feat:

1. 建立问题库大纲，对应每个人生阶段槽位
2. 鼓励使用更生活化的交流语言共情与总结
3. 降低评审模型可能发生截断的概率
4. 成稿质量维度强化情感表达和上下文连贯性

api/app/agents/chat/interview_agent.py

@@ -170,7 +170,10 @@ class InterviewAgent:
                     omit_system_body=settings.agent_log_omit_system_message_body,
                 ),
             )
-            chat_llm = self.llm.bind(max_tokens=max_tokens)
+            chat_llm = self.llm.bind(
+                max_tokens=max_tokens,
+                temperature=float(settings.chat_interview_temperature),
+            )
             prompt_chars = _message_contents_char_count(messages)
             llm_t0 = time.perf_counter()
             with agent_span(
@@ -292,7 +295,10 @@ class InterviewAgent:
                     omit_system_body=settings.agent_log_omit_system_message_body,
                 ),
             )
-            opening_llm = self.llm.bind(max_tokens=settings.chat_opening_max_tokens)
+            opening_llm = self.llm.bind(
+                max_tokens=settings.chat_opening_max_tokens,
+                temperature=float(settings.chat_interview_temperature),
+            )
             prompt_chars = _message_contents_char_count(messages)
             llm_t0 = time.perf_counter()
             with agent_span(