fix(auth): surface the silent /auth/me store-failure that bounces login to signin (#3734)

app/src/utils/__tests__/desktopDeepLinkListener.test.ts

@@ -10,6 +10,7 @@ import {
 } from '../../store/deepLinkAuthState';
 import { getStoredCoreMode } from '../configPersistence';
 import {
+  classifyAuthStoreFailure,
   registerAuthDeepLinkState,
   setupDesktopDeepLinkListener,
 } from '../desktopDeepLinkListener';
@@ -227,6 +228,43 @@ describe('desktopDeepLinkListener', () => {
     expect(state.errorMessage).toBe('Sign-in failed. Please try again.');
   });
 
+  it('injection #1: store-time /auth/me failure bounces to signin — no session applied, no /home nav', async () => {
+    // Root-cause hypothesis: `auth_store_session` validates the JWT against the
+    // backend GET /auth/me BEFORE persisting (credentials/ops.rs). If that call
+    // errors/times out, store_session returns Err → applySessionToken rethrows →
+    // the session is NEVER persisted and the login event NEVER fires, so the user
+    // stays on the signin page even though OAuth "succeeded".
+    vi.mocked(storeSession).mockRejectedValueOnce(
+      new Error('Session validation failed (GET /auth/me): 503 Service Unavailable')
+    );
+
+    // The `core-state:session-token-updated` event is the ONLY trigger that drives
+    // CoreStateProvider → refresh → authenticated React state. If it never fires,
+    // the app cannot leave the signin page.
+    const sessionTokenUpdated = vi.fn();
+    window.addEventListener('core-state:session-token-updated', sessionTokenUpdated);
+    window.location.hash = '#/'; // reset any prior test's navigation
+
+    try {
+      vi.mocked(getCurrent).mockResolvedValue([authDeepLinkWithState('token=abc&key=auth')]);
+      await setupDesktopDeepLinkListener();
+      await waitForAuthSettled();
+
+      // store WAS attempted (we reached the persistence call)...
+      expect(storeSession).toHaveBeenCalledWith('abc', {});
+      // ...but it FAILED, so the session-applied event was never dispatched...
+      expect(sessionTokenUpdated).not.toHaveBeenCalled();
+      // ...and we never navigated to /home (ProtectedRoute/PublicRoute keep signin).
+      expect(window.location.hash).not.toBe('#/home');
+      // Surfaced as the generic toast; processing cleared.
+      const state = getDeepLinkAuthState();
+      expect(state.errorMessage).toBe('Sign-in failed. Please try again.');
+      expect(state.isProcessing).toBe(false);
+    } finally {
+      window.removeEventListener('core-state:session-token-updated', sessionTokenUpdated);
+    }
+  });
+
   it('does not make the E2E deep-link helper wait for auth readiness', async () => {
     let resolveReadiness!: (_value: { ready: true }) => void;
     waitForOAuthAuthReadiness.mockReturnValueOnce(
@@ -320,3 +358,37 @@ describe('desktopDeepLinkListener', () => {
     expect(suppressEvents[suppressEvents.length - 1].until).toBe(0);
   });
 });
+
+describe('classifyAuthStoreFailure', () => {
+  it.each([
+    ['Session validation failed (GET /auth/me): operation timed out', 'auth_me_timeout'],
+    ['error sending request: deadline has elapsed', 'auth_me_timeout'],
+    ['GET /auth/me failed (401 Unauthorized): bad token', 'auth_me_unauthorized'],
+    ['Session validation failed (GET /auth/me): 503 Service Unavailable', 'auth_me_gateway'],
+    ['upstream returned 502 Bad Gateway', 'auth_me_gateway'],
+    ['fetch failed: ECONNREFUSED', 'network'],
+    ['Session validation failed (GET /auth/me): something odd', 'auth_me_other'],
+    ['totally unrelated explosion', 'other'],
+  ])('classifies %j as %s', (message, expected) => {
+    expect(classifyAuthStoreFailure(message)).toBe(expected);
+  });
+
+  // Contract pin: the classifier matches substrings of the Rust-produced error
+  // (credentials/ops.rs: `Session validation failed (GET /auth/me): {reason}`,
+  // with {reason} from rest.rs `GET /auth/me failed ({status}): {text}` or a
+  // reqwest transport error). If Rust rewords that prefix, these must fail CI
+  // rather than letting an arm silently degrade to 'other'.
+  it('pins the real Rust store_session failure strings to meaningful kinds', () => {
+    const gateway =
+      'Session validation failed (GET /auth/me): GET /auth/me failed (503): {"error":"unavailable"}';
+    const timeout =
+      'Session validation failed (GET /auth/me): error sending request for url (https://api.tinyhumans.ai/auth/me): operation timed out';
+    const bare = 'Session validation failed (GET /auth/me): something unexpected';
+
+    expect(classifyAuthStoreFailure(gateway)).toBe('auth_me_gateway');
+    expect(classifyAuthStoreFailure(timeout)).toBe('auth_me_timeout');
+    // The bare prefix is still recognized via the auth/me anchor — NOT 'other'.
+    expect(classifyAuthStoreFailure(bare)).toBe('auth_me_other');
+    expect(classifyAuthStoreFailure(bare)).not.toBe('other');
+  });
+});

