From 44cab88546c6511483f495a36c9cad2499699421 Mon Sep 17 00:00:00 2001
From: butubb <1422726308@qq.com>
Date: Fri, 4 Sep 2026 13:23:04 +0800
Subject: [PATCH] =?UTF-8?q?fix:=20Agent=20=E6=AD=A5=E9=AA=A4=E4=B8=8A?=
 =?UTF-8?q?=E9=99=90=E4=BC=98=E5=8C=96=E2=80=94=E2=80=94max=5Fsteps=2020?=
 =?UTF-8?q?=E2=86=9240=EF=BC=88=E5=A4=9A=E6=AD=A5=E4=BB=BB=E5=8A=A1?=
 =?UTF-8?q?=E5=A6=82=E6=89=93=E5=BC=80=E6=8A=96=E9=9F=B3=E6=90=9C=E7=B4=A2?=
 =?UTF-8?q?=EF=BC=89=EF=BC=8Csystem=20prompt=20=E5=8A=A0=E6=94=B6=E6=95=9B?=
 =?UTF-8?q?=E8=A7=84=E8=8C=83=EF=BC=88=E6=97=A0=E8=BF=9B=E5=B1=95=206=20?=
 =?UTF-8?q?=E6=AD=A5=E4=B8=BB=E5=8A=A8=E5=81=9C=E6=AD=A2=E6=80=BB=E7=BB=93?=
 =?UTF-8?q?=EF=BC=89=EF=BC=8C=E8=B6=85=E9=99=90=E6=97=B6=E6=97=A0=E5=B7=A5?=
 =?UTF-8?q?=E5=85=B7=E6=94=B6=E5=B0=BE=E8=AF=B7=E6=B1=82=E8=AE=A9=E6=A8=A1?=
 =?UTF-8?q?=E5=9E=8B=E7=BB=99=E6=9C=89=E6=84=8F=E4=B9=89=E6=80=BB=E7=BB=93?=
 =?UTF-8?q?=EF=BC=88=E4=B8=8D=E5=86=8D=E6=9C=BA=E6=A2=B0=E6=8F=90=E7=A4=BA?=
 =?UTF-8?q?=EF=BC=89=EF=BC=9Bweb=20=E8=B6=85=E6=97=B6=20600=E2=86=92900s?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 mcp_agent/agent.py  | 25 ++++++++++++++++++++++++-
 mcp_agent/config.py |  2 +-
 web/agent_api.py    |  2 +-
 3 files changed, 26 insertions(+), 3 deletions(-)

diff --git a/mcp_agent/agent.py b/mcp_agent/agent.py
index 41f4781..15bca89 100644
--- a/mcp_agent/agent.py
+++ b/mcp_agent/agent.py
@@ -32,6 +32,8 @@ SYSTEM_PROMPT = """你是手机自动化控制助手。你通过工具实时操
 5. 每次关键操作后再次 de_screenshot 验证结果，直到完成用户目标
 6. 完成或失败时用中文总结：做了什么、当前状态、需要用户注意的事项
 7. 设备不可用/操作失败时如实报告错误，不要臆测成功
+8. 效率：界面未变化时不要重复截图/点击同一位置；每步都要推进目标；
+   若连续 6 步无进展（截图内容未变/操作无效），停止并总结原因，不要空转
 
 可用工具清单将由系统提供。"""
 
@@ -214,7 +216,28 @@ class Agent:
             # 无工具调用：本轮即最终回答
             return full_content
 
-        return "（达到最大步骤数未完成，请检查操作是否卡在循环）"
+        # 步骤超限：不带工具让模型做最终总结（避免机械提示，给用户有意义的结论）
+        try:
+            _log.warning("达到最大步骤数，请求模型收尾总结")
+            saved_tools = self.tools_schema
+            self.tools_schema = []
+            self.messages.append({"role": "user",
+                                  "content": "已达最大操作步骤数，请立即用中文总结："
+                                             "已完成的部分、当前设备状态、未能完成的原因与下一步建议。"
+                                             "不要调用任何工具。"})
+            parts = []
+            async for chunk in self._chat_stream():
+                delta = (chunk.get("choices") or [{}])[0].get("delta") or {}
+                text = delta.get("content")
+                if text:
+                    parts.append(text)
+                    if on_delta:
+                        on_delta(text, "content")
+            self.tools_schema = saved_tools
+            summary = "".join(parts)
+            return summary or "（已达步骤上限，模型未能生成总结）"
+        except Exception as e:
+            return f"（已达最大步骤数，且收尾总结失败: {e}）"
 
     # ---------- 非流式（CLI） ----------
     async def run(self, prompt: str, serial: str = "") -> str:
diff --git a/mcp_agent/config.py b/mcp_agent/config.py
index 6c6c49c..74ed891 100644
--- a/mcp_agent/config.py
+++ b/mcp_agent/config.py
@@ -23,7 +23,7 @@ class AgentSettings:
     default_serial = _env("AGENT_DEFAULT_SERIAL", "")
 
     # Agent 循环上限与请求超时
-    max_steps = int(_env("AGENT_MAX_STEPS", "20"))
+    max_steps = int(_env("AGENT_MAX_STEPS", "40"))
     request_timeout = float(_env("AGENT_TIMEOUT", "120"))
 
     # system prompt 语言
diff --git a/web/agent_api.py b/web/agent_api.py
index 6d2db18..d1e1759 100644
--- a/web/agent_api.py
+++ b/web/agent_api.py
@@ -201,7 +201,7 @@ def _agent_thread(run_id, prompt, serial, cfg):
                                           on_delta=on_delta, on_tool=on_tool)
 
         # 整体超时保护：卡死时结束，释放单实例
-        answer = asyncio.run(asyncio.wait_for(_execute(), timeout=600))
+        answer = asyncio.run(asyncio.wait_for(_execute(), timeout=900))
         with _lock:
             _run["state"] = "done"
             _run["answer"] = answer