personal memory agent
Fix fallback error event leaking backup provider errors
When provider fallback retries with a backup provider that also fails,
suppress the backup's error events from the callback and clear the
_evented flag on the original exception so the caller reports the
correct (original) error.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
+13
-1
+13
-1
think/agents.py
+13
-1
think/agents.py
···
625
625
config["model"] = backup_model
626
626
627
627
backup_mod = get_provider_module(backup)
628
+
629
+
# Suppress error events from backup provider — if backup also fails
630
+
# we report the original error, not the backup's error.
631
+
def backup_emit(data: Event) -> None:
632
+
if data.get("event") == "error":
633
+
return
634
+
agent_emit_event(data)
635
+
628
636
try:
629
-
await backup_mod.run_cogitate(config=config, on_event=agent_emit_event)
637
+
await backup_mod.run_cogitate(config=config, on_event=backup_emit)
630
638
except Exception:
639
+
# Ensure the original error is reported by the caller even if the
640
+
# primary provider already emitted its own error event (_evented).
641
+
if hasattr(exc, "_evented"):
642
+
delattr(exc, "_evented")
631
643
raise exc
632
644
finally:
633
645
if config.get("health_stale"):