fix(core): prefer _llm_gateway over _llm_client in _handle_llm_generate

AgentPool.create_agent() injects only llm_gateway (v2 path) into
ConfigDrivenAgent and never sets the legacy _llm_client field. The
previous _handle_llm_generate implementation gated the entire LLM
call on `self._llm_client is None`, so every expert/board agent
created via the pool silently fell through to the
`llm_generate_no_client` placeholder — which is why the private
board moderator/experts reported "LLM 不可用" even when the gateway
and providers were healthy.

Mirror the same precedence used by `_handle_direct`: prefer
`_llm_gateway` (with agent_name/task_type for usage tracking), fall
back to `_llm_client` for backward compatibility, and only return
the no-client placeholder when both are unavailable.

Add a unit test that constructs an agent with `llm_gateway` only
(_llm_client is None) and asserts the gateway.chat path is taken.

src/agentkit/core/config_driven.py

@@ -1358,7 +1358,14 @@ class ConfigDrivenAgent(BaseAgent, EvolutionMixin):
         return gateway
 
     async def _handle_llm_generate(self, task: TaskMessage) -> dict:
-        """LLM 生成模式：渲染 Prompt → 调用 LLM → 解析输出"""
+        """LLM 生成模式：渲染 Prompt → 调用 LLM → 解析输出
+
+        ponytail: 优先使用 ``_llm_gateway``（v2 路径，``AgentPool`` 注入的），
+        仅在 gateway 缺失时回退到 ``_llm_client``（legacy 字段）。原实现只检查
+        ``_llm_client``，导致 ``AgentPool`` 创建的 expert/board agent 在
+        ``_llm_client is None`` 但 ``_llm_gateway`` 已注入时错误降级为
+        ``llm_generate_no_client``。
+        """
         if not self._prompt_template:
             raise ConfigValidationError(
                 agent_name=self.name,
@@ -1371,7 +1378,19 @@ class ConfigDrivenAgent(BaseAgent, EvolutionMixin):
         variables["task_type"] = task.task_type
         messages = self._prompt_template.render(variables=variables)
 
-        # 调用 LLM
+        # v2 路径：_llm_gateway 优先（AgentPool 仅注入 gateway，不设 _llm_client）
+        if self._llm_gateway is not None:
+            llm_params = self._config.llm.copy() if self._config.llm else {}
+            model = llm_params.get("model", "default")
+            response = await self._llm_gateway.chat(
+                messages=messages,
+                model=model,
+                agent_name=self.name,
+                task_type=task.task_type,
+            )
+            return self._parse_llm_response(response.content)
+
+        # 兼容旧路径：_llm_client（legacy）
         if self._llm_client is None:
             # 无 LLM 客户端时返回渲染后的 Prompt（降级模式）
             return {

tests/unit/test_config_driven.py

@@ -144,6 +144,52 @@ class TestConfigDrivenAgent:
         assert result["mode"] == "llm_generate_no_client"
         assert len(result["messages"]) == 2  # system + user
 
+    async def test_llm_generate_with_gateway_only(self):
+        """仅注入 LLMGateway（_llm_client=None）时应正常调用 LLM。
+
+        复现：AgentPool.create_agent() 只注入 llm_gateway，不设 _llm_client。
+        原实现错误降级为 llm_generate_no_client，导致私董会专家返回占位文本。
+        """
+
+        class _StubResponse:
+            def __init__(self, content: str) -> None:
+                self.content = content
+
+        class _StubGateway:
+            def __init__(self) -> None:
+                self.calls: list[dict] = []
+
+            async def chat(self, *, messages, model, agent_name, task_type, **kwargs):
+                self.calls.append(
+                    {
+                        "messages": messages,
+                        "model": model,
+                        "agent_name": agent_name,
+                        "task_type": task_type,
+                    }
+                )
+                return _StubResponse(json.dumps({"title": "FromGateway", "content": "OK"}))
+
+        gateway = _StubGateway()
+        config = AgentConfig.from_dict(_sample_llm_config())
+        agent = ConfigDrivenAgent(config=config, llm_gateway=gateway)
+
+        # 验证前置条件：_llm_client 必须为 None，模拟 AgentPool 注入路径
+        assert agent._llm_client is None
+        assert agent._llm_gateway is gateway
+
+        task = _make_task()
+        result = await agent.handle_task(task)
+
+        # 不应再走 no_client 降级
+        assert "mode" not in result or result.get("mode") != "llm_generate_no_client"
+        assert result["title"] == "FromGateway"
+        assert result["content"] == "OK"
+        # 验证 gateway.chat 被实际调用
+        assert len(gateway.calls) == 1
+        assert gateway.calls[0]["agent_name"] == config.name
+        assert gateway.calls[0]["model"] == "gpt-4"
+
     async def test_llm_generate_with_client(self):
         """有 LLM 客户端时调用 LLM 并解析结果"""