修复了深度思考模式

src/config/agents.py

@@ -8,13 +8,13 @@
 
 # Define agent-LLM mapping
 AGENT_LLM_MAP: dict[str, LLMType] = {
-    "coordinator": "basic",
-    "planner": "basic",
-    "researcher": "basic",
-    "coder": "basic",
-    "reporter": "basic",
-    "podcast_script_writer": "basic",
-    "ppt_composer": "basic",
-    "prose_writer": "basic",
-    "prompt_enhancer": "basic",
+    "coordinator": "reasoning",
+    "planner": "reasoning",
+    "researcher": "reasoning",
+    "coder": "reasoning",
+    "reporter": "reasoning",
+    "podcast_script_writer": "reasoning",
+    "ppt_composer": "reasoning",
+    "prose_writer": "reasoning",
+    "prompt_enhancer": "reasoning",
 }

src/graph/nodes.py

@@ -104,10 +104,7 @@ def planner_node(
     if configurable.enable_deep_thinking:
         llm = get_llm_by_type("reasoning")
     elif AGENT_LLM_MAP["planner"] == "basic":
-        llm = get_llm_by_type("basic").with_structured_output(
-            Plan,
-            method="json_mode",
-        )
+        llm = get_llm_by_type("basic")
     else:
         llm = get_llm_by_type(AGENT_LLM_MAP["planner"])
 
@@ -116,13 +113,9 @@ def planner_node(
         return Command(goto="reporter")
 
     full_response = ""
-    if AGENT_LLM_MAP["planner"] == "basic" and not configurable.enable_deep_thinking:
-        response = llm.invoke(messages)
-        full_response = response.model_dump_json(indent=4, exclude_none=True)
-    else:
-        response = llm.stream(messages)
-        for chunk in response:
-            full_response += chunk.content
+    response = llm.stream(messages)
+    for chunk in response:
+        full_response += chunk.content
     logger.debug(f"Current state messages: {state['messages']}")
     logger.info(f"Planner response: {full_response}")
 

src/podcast/graph/script_writer_node.py

@@ -17,14 +17,32 @@
 
 def script_writer_node(state: PodcastState):
     logger.info("Generating script for podcast...")
-    model = get_llm_by_type(
-        AGENT_LLM_MAP["podcast_script_writer"]
-    ).with_structured_output(Script, method="json_mode")
-    script = model.invoke(
+    model = get_llm_by_type(AGENT_LLM_MAP["podcast_script_writer"])
+    response = model.invoke(
         [
             SystemMessage(content=get_prompt_template("podcast/podcast_script_writer")),
             HumanMessage(content=state["input"]),
         ],
     )
+
+    # 手动解析 JSON 响应
+    try:
+        import json
+        from src.utils.json_utils import repair_json_output
+
+        # 修复和解析 JSON 响应
+        content = repair_json_output(response.content)
+        script_data = json.loads(content)
+
+        # 创建 Script 对象
+        script = Script(**script_data)
+    except Exception as e:
+        logger.error(f"Failed to parse script JSON: {e}")
+        # 如果解析失败，创建一个默认的 Script 对象
+        script = Script(
+            locale="en",
+            lines=[]
+        )
+
     print(script)
     return {"script": script, "audio_chunks": []}