Normalize OAI messages before conversion and update tool-call prompt

llmcore.py

@@ -623,10 +623,9 @@ class NativeClaudeSession(BaseSession):
         return MockResponse(thinking, content, tool_calls, str(content_blocks))
 
 class NativeOAISession(NativeClaudeSession):
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
     def raw_ask(self, messages):
         """OpenAI streaming. yields text chunks, generator return = list[content_block]"""
+        messages = _fix_messages(messages)
         msgs = ([{"role": "system", "content": self.system}] if self.system else []) + _msgs_claude2oai(messages)
         return (yield from _openai_stream(self.api_base, self.api_key, msgs, self.model, self.api_mode,
                                           temperature=self.temperature, max_tokens=self.max_tokens,
@@ -893,17 +892,17 @@ class MixinSession:
 
 THINKING_PROMPT_ZH = """
 ### 行动规范（持续有效）
-每次回复请遵循：
+每次回复请先在回复文字中包含：
 1. 在 <thinking></thinking> 标签中先分析现状和策略
 2. 在 <summary></summary> 中输出极简单行（<30字）物理快照：上次结果新信息+本次意图。此内容进入长期工作记忆。
-3. 然后才能输出工具调用
+\n**除了最后回答，必须进行工具调用！**
 """.strip()
 THINKING_PROMPT_EN = """
 ### Action Protocol (always in effect)
-For every reply, follow these steps:
+The reply body should first include:
 1. Analyze the current situation and strategy inside <thinking></thinking>
 2. Output a minimal one-line (<30 words) physical snapshot in <summary></summary>: new info from last result + current intent. This goes into long-term working memory.
-3. Then output tool calls
+\n**Tool calls are required for every turn except the final answer!**
 """.strip()
 
 class NativeToolClient: