feat(workflow): enhance message stages and image routing

backend/tests/test_brains.py

@@ -1021,7 +1021,7 @@ class WorkflowBrainTests(unittest.IsolatedAsyncioTestCase):
                         "type": "message",
                         "data": {
                             "speech": "请问您怎么称呼？",
-                            "showMessage": False,
+                            "completionPolicy": "playback",
                         },
                     },
                     {
@@ -1457,11 +1457,10 @@ class WorkflowBrainTests(unittest.IsolatedAsyncioTestCase):
                             "type": "message",
                             "data": {
                                 "speech": "请先确认 {{customer}} 的重要信息。",
-                                "showMessage": True,
                                 "title": "重要提示",
                                 "message": "请核对客户信息。",
                                 "confirmLabel": "确认",
-                                "requireConfirmation": True,
+                                "completionPolicy": "confirmation",
                             },
                         },
                     ],
@@ -1585,6 +1584,131 @@ class WorkflowBrainTests(unittest.IsolatedAsyncioTestCase):
         self.assertTrue(result.succeeded)
         self.assertEqual(input_states, [False, True])
 
+    async def test_interruptible_message_forwards_multimodal_input_once(self):
+        graph = {
+            "specVersion": 3,
+            "settings": {},
+            "nodes": [
+                {"id": "start", "type": "start", "data": {}},
+                {
+                    "id": "message",
+                    "type": "message",
+                    "data": {
+                        "speech": "请按提示操作，也可以直接告诉我需求。",
+                        "completionPolicy": "interruptible",
+                    },
+                },
+                {
+                    "id": "agent",
+                    "type": "agent",
+                    "data": {
+                        "prompt": "处理用户输入",
+                        "contextPolicy": "fresh",
+                    },
+                },
+            ],
+            "edges": [
+                {
+                    "id": "start-message",
+                    "source": "start",
+                    "target": "message",
+                    "data": {"mode": "always"},
+                },
+                {
+                    "id": "message-agent",
+                    "source": "message",
+                    "target": "agent",
+                    "data": {"mode": "always"},
+                },
+            ],
+        }
+        brain = WorkflowBrain(graph)
+        queued = []
+        input_states = []
+
+        class PlaybackCallEnd(FakeCallEnd):
+            def __init__(self):
+                super().__init__()
+                self.completion = None
+
+            def track_speech(self):
+                self.completion = asyncio.get_running_loop().create_future()
+                return self.completion
+
+        class FakeManager:
+            def __init__(self):
+                self.current_node = None
+                self.configs = []
+
+            async def initialize(self, config):
+                self.current_node = config["name"]
+                self.configs.append(config)
+
+            async def set_node_from_config(self, config):
+                self.current_node = config["name"]
+                self.configs.append(config)
+
+        async def queue_frame(frame):
+            queued.append(frame)
+
+        call_end = PlaybackCallEnd()
+        manager = FakeManager()
+        brain._runtime = BrainRuntime(
+            context=LLMContext(messages=[]),
+            llm=FakeLLM(),
+            queue_frame=queue_frame,
+            set_system_prompt=lambda _prompt: None,
+            set_tools=lambda _tools: None,
+            call_end=call_end,
+            set_input_enabled=input_states.append,
+        )
+        brain._manager = manager
+
+        await brain.on_connected()
+        await asyncio.sleep(0)
+        self.assertEqual(manager.current_node, "message")
+        self.assertIsNotNone(call_end.completion)
+        self.assertTrue(all(input_states))
+
+        image_message = {
+            "role": "user",
+            "content": [
+                {"type": "text", "text": "已发送一张图片"},
+                {
+                    "type": "image_url",
+                    "image_url": {"url": "data:image/jpeg;base64,AA=="},
+                },
+            ],
+        }
+        await brain.on_user_turn_end(
+            "已发送一张图片",
+            user_message=image_message,
+        )
+        await asyncio.sleep(0)
+
+        self.assertEqual(manager.current_node, "agent")
+        self.assertEqual(
+            [config["name"] for config in manager.configs].count("agent"),
+            1,
+        )
+        self.assertTrue(call_end.completion.cancelled())
+        self.assertTrue(all(input_states))
+        self.assertEqual(
+            manager.configs[-1]["task_messages"],
+            [image_message],
+        )
+        self.assertEqual(
+            sum(isinstance(frame, LLMRunFrame) for frame in queued),
+            1,
+        )
+        self.assertTrue(
+            any(
+                isinstance(frame, OutputTransportMessageUrgentFrame)
+                and frame.message.get("event") == "message_interrupted"
+                for frame in queued
+            )
+        )
+
     async def test_message_between_agents_resumes_after_playback(self):
         graph = {
             "specVersion": 3,