Merge branch 'lsdefine:main' into fix/linux-desktop-pet-pyside6

assets/sys_prompt.txt

@@ -1,6 +1,6 @@
 # Role: 物理级全能执行者
 你拥有文件读写、脚本执行、用户浏览器JS注入、系统级干预的物理操作权限。禁止推诿"无法操作"——不空想，用工具探测。
 ## 行动原则
-调用工具前在 <thinking> 内推演：当前阶段、上步结果是否符合预期、下步策略。
+调用工具前在<thinking>内推演：当前阶段、上步结果是否符合预期、下步策略；<summary>内输出极简总结。
 - 探测优先：失败时先充分获取信息（日志/状态/上下文），关键信息存入工作记忆，再决定重试或换方案。不可逆操作先询问用户。
 - 失败升级：1次→读错误理解原因，2次→探测环境状态，3次→深度分析后换方案或问用户。禁止无新信息的重复操作。

ga.py

@@ -504,7 +504,7 @@ class GenericAgentHandler(BaseHandler):
 
     def _get_anchor_prompt(self, skip=False):
         if skip: return "\n"
-        h_str = "\n".join(self.history_info[-20:])
+        h_str = "\n".join(self.history_info[-40:])
         prompt = f"\n### [WORKING MEMORY]\n<history>\n{h_str}\n</history>"
         prompt += f"\nCurrent turn: {self.current_turn}\n"
         if self.working.get('key_info'): prompt += f"\n<key_info>{self.working.get('key_info')}</key_info>"
@@ -523,7 +523,7 @@ class GenericAgentHandler(BaseHandler):
             clean_args = {k: v for k, v in args.items() if not k.startswith('_')}
             summary = f"调用工具{tool_name}, args: {clean_args}"
             if tool_name == 'no_tool': summary = "直接回答了用户问题"
-            next_prompt += "\n[DANGER] 上一轮遗漏了<summary>，需要按协议在<summary>中输出极简单行摘要！" 
+            next_prompt += "\n[DANGER] 你遗漏了<summary>，必须按协议一直在每次回复中用<summary>中输出极简单行摘要！" 
         summary = smart_format(summary, max_str_len=100)
         self.history_info.append(f'[Agent] {summary}')
         if turn % 65 == 0 and 'plan' not in str(self.working.get('related_sop')):

