Make the world answer first
Isekai-LifeSim is not interesting just because an AI model can write prose. The key is whether each choice leaves state that the next scene can read: character status, location, tasks, and branch text need to agree.
I treat the flow as a replayable story line, not a loose chat log.
Branch rules before flavor
Branch narrative gets weak when every result is random. I need to know which variables change the ending and which variables only change tone, so the Next.js UI, TypeScript types, and prompt payload stay aligned.
When a node lacks required state, the safer move is to fall back to a conservative scene.
Next pass
The next pass should make event data easier to inspect: required flags, optional flavor text, and debug labels for why a branch appeared. After that, affection, items, and location events can become richer without making the story graph fragile.
要和 XingHuiSama 参考页一样显示 Gitalk 评论框，需要配置你自己的 GitHub OAuth App： `NEXT_PUBLIC_GITALK_CLIENT_ID` 和服务端 `GITHUB_CLIENT_SECRET`。
comments: { provider: "gitalk", owner: "user", repo: "repo", clientId: "..." }