fix(connect): hard-fail OAuth callback when Redis unavailable

[hariom888]

Summary

Fixes #438.

/connect/github/callback guarded its CSRF nonce check behind if (app.redis && ...). A Redis outage silently removed the entire guard while still allowing the OAuth exchange to proceed — with userId taken directly from the attacker-controlled base64 state parameter. An attacker who could predict or observe the state format could store a token under any arbitrary user's account.

Root cause

// Before — Redis absence silently skips the entire CSRF check
const storedUserId = app.redis
  ? await app.redis.get(`oauth:nonce:${decodedState.nonce}`)
  : null;

if (app.redis && (!storedUserId || storedUserId !== decodedState.userId)) {
  // ...reject
}
// Redis down → both branches skipped → falls through to token exchange

Fix

Hard-fail immediately if Redis is not ready. The callback cannot safely proceed without nonce verification.

// After — explicit 503 gate before any token exchange
if (!app.redis || app.redis.status !== 'ready') {
  return reply.status(503).send({ error: 'Service temporarily unavailable. Please try again.' });
}

const storedUserId = await app.redis.get(`oauth:nonce:${decodedState.nonce}`);
if (!storedUserId || storedUserId !== decodedState.userId) {
  return reply.redirect(`${process.env.PUBLIC_APP_URL}/settings?error=invalid_state`);
}

await app.redis.del(`oauth:nonce:${decodedState.nonce}`);

This mirrors the strictness of the initiator (GET /github), which already unconditionally calls app.redis.set — making Redis a hard dependency for both ends of the flow.

Test coverage

Scenario	Expected
Redis status !== 'ready'	503, no token written
app.redis is null	503, no token written
Crafted state, nonce never issued	302 → invalid_state
Nonce exists but userId mismatch	302 → invalid_state
Valid round-trip	302 → connected=github, nonce consumed
Nonce replay (second request)	302 → invalid_state
Missing code or state	302 → missing_params
Malformed base64 state	302 → connect_failed

Files changed

• apps/backend/src/routes/connect.ts
• apps/backend/src/__tests__/connect.test.ts

[vercel]

@hariom888 is attempting to deploy a commit to the Prashantkumar Khatri's projects Team on Vercel.

A member of the Team first needs to authorize it.

[github-actions]

CI — Checks Failed

Backend — FAIL

Check	Result
Lint	PASS
Test	PASS
Typecheck	FAIL

Mobile — SKIP

Check	Result
Lint	-
Test	-

Web — SKIP

Check	Result
Check	-
Build	-

---

Last updated: Fri, 05 Jun 2026 11:16:33 GMT

[hariom888]

@Harxhit

Typecheck failure is pre-existing on main — not introduced by this PR

All typecheck errors are in files I did not touch:

• src/routes/team.ts — 6 errors (missing TeamRole export, implicit any, PrismaClientKnownRequestError)
• src/__tests__/team.test.ts — 1 error (missing TeamRole export)
• src/services/publicService.ts — 1 error (implicit any)
• src/utils/error.util.ts — 7 errors (PrismaClientKnownRequestError, PrismaClientValidationError, unknown type)

To confirm, run tsc --noEmit on main directly — the same 15 errors appear before any changes from this PR.

Lint: ✅ PASS
Test: ✅ PASS
Typecheck: ❌ pre-existing failure unrelated to this PR

[Harxhit]

Fix lint error in connect.ts file.