apps/server/src/orchestration/Layers/ProviderRuntimeIngestion.ts (1)

852-863: ⚠️ Potential issue | 🟠 Major

Gate turn.aborted on the current active turn.

turn.aborted is now using the turn.completed fallback path, so a late/duplicate abort with no active turn can still move the session to "interrupted" and clear lastError. Also, the new finalization block does not consult the lifecycle guard, so an abort for a different turn can still flush buffered assistant text / proposed plans from that stale turn. Please restrict abort handling to a matching activeTurnId, and add a regression for late/different-turn aborts.

Possible tightening

-          case "turn.completed":
-          case "turn.aborted":
+          case "turn.completed":
             if (conflictsWithActiveTurn || missingTurnForActiveTurn) {
               return false;
             }
             // Only the active turn may close the lifecycle state.
             if (activeTurnId !== null && eventTurnId !== undefined) {
               return sameId(activeTurnId, eventTurnId);
             }
             // If no active turn is tracked, accept completion scoped to this thread.
             return true;
+          case "turn.aborted":
+            return (
+              activeTurnId !== null &&
+              eventTurnId !== undefined &&
+              sameId(activeTurnId, eventTurnId)
+            );

-      if (event.type === "turn.completed" || event.type === "turn.aborted") {
+      const shouldFinalizeTurn =
+        event.type === "turn.completed" ||
+        (event.type === "turn.aborted" &&
+          activeTurnId !== null &&
+          eventTurnId !== undefined &&
+          sameId(activeTurnId, eventTurnId));
+      if (shouldFinalizeTurn) {

As per coding guidelines, apps/server/**: Prioritize correctness, reliability, and predictable behavior under reconnects, session restarts, partial streams, and provider failures.

Also applies to: 1044-1075

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@apps/server/src/orchestration/Layers/ProviderRuntimeIngestion.ts` around
lines 852 - 863, The handling for "turn.aborted" currently falls through to the
"turn.completed" path allowing late or out-of-thread aborts to clear session
state and flush stale assistant content; change the switch so "turn.aborted" is
explicitly gated to only succeed when activeTurnId is non-null and
sameId(activeTurnId, eventTurnId) is true (i.e., mirror the explicit active-turn
check used for completions) and ensure the finalization block consults the
lifecycle guard before clearing lastError or flushing buffered assistant
text/proposed plans; update ProviderRuntimeIngestion's turn finalization logic
(the block referencing activeTurnId, eventTurnId, sameId, and the lifecycle
guard) and add a regression test that sends a late/different-turn "turn.aborted"
to assert that session state, lastError, and buffered assistant output remain
unchanged.