app/src/utils/desktopDeepLinkListener.ts

@@ -265,6 +265,25 @@ const handleAuthDeepLink = async (parsed: URL, requireStateNonce = true) => {
         { requiresAppDataReset: true }
       );
     } else {
+      const kind = classifyAuthStoreFailure(rawMessage);
+      // Capture a SYNTHETIC error keyed only by `kind` — never the raw error.
+      // Two reasons (both raised in review):
+      //  1. PII: the upstream `/auth/me` failure embeds the verbatim backend
+      //     response body (`rest.rs`: `GET /auth/me failed ({status}): {text}`),
+      //     and `beforeSend` does NOT scrub `exception.values[].value`. Severing
+      //     the message (vs. scrubbing) guarantees no body/email/token-adjacent
+      //     text ships.
+      //  2. Timeout shape: a hang surfaces as `CoreRpcError(kind='timeout')`,
+      //     which `beforeSend` drops via `isCoreRpcTimeoutError(originalException)`
+      //     BEFORE our tag applies. A plain `Error` makes `originalException`
+      //     non-matching, so the lead cause finally reaches Sentry.
+      // The PII-free `kind` tag + stable fingerprint are all we need to group.
+      Sentry.captureException(new Error(`auth store failed: ${kind}`), {
+        level: 'error',
+        tags: { surface: 'react', phase: 'deep-link-auth-store', auth_store_failure: kind },
+        fingerprint: ['deep-link-auth', 'session-store-failed', kind],
+      });
+      console.warn('[DeepLink][auth] session store failed — staying on signin (kind=%s)', kind);
       failDeepLinkAuthProcessing('Sign-in failed. Please try again.');
     }
   }
@@ -279,6 +298,28 @@ const isDecryptionFailure = (message: string): boolean => {
   );
 };
 
+/**
+ * Classify a sign-in *store* failure into a short, PII-free kind. A store-time
+ * `/auth/me` failure (esp. a timeout) is the lead root cause of "OAuth succeeded
+ * but the app is back on the login page", yet it currently emits NO Sentry signal
+ * on any layer: the FE has no console-capture integration, the Rust core drops
+ * `"timeout"`/408/504 as transient (`observability.rs`), and the backend only
+ * pages genuine 500s (`shouldHandleError: status === 500`, BACKEND-ALPHAHUMAN-40)
+ * — so gateway/timeout 5xx never reach Sentry. Tagging the kind here is the one
+ * place the bounce becomes debuggable. Returns a stable enum-like string (no URLs,
+ * no tokens) safe to use as a Sentry tag / fingerprint.
+ */
+export const classifyAuthStoreFailure = (message: string): string => {
+  const m = message.toLowerCase();
+  if (/timed out|timeout|operation timed out|deadline/.test(m)) return 'auth_me_timeout';
+  if (/\b401\b|unauthorized/.test(m)) return 'auth_me_unauthorized';
+  if (/\b50[234]\b|bad gateway|service unavailable|gateway timeout/.test(m))
+    return 'auth_me_gateway';
+  if (/network|fetch failed|connection|dns|unreachable/.test(m)) return 'network';
+  if (/auth\/me|session validation failed/.test(m)) return 'auth_me_other';
+  return 'other';
+};
+
 /**
  * Handle `openhuman://payment/success?session_id=...` deep links.
  * Fired when a Stripe checkout session completes and the browser redirects

src/openhuman/credentials/ops.rs

@@ -151,10 +151,26 @@ pub async fn store_session(
             .ok_or_else(|| "local session requires a user payload".to_string())?
     } else {
         let client = BackendOAuthClient::new(&api_url).map_err(|e| e.to_string())?;
-        client
-            .fetch_current_user(trimmed_token)
-            .await
-            .map_err(|e| format!("Session validation failed (GET /auth/me): {e:#}"))?
+        match client.fetch_current_user(trimmed_token).await {
+            Ok(user) => user,
+            Err(e) => {
+                // This is the store-time validation gate: if it fails the profile
+                // is NEVER persisted, so the user bounces straight back to the
+                // signin page after a "successful" OAuth. Timeouts/gateway 5xx are
+                // otherwise dropped by the Sentry transient classifier, so log an
+                // explicit, grep-friendly WARN to the app log regardless.
+                let reason = format!("{e:#}");
+                tracing::warn!(
+                    domain = "credentials",
+                    operation = "store_session",
+                    "[credentials][auth-store] GET /auth/me validation FAILED on {} — session NOT persisted; user will bounce to signin: {reason}",
+                    api_url.trim_end_matches('/')
+                );
+                return Err(format!(
+                    "Session validation failed (GET /auth/me): {reason}"
+                ));
+            }
+        }
     };
 
