itself is mostly vibe coded if you consider that impressive. In general I’m not aware of any vibe coded projects with a substantial amount of iteration put into it rather than being close to one shot, which is why I shared mines. Any project someone can one shot is also something anyone else can vibe code for themselves.
What definition of "vibe coding" are you using here? I seriously doubt Claude Code was made "mostly" (or even "partly") by telling a prompt what they want, and accepting the output when it looks functionally acceptable without regard to how the code looks or works under the hood.
I don't know if or how you've been using AI for coding, but Claude Code is almost certainly being used to develop itself, and I have a hard time imagining their devs not having the best setup possible to make it as easy as possible for Claude Code to iterate on itself with minimal oversight. Once a codebase is well structured and you supply a well thought out meta prompt with something like CLAUDE.md, then there's no reason to check the code output/changes beyond a quick skim.
