diff --git a/aios/scheduler/rr_scheduler.py b/aios/scheduler/rr_scheduler.py
index efd79c017..76a7ebf0f 100644
--- a/aios/scheduler/rr_scheduler.py
+++ b/aios/scheduler/rr_scheduler.py
@@ -178,7 +178,9 @@ def process_llm_requests(self) -> None:
         while self.active:
             try:
                 llm_syscall = self.get_llm_syscall()
-                self._execute_batch_syscalls(llm_syscall, self.llm.execute_llm_syscalls, "LLM")
+                self._execute_batch_syscalls(
+                    [llm_syscall], self.llm.execute_llm_syscalls, "LLM"
+                )
             except Empty:
                 pass
 
diff --git a/tests/modules/scheduler/test_rr_scheduler.py b/tests/modules/scheduler/test_rr_scheduler.py
new file mode 100644
index 000000000..8139fed0f
--- /dev/null
+++ b/tests/modules/scheduler/test_rr_scheduler.py
@@ -0,0 +1,71 @@
+# Tests for round-robin dispatch of individual queued LLM syscalls.
+
+from queue import Empty, Queue
+from unittest.mock import Mock, call
+
+import pytest
+
+from aios.hooks.stores.queue import getMessage
+from aios.scheduler.rr_scheduler import RRScheduler
+from aios.syscall import Syscall
+
+
+def make_scheduler():
+    """Build a scheduler without starting services or worker threads."""
+    return RRScheduler(
+        llm=Mock(),
+        memory_manager=Mock(),
+        storage_manager=Mock(),
+        tool_manager=Mock(),
+        log_mode="console",
+        get_llm_syscall=Mock(),
+        get_memory_syscall=Mock(),
+        get_storage_syscall=Mock(),
+        get_tool_syscall=Mock(),
+        time_slice=0.05,
+    )
+
+
+@pytest.mark.parametrize("request_count", [0, 1, 2])
+def test_dispatches_queued_syscalls_as_single_item_batches(request_count):
+    """An individual queue item must reach the batch executor in a list."""
+    scheduler = make_scheduler()
+    queue = Queue()
+    syscalls = [Syscall(f"agent-{i}", None) for i in range(request_count)]
+    for syscall in syscalls:
+        queue.put(syscall)
+
+    def get_syscall():
+        try:
+            return getMessage(queue)
+        except Empty:
+            scheduler.active = False
+            raise
+
+    scheduler.get_llm_syscall = get_syscall
+    scheduler.active = True
+    scheduler.process_llm_requests()
+
+    assert scheduler.llm.execute_llm_syscalls.call_args_list == [
+        call([syscall]) for syscall in syscalls
+    ]
+    assert queue.empty()
+    for syscall in syscalls:
+        assert syscall.get_status() == "executing"
+        assert syscall.get_time_limit() == scheduler.time_slice
+        assert syscall.get_start_time() is not None
+
+
+def test_batch_executor_still_accepts_multiple_syscalls():
+    """Keep the existing list-based batch API unchanged."""
+    scheduler = make_scheduler()
+    syscalls = [Syscall("first", None), Syscall("second", None)]
+
+    scheduler._execute_batch_syscalls(
+        syscalls, scheduler.llm.execute_llm_syscalls, "LLM"
+    )
+
+    scheduler.llm.execute_llm_syscalls.assert_called_once_with(syscalls)
+    for syscall in syscalls:
+        assert syscall.get_time_limit() == scheduler.time_slice
+        assert syscall.get_status() == "executing"