Enhance DuplexPipeline to support follow-up context for manual opener tool calls

- Introduced logic to trigger a follow-up turn when the manual opener greeting is empty.
- Updated `_execute_manual_opener_tool_calls` to return structured tool call and result data.
- Added `_build_manual_opener_follow_up_context` method to construct context for follow-up turns.
- Modified `_handle_turn` to accept system context for improved conversation management.
- Enhanced tests to validate the new follow-up behavior and ensure proper context handling.

engine/tests/test_tool_call_flow.py

@@ -350,6 +350,76 @@ async def test_manual_opener_legacy_voice_message_prompt_is_normalized(monkeypat
     assert tool_events[0].get("arguments") == {"msg": "您好"}
 
 
+@pytest.mark.asyncio
+async def test_manual_opener_empty_greeting_triggers_follow_up_turn(monkeypatch):
+    pipeline, _events = _build_pipeline(monkeypatch, [[LLMStreamEvent(type="done")]])
+    pipeline.apply_runtime_overrides(
+        {
+            "generatedOpenerEnabled": False,
+            "greeting": "",
+            "output": {"mode": "text"},
+            "tools": [
+                {
+                    "type": "function",
+                    "executor": "client",
+                    "waitForResponse": True,
+                    "function": {
+                        "name": "text_choice_prompt",
+                        "description": "Prompt choice",
+                        "parameters": {"type": "object", "properties": {"question": {"type": "string"}}},
+                    },
+                }
+            ],
+            "manualOpenerToolCalls": [
+                {
+                    "toolName": "text_choice_prompt",
+                    "arguments": {
+                        "question": "请选择业务类型",
+                        "options": ["账单", "报修"],
+                    },
+                }
+            ],
+        }
+    )
+
+    async def _fake_manual_opener_calls() -> Dict[str, List[Dict[str, Any]]]:
+        return {
+            "toolCalls": [
+                {
+                    "tool_call_id": "call_opener_1",
+                    "tool_name": "text_choice_prompt",
+                    "arguments": {"question": "请选择业务类型", "options": ["账单", "报修"]},
+                }
+            ],
+            "toolResults": [
+                {
+                    "tool_call_id": "call_opener_1",
+                    "name": "text_choice_prompt",
+                    "output": {"selected": "报修"},
+                    "status": {"code": 200, "message": "ok"},
+                }
+            ],
+        }
+
+    called: Dict[str, Any] = {}
+    waiter = asyncio.Event()
+
+    async def _fake_handle_turn(user_text: str, system_context: str | None = None) -> None:
+        called["user_text"] = user_text
+        called["system_context"] = system_context or ""
+        waiter.set()
+
+    monkeypatch.setattr(pipeline, "_execute_manual_opener_tool_calls", _fake_manual_opener_calls)
+    monkeypatch.setattr(pipeline, "_handle_turn", _fake_handle_turn)
+
+    await pipeline.emit_initial_greeting()
+    await asyncio.wait_for(waiter.wait(), timeout=1.0)
+
+    assert called.get("user_text") == ""
+    assert "opener_tool_results" in called.get("system_context", "")
+    assert "报修" in called.get("system_context", "")
+
+
 @pytest.mark.asyncio
 async def test_ws_message_parses_tool_call_results():
     msg = parse_client_message(