Add presence probe configuration to Assistant model and API. Introduce new fields for enabling presence probes, idle and cooldown durations, maximum prompts, context inclusion, and custom questions. Update schemas, routers, and frontend components to support these features, along with corresponding tests to ensure functionality.

engine/tests/test_tool_call_flow.py

@@ -1,5 +1,6 @@
 import asyncio
 import json
+import time
 from typing import Any, Dict, List
 
 import pytest
@@ -86,7 +87,12 @@ def _build_pipeline(monkeypatch, llm_rounds: List[List[LLMStreamEvent]]) -> tupl
     async def _capture_event(event: Dict[str, Any], priority: int = 20):
         events.append(event)
 
-    async def _noop_speak(_text: str, fade_in_ms: int = 0, fade_out_ms: int = 8):
+    async def _noop_speak(
+        _text: str,
+        fade_in_ms: int = 0,
+        fade_out_ms: int = 8,
+        **_kwargs,
+    ):
         return None
 
     monkeypatch.setattr(pipeline, "_send_event", _capture_event)
@@ -362,6 +368,89 @@ async def test_duplicate_tool_results_are_ignored(monkeypatch):
     assert result.get("output", {}).get("value") == 1
 
 
+@pytest.mark.asyncio
+async def test_presence_probe_emits_contextual_direct_prompt(monkeypatch):
+    pipeline, events = _build_pipeline(monkeypatch, [[LLMStreamEvent(type="done")]])
+    pipeline.apply_runtime_overrides(
+        {
+            "presenceProbe": {
+                "enabled": True,
+                "idleSeconds": 5,
+                "cooldownSeconds": 5,
+                "maxPrompts": 2,
+                "includeContext": True,
+            }
+        }
+    )
+    await pipeline._shutdown_presence_probe_task()
+    await pipeline.conversation.add_assistant_turn("请把你的订单号告诉我，我继续帮你处理。")
+    pipeline._last_user_activity_ms = (time.monotonic() * 1000.0) - 8000.0
+
+    fired = await pipeline._run_presence_probe_once()
+
+    assert fired is True
+    probe_text_events = [e for e in events if e.get("type") == "assistant.response.final"]
+    assert probe_text_events
+    assert "订单号" in str(probe_text_events[-1].get("text") or "")
+    assert any(e.get("type") == "output.audio.start" for e in events)
+    assert not any(e.get("type") == "assistant.tool_call" for e in events)
+
+
+@pytest.mark.asyncio
+async def test_presence_probe_respects_max_prompts_limit(monkeypatch):
+    pipeline, events = _build_pipeline(monkeypatch, [[LLMStreamEvent(type="done")]])
+    pipeline.apply_runtime_overrides(
+        {
+            "presenceProbe": {
+                "enabled": True,
+                "idleSeconds": 5,
+                "cooldownSeconds": 5,
+                "maxPrompts": 1,
+                "waitForResponse": False,
+            }
+        }
+    )
+    await pipeline._shutdown_presence_probe_task()
+    await pipeline.conversation.add_assistant_turn("我们继续。")
+    pipeline._last_user_activity_ms = (time.monotonic() * 1000.0) - 8000.0
+
+    first_fired = await pipeline._run_presence_probe_once()
+    second_fired = await pipeline._run_presence_probe_once(
+        now_ms=(time.monotonic() * 1000.0) + 10000.0
+    )
+
+    assert first_fired is True
+    assert second_fired is False
+    assert len([e for e in events if e.get("type") == "assistant.response.final"]) == 1
+
+
+@pytest.mark.asyncio
+async def test_presence_probe_text_mode_emits_text_only(monkeypatch):
+    pipeline, events = _build_pipeline(monkeypatch, [[LLMStreamEvent(type="done")]])
+    pipeline.apply_runtime_overrides(
+        {
+            "output": {"mode": "text"},
+            "presenceProbe": {
+                "enabled": True,
+                "idleSeconds": 5,
+                "cooldownSeconds": 5,
+                "maxPrompts": 1,
+                "waitForResponse": False,
+            },
+        }
+    )
+    await pipeline._shutdown_presence_probe_task()
+    await pipeline.conversation.add_assistant_turn("我们继续。")
+    pipeline._last_user_activity_ms = (time.monotonic() * 1000.0) - 8000.0
+
+    fired = await pipeline._run_presence_probe_once()
+
+    assert fired is True
+    assert any(e.get("type") == "assistant.response.final" for e in events)
+    assert not any(e.get("type") == "assistant.tool_call" for e in events)
+    assert not any(e.get("type") == "output.audio.start" for e in events)
+
+
 @pytest.mark.asyncio
 async def test_server_calculator_emits_tool_result(monkeypatch):
     pipeline, events = _build_pipeline(