回顾与练习

关键要点

• 上线是一套工作流，而不是一个瞬间：部署、观察、修补、重复 — 并且把回滚做成你已经测试过一次的东西。
• 在宣布任何东西之前先过一遍上线前检查清单，让 AI 逐项核验，而不要依赖你的记忆。
• 让测量来驱动优化 — Lighthouse 报告永远胜过"感觉很慢"的直觉。
• 密钥放在生产配置里，绝不放进仓库；.env 加入 gitignore，.env.example 保持最新。
• 第一个版本是假设，真实用户才是实验，所以要用小而快的修复来迭代。

动手试试

挑一个你已部署的应用(或现在就部署一个很小的),逐行对照走一遍完整的上线前检查清单。对每一项你无法诚实勾选的，把失败的那一项贴给 AI，请它给出弥补差距的最小改动。最后点击线上站点的一个深层链接 — 而不只是首页 — 确认生产环境的行为和开发环境一致。

本章提示词

Here is my deployed app and its stack: [describe app + hosting + services].
Act as my launch reviewer. Walk this pre-launch checklist item by item
and, for each, tell me how to verify it on the LIVE site (not in dev):
- production URL loads over valid HTTPS on the custom domain
- all secrets set in production; none committed to the repo
- core happy path (sign-up, login, main action) works in production
- payments tested end-to-end including the webhook (if any)
- error tracking and uptime monitoring are live
- I can roll back, and have tested it once
- mobile layout works on a real phone
- database backups are enabled
For every item that fails, give me the SMALLEST change that fixes it.
Do not assume anything passes — make me prove each one.