fix(fallback): trigger eager fallback on 503/529 provider overload

agent/error_classifier.py

@@ -611,7 +611,7 @@ def _classify_by_status(
         return result_fn(FailoverReason.server_error, retryable=True)
 
     if status_code in (503, 529):
-        return result_fn(FailoverReason.overloaded, retryable=True)
+        return result_fn(FailoverReason.overloaded, retryable=True, should_fallback=True)
 
     # Other 4xx — non-retryable
     if 400 <= status_code < 500:

run_agent.py

@@ -11144,6 +11144,21 @@ def _stop_spinner():
                                 primary_recovery_attempted = False
                                 continue
 
+                    # Eager fallback for provider overload (503/529).
+                    # Provider-side overload cannot be fixed by credential rotation,
+                    # so bypass the pool check and switch to a fallback immediately.
+                    # Fixes #11314 / #10210.
+                    if (
+                        classified.reason == FailoverReason.overloaded
+                        and self._fallback_index < len(self._fallback_chain)
+                    ):
+                        self._emit_status("⚠️ Provider overloaded — switching to fallback provider...")
+                        if self._try_activate_fallback(reason=classified.reason):
+                            retry_count = 0
+                            compression_attempts = 0
+                            primary_recovery_attempted = False
+                            continue
+
                     # ── Nous Portal: record rate limit & skip retries ─────
                     # When Nous returns a 429, record the reset time to a
                     # shared file so ALL sessions (cron, gateway, auxiliary)

tests/agent/test_error_classifier.py

@@ -283,11 +283,15 @@ def test_503_overloaded(self):
         e = MockAPIError("Service Unavailable", status_code=503)
         result = classify_api_error(e)
         assert result.reason == FailoverReason.overloaded
+        assert result.should_fallback is True
+        assert result.retryable is True
 
     def test_529_anthropic_overloaded(self):
         e = MockAPIError("Overloaded", status_code=529)
         result = classify_api_error(e)
         assert result.reason == FailoverReason.overloaded
+        assert result.should_fallback is True
+        assert result.retryable is True
 
     # ── Model not found ──
 

tests/run_agent/test_provider_fallback.py

@@ -5,6 +5,13 @@
 advancement through multiple providers.
 """
 
+import sys
+import types
+
+# Prevent heavy imports from failing in lightweight test environments.
+sys.modules.setdefault("fire", types.SimpleNamespace(Fire=lambda *a, **k: None))
+sys.modules.setdefault("firecrawl", types.SimpleNamespace(Firecrawl=object))
+
 from unittest.mock import MagicMock, patch
 
 from run_agent import AIAgent, _pool_may_recover_from_rate_limit