     let mut metadata = std::collections::HashMap::new();

tests/app_state_credentials_raw_coverage_e2e.rs

@@ -8,6 +8,7 @@ use openhuman_core::openhuman::app_state::{
     StoredOnboardingTasks,
 };
 use openhuman_core::openhuman::config::rpc as config_rpc;
+use openhuman_core::openhuman::credentials::ops::store_session;
 use openhuman_core::openhuman::credentials::profiles::{
     AuthProfile, AuthProfileKind, AuthProfilesStore, TokenSet,
 };
@@ -188,6 +189,89 @@ async fn auth_me_server(
     (url, task, shutdown_tx)
 }
 
+/// Like `auth_me_server` but always replies HTTP 500, so `store_session`'s
+/// `GET /auth/me` validation gate fails — exercising the WARN + `Err` path that
+/// leaves the session unpersisted (the "OAuth succeeded but app is back on the
+/// signin page" bug).
+async fn auth_me_failing_server() -> (
+    String,
+    tokio::task::JoinHandle<()>,
+    tokio::sync::oneshot::Sender<()>,
+) {
+    let listener = tokio::net::TcpListener::bind(("127.0.0.1", 0))
+        .await
+        .expect("bind failing auth/me listener");
+    let url = format!("http://{}", listener.local_addr().expect("listener addr"));
+    let (shutdown_tx, mut shutdown_rx) = tokio::sync::oneshot::channel::<()>();
+    let task = tokio::spawn(async move {
+        loop {
+            tokio::select! {
+                _ = &mut shutdown_rx => break,
+                accepted = listener.accept() => {
+                    let Ok((mut stream, _)) = accepted else { break; };
+                    let mut req = [0_u8; 2048];
+                    let _ = stream.read(&mut req).await;
+                    let body = "{\"error\":\"mock /auth/me 500\"}";
+                    let response = format!(
+                        "HTTP/1.1 500 Internal Server Error\r\ncontent-type: application/json\r\ncontent-length: {}\r\nconnection: close\r\n\r\n{}",
+                        body.len(),
+                        body
+                    );
+                    let _ = stream.write_all(response.as_bytes()).await;
+                    let _ = stream.shutdown().await;
+                }
+            }
+        }
+    });
+    (url, task, shutdown_tx)
+}
+
+/// Store-time `/auth/me` failure must surface as `Err` (and NOT persist a
+/// profile), which is what bounces the user back to signin after a "successful"
+/// OAuth. Covers the WARN/`Err` gate added in `credentials::ops::store_session`.
+#[tokio::test]
+async fn store_session_auth_me_failure_returns_err_and_does_not_persist() {
+    let _lock = env_lock();
+    let (api_url, server_task, shutdown_tx) = auth_me_failing_server().await;
+    let harness = setup(&api_url);
+    let config = harness.config().await;
+
+    // Non-local token (3 dot-parts, not ".local") forces the backend
+    // `GET /auth/me` validation path inside `store_session`.
+    let result = store_session(&config, "header.payload.signature", None, None).await;
+
+    let _ = shutdown_tx.send(());
+    let _ = server_task.await;
+
+    let err = result.expect_err("store_session must fail when GET /auth/me returns 500");
+    // Lock the cross-layer error-string contract: this exact prefix is what the
+    // frontend `classifyAuthStoreFailure` matches on. Assert `starts_with` (not
+    // just `contains`) so a reword in `store_session`/`rest.rs` fails CI here
+    // instead of silently degrading the FE classifier to 'other'.
+    assert!(
+        err.starts_with("Session validation failed (GET /auth/me):"),
+        "store_session error must keep the contract prefix; got: {err}"
+    );
+
+    // Explicitly verify the failure path persisted NOTHING — the gate returns
+    // before the persist step, so the snapshot must read back as unauthenticated
+    // with no session token (a partial regression that wrote a profile would
+    // otherwise still pass on the Err check alone). This is what leaves the user
+    // on the signin page.
+    let snap = snapshot()
+        .await
+        .expect("snapshot after failed store_session")
+        .value;
+    assert!(
+        !snap.auth.is_authenticated,
+        "auth must remain unauthenticated after a failed store_session"
+    );
+    assert!(
+        snap.session_token.is_none(),
+        "session token must not be persisted after a failed store_session"
+    );
+}
+
 #[tokio::test]
 async fn round14_snapshot_preserves_rich_local_state_with_backend_or_stored_user() {
     let _lock = env_lock();