No auth in apps or APIs — login is a separate project

Apps/APIs 100% public, login redirects to dedicated login-manager

No auth in apps or APIs — login is a separate project

Decision

Apps and APIs across the fleet are 100% public. No sign-in UI, no session check, no auth SDK imports in any app shell or API handler. Login functionality moves to a dedicated login-manager project (separate repo, separate subdomain) — apps that need an authenticated user redirect to it and never embed the auth flow.

Why

• Simpler shells: every app drops its auth wiring, session middleware, and user-context plumbing — the shell is just the tool.
• Kill cross-repo coupling: no more shared @oriz/account-widget package, no Clerk SDK pinned across 18+ apps, no JWT verifier worker in every API.
• Donations-only is coherent without accounts: donors don’t need a logged-in identity to give via BuyMeACoffee / GitHub Sponsors / UPI — auth was solving a problem we don’t have.
• No paid users yet: embedded auth is a maintenance tax with zero revenue against it. The minute we charge, login-manager spins up — until then, friction = lost users.
• Friction-free for visitors: the fleet’s pitch is “free public tools” — a sign-in wall (even optional) signals lock-in and contradicts the build-gate-vs-paywalled-competitors stance.

What this kills

• The Clerk JWT pattern locked an hour ago (auth-clerk-emergency-migrate-2026-06-25).
• login.oriz.in + account.oriz.in subdomain plan.
• The repos/infra/auth integration repo (mark for deletion or repurpose as the login-manager skeleton).
• AccountWidget React island in the template.
• apps.ts AccountWidget references.
• The @oriz/account-widget package (if it exists; drop from dependency arrays).

What stays

• The 24 npm packages — none of them assumed auth in the first place.
• The 18 APIs — already public per Spark plan; just confirm no auth code crept in.

Login-manager project (TBD)

• Lives at TBD subdomain (probably one of: login.oriz.in, accounts.oriz.in, id.oriz.in).
• Hosted as its own repo (probably repos/infra/login-manager/ — to be created).
• Apps that need login redirect: https://login.oriz.in/?return=<encoded-url> ? callback to app’s /auth/callback route.
• Decision deferred: which backend (still TBD, could be Clerk, Lucia, or self-rolled JWT).

Implications

• Audit pass: grep every app and API for Clerk/Firebase SDK imports, session middleware, JWT verifier code — strip on sight.
• Template repo: remove the auth island and any useUser() hooks from the Astro+React baseline.
• repos/infra/auth repo: either delete or rename to login-manager as the skeleton for the future separate project.
• Rate limiting on APIs: switch to IP-based throttling (per oriz-rate-limit-npm-pkg) since there’s no API key check.
• Donation footers stay anonymous — no “logged-in donor” tracking.
• When the login-manager project ships, it’ll be a single OAuth/redirect surface; apps never embed it, only redirect to it.

Cross-refs

• Supersedes: auth-clerk-emergency-migrate-2026-06-25
• Supersedes:
• Related rule:
• Related rule: no-card-on-file
• Related memory: [[no-auth-in-apps-or-apis]], [[donations-only-no-pro-no-ads]]

Edit on GitHub · Back to index