feat: enforce full AI outbound generation and reduce template replies

tests/test_regression_pipeline.py

@@ -432,6 +432,31 @@ class RegressionPipelineTest(unittest.IsolatedAsyncioTestCase):
         self.assertTrue(resp.should_reply)
         self.assertIn(resp.reply, ("嗯咯", "嗯啦", "嗯", "哦"))
 
+    async def test_meaningless_short_text_not_ping_when_collecting(self):
+        agent = CustomerServiceAgent()
+        st = agent._get_conversation_state(self.customer_id)
+        st.pending_image_urls = ["https://img.alicdn.com/a.jpg"]
+        st.quote_phase = "collecting"
+        agent._sync_pending_quote_state(self.customer_id, st)
+
+        msg = CustomerMessage(
+            msg_id="m-meaningless-collecting",
+            acc_id="test_shop",
+            msg="好的",
+            from_id=self.customer_id,
+            from_name="t",
+            cy_id=self.customer_id,
+            acc_type="AliWorkbench",
+            msg_type=0,
+            cy_name="t",
+            goods_name="专业找图",
+            goods_order="",
+        )
+        resp = await agent.process_message(msg)
+        self.assertTrue(resp.should_reply)
+        self.assertTrue((resp.reply or "").strip())
+        self.assertNotIn(resp.reply, ("嗯咯", "嗯啦", "嗯", "哦"))
+
     def tearDown(self):
         db.clear_pending_quote_state(self.customer_id)