llmcore.py

@@ -229,6 +229,7 @@ def _parse_openai_sse(resp_lines, api_mode="chat_completions"):
         return blocks
     else:
         tc_buf = {}  # index -> {id, name, args}
+        reasoning_text = ""
         for line in resp_lines:
             if not line: continue
             line = line.decode('utf-8', errors='replace') if isinstance(line, bytes) else line
@@ -239,6 +240,8 @@ def _parse_openai_sse(resp_lines, api_mode="chat_completions"):
             except: continue
             ch = (evt.get("choices") or [{}])[0]
             delta = ch.get("delta") or {}
+            if delta.get("reasoning_content"):
+                reasoning_text += delta["reasoning_content"]
             if delta.get("content"):
                 text = delta["content"]; content_text += text; yield text
             for tc in (delta.get("tool_calls") or []):
@@ -253,6 +256,7 @@ def _parse_openai_sse(resp_lines, api_mode="chat_completions"):
             usage = evt.get("usage")
             if usage: _record_usage(usage, api_mode)
         blocks = []
+        if reasoning_text: blocks.append({"type": "thinking", "thinking": reasoning_text})
         if content_text: blocks.append({"type": "text", "text": content_text})
         for idx in sorted(tc_buf):
             tc = tc_buf[idx]
@@ -294,6 +298,9 @@ def _parse_openai_json(data, api_mode="chat_completions"):
     else:
         _record_usage(data.get("usage") or {}, api_mode)
         msg = (data.get("choices") or [{}])[0].get("message", {})
+        reasoning = msg.get("reasoning_content", "")
+        if reasoning:
+            blocks.append({"type": "thinking", "thinking": reasoning})
         content = msg.get("content", "")
         if content:
             blocks.append({"type": "text", "text": content}); yield content
@@ -330,6 +337,7 @@ def _openai_stream(api_base, api_key, messages, model, api_mode='chat_completion
         payload = {"model": model, "input": _to_responses_input(messages), "stream": stream, 
                    "prompt_cache_key": _RESP_CACHE_KEY, "instructions": system or "You are an Omnipotent Executor."}
         if reasoning_effort: payload["reasoning"] = {"effort": reasoning_effort}
+        if max_tokens: payload["max_output_tokens"] = max_tokens
     else:
         url = auto_make_url(api_base, "chat/completions")
         if system: messages = [{"role": "system", "content": system}] + messages
@@ -337,7 +345,7 @@ def _openai_stream(api_base, api_key, messages, model, api_mode='chat_completion
         payload = {"model": model, "messages": messages, "stream": stream}
         if stream: payload["stream_options"] = {"include_usage": True}
         if temperature != 1: payload["temperature"] = temperature
-        if max_tokens: payload["max_tokens"] = max_tokens
+        if max_tokens: payload["max_completion_tokens" if ml.startswith(("gpt-5", "o1", "o2", "o3", "o4")) else "max_tokens"] = max_tokens
         if reasoning_effort: payload["reasoning_effort"] = reasoning_effort
     if tools: payload["tools"] = _prepare_oai_tools(tools, api_mode)
     RETRYABLE = {408, 409, 425, 429, 500, 502, 503, 504, 529}
@@ -412,8 +420,8 @@ def _to_responses_input(messages):
                 elif ptype == "image_url":
                     url = (part.get("image_url") or {}).get("url", "")
                     if url and role != "assistant": parts.append({"type": "input_image", "image_url": url})
-        if len(parts) == 0: parts = [{"type": text_type, "text": str(content) or '[empty]'}]
-        result.append({"role": role, "content": parts})
+        if len(parts) == 0 and not isinstance(content, list): parts = [{"type": text_type, "text": str(content) or '[empty]'}]
+        if parts: result.append({"role": role, "content": parts})
         pending = []
         for tc in (msg.get("tool_calls") or []):
             f = tc.get("function", {})
@@ -430,16 +438,18 @@ def _msgs_claude2oai(messages):
         content = msg.get("content", "")
         blocks = content if isinstance(content, list) else [{"type": "text", "text": str(content)}]
         if role == "assistant":
-            text_parts, tool_calls = [], []
+            text_parts, tool_calls, reasoning = [], [], ""
             for b in blocks:
                 if not isinstance(b, dict): continue
-                if b.get("type") == "text" and b.get("text"): text_parts.append({"type": "text", "text": b.get("text", "")})
+                if b.get("type") == "thinking" and b.get("thinking"): reasoning = b["thinking"]
+                elif b.get("type") == "text" and b.get("text"): text_parts.append({"type": "text", "text": b.get("text", "")})
                 elif b.get("type") == "tool_use":
                     tool_calls.append({
                         "id": b.get("id") or '', "type": "function",
                         "function": {"name": b.get("name", ""), "arguments": json.dumps(b.get("input", {}), ensure_ascii=False)}
                     })
             m = {"role": "assistant"}
+            if reasoning: m["reasoning_content"] = reasoning
             if text_parts: m["content"] = text_parts
             else: m["content"] = ""
             if tool_calls: m["tool_calls"] = tool_calls
@@ -525,6 +535,11 @@ class BaseSession:
             if not content.startswith("!!!Error:"): self.history.append({"role": "assistant", "content": [{"type": "text", "text": content}]})
         return _ask_gen() if stream else ''.join(list(_ask_gen()))
 
+def _keep_claude_block(b): return not isinstance(b, dict) or b.get("type") != "thinking" or b.get("signature")
+def _drop_unsigned_thinking(messages):
+    for m in messages: m["content"] = [b for b in m["content"] if _keep_claude_block(b)]
+    return messages
+
 class ClaudeSession(BaseSession):
     def raw_ask(self, messages):
         headers = {"x-api-key": self.api_key, "Content-Type": "application/json", "anthropic-version": "2023-06-01", "anthropic-beta": "prompt-caching-2024-07-31"}
@@ -540,7 +555,7 @@ class ClaudeSession(BaseSession):
             yield (err := f"!!!Error: {e}")
             return [{"type": "text", "text": err}]
     def make_messages(self, raw_list):
-        msgs = [{"role": m['role'], "content": list(m['content'])} for m in raw_list]
+        msgs = _drop_unsigned_thinking([{"role": m['role'], "content": list(m['content'])} for m in raw_list])
         user_idxs = [i for i, m in enumerate(msgs) if m['role'] == 'user']
         for idx in user_idxs[-2:]:
             msgs[idx]["content"][-1] = dict(msgs[idx]["content"][-1], cache_control={"type": "ephemeral"})
@@ -582,7 +597,7 @@ class NativeClaudeSession(BaseSession):
         self._device_id = uuid.uuid4().hex + uuid.uuid4().hex[:32]
         self.tools = None
     def raw_ask(self, messages):
-        messages = _fix_messages(messages)
+        messages = _drop_unsigned_thinking(_fix_messages(messages))
         model = self.model
         beta_parts = ["claude-code-20250219", "interleaved-thinking-2025-05-14", "redact-thinking-2026-02-12", "prompt-caching-scope-2026-01-05"]
         if "[1m]" in model.lower():