feat: persist photos and enrich conversation history

backend/tests/test_conversation_history.py

@@ -105,6 +105,54 @@ class ConversationRecorderTest(unittest.IsolatedAsyncioTestCase):
         self.assertEqual(saved["sessionId"], "conv_test")
         self.assertEqual(saved["sequence"], 1)
 
+    async def test_image_is_persisted_as_message_with_artifact(self):
+        conversation = SimpleNamespace(message_count=0)
+
+        class FakeSession:
+            def __init__(self):
+                self.added = []
+                self.committed = False
+
+            async def __aenter__(self):
+                return self
+
+            async def __aexit__(self, *_args):
+                return None
+
+            def add(self, value):
+                self.added.append(value)
+
+            async def get(self, _model, _session_id):
+                return conversation
+
+            async def commit(self):
+                self.committed = True
+
+        session = FakeSession()
+        recorder = ConversationRecorder("conv_test")
+        with (
+            patch("services.conversation_history.SessionLocal", return_value=session),
+            patch("services.conversation_history.put_object") as put_object,
+        ):
+            await recorder._record_image(
+                b"jpeg-data",
+                input_id="input_photo",
+                timestamp="2026-08-05T10:00:00+08:00",
+                mime_type="image/jpeg",
+            )
+
+        self.assertTrue(session.committed)
+        self.assertEqual(len(session.added), 2)
+        message, artifact = session.added
+        self.assertEqual(message.content_type, "image")
+        self.assertEqual(message.role, "user")
+        self.assertEqual(message.extra["input_id"], "input_photo")
+        self.assertEqual(artifact.message_id, message.id)
+        self.assertEqual(artifact.kind, "image")
+        self.assertEqual(artifact.size_bytes, len(b"jpeg-data"))
+        self.assertEqual(conversation.message_count, 1)
+        put_object.assert_called_once()
+
 
 if __name__ == "__main__":
     unittest.main()

backend/tests/test_user_input.py

@@ -32,7 +32,7 @@ class UserInputParserTests(unittest.TestCase):
         self.assertIsNotNone(value)
         self.assertEqual(value.text, "帮我看看")
         self.assertTrue(value.has_camera_frame)
-        self.assertEqual(value.transcript_text, "帮我看看\n已发送一张图片")
+        self.assertEqual(value.transcript_text, "帮我看看")
 
     def test_rejects_legacy_and_unsupported_image_sources(self):
         self.assertIsNone(parse_user_input({"type": "user-text", "text": "旧协议"}))
@@ -54,6 +54,25 @@ class UserInputParserTests(unittest.TestCase):
                 }
             )
 
+    def test_image_only_input_has_no_synthetic_chat_text(self):
+        value = parse_user_input(
+            {
+                "type": "user-input",
+                "schema_version": 1,
+                "input_id": "input_photo",
+                "parts": [
+                    {
+                        "type": "input_image",
+                        "source": {"type": "camera_frame", "frame": "current"},
+                    }
+                ],
+            }
+        )
+
+        self.assertIsNotNone(value)
+        self.assertEqual(value.transcript_text, "")
+        self.assertEqual(value.prompt_text, "请根据用户刚提交的图片进行回复。")
+
     def test_native_image_uses_the_standard_multimodal_user_turn_path(self):
         image = UserImageRawFrame(
             image=bytes([220, 40, 40] * 16 * 16),