fix: prioritize child idle grace retries

dist/server.js

@@ -2330,7 +2330,7 @@ var server = async ({ client }, options) => {
           return;
         }
         taskDeferredSessions.add(sessionID);
-        scheduleSettledContinuation(sessionID, taskStatus.retryAt != null ? taskStatus.retryAt - Date.now() : TASK_BLOCK_RETRY_MS, scheduled != null);
+        scheduleSettledContinuation(sessionID, taskStatus.retryAt != null ? taskStatus.retryAt - Date.now() : TASK_BLOCK_RETRY_MS, scheduled != null || taskStatus.retryAt != null);
         return;
       }
       if (busySessions.has(sessionID))

src/server.ts

@@ -1348,11 +1348,13 @@ const server: Plugin = async ({ client }, options?: Options) => {
         // Always re-arm. A task block is the only deferral that records nothing on the
         // goal, so without a scheduled retry a child that never reports a terminal state
         // silently ends auto-continuation: nothing refreshes live children again and the
-        // goal keeps reading active with no stop reason.
+        // goal keeps reading active with no stop reason. When a fresh child snapshot adds
+        // a shorter idle-grace deadline, replace the older fallback timer so the stale
+        // running record is revisited promptly even if a poll was already queued.
         scheduleSettledContinuation(
           sessionID,
           taskStatus.retryAt != null ? taskStatus.retryAt - Date.now() : TASK_BLOCK_RETRY_MS,
-          scheduled != null,
+          scheduled != null || taskStatus.retryAt != null,
         )
         return
       }