尚未收集仓库信息。
尚未检测到更新。GitHub 发布版本后会显示在这里。
制作者的介绍仅有上面的一句话。 更多详情请阅读仓库 README。 仓库 README ↗
正在加载完整 README。
查看完整 README ↗Real sign-up, real login, a real database, row-level security switched on from the start. This is the foundation everything else gets built on.
supabase/schema.sql — run this once in Supabase to create the two tables this needs/signup — create an account with a screen name (never your real name), age group, email and password/login — log back in/dashboard — a protected page, redirects you to /login if you're not signed in, shows your profile and whether you've completed the personality assessment yet (not ported over yet, that's step 2)1. Run the database schema first, before deploying anything.
Go to your Supabase project, click "SQL Editor" in the left sidebar, "New Query", paste in the entire contents of supabase/schema.sql, and run it. This creates two tables, profiles and assessment_results, both with row-level security switched on so nobody can see anyone else's data.
2. Deploy the same way as before.
New empty GitHub repo, unzip this, drag everything inside the folder in one go (not the folder itself), import into Vercel, add your two environment variables in Vercel's Settings before deploying (NEXT_PUBLIC_SUPABASE_URL and NEXT_PUBLIC_SUPABASE_ANON_KEY), then deploy.
3. Test the whole loop once it's live.
Go to your URL, click "Create account", fill in a screen name and your details, submit. You should land on /dashboard showing your screen name and email. Log out, log back in with /login, you should land back on the same dashboard. That confirms accounts genuinely work.
Porting the actual personality assessment onto this, saving your type to the assessment_results table instead of asking you to retake it. After that, step 3 is the training log itself, and the fair, effort-based comparison scoring across different training styles.