diff --git a/agentmain.py b/agentmain.py
index 6551b995b..99e9b215e 100644
--- a/agentmain.py
+++ b/agentmain.py
@@ -160,6 +160,7 @@ def run(self):
             if raw_query is None:
                 self.task_queue.task_done(); continue
             self.is_running = True; self._current_queue = display_queue
+            is_autonomous = raw_query.lstrip().startswith('[AUTO]')
             if len(raw_query) > 2000:
                 task_file = os.path.join(script_dir, 'temp', f'user_prompt_{os.getpid()}_{time.time_ns()}.md')
                 with open(task_file, 'w', encoding='utf-8') as f: f.write(raw_query)
@@ -170,7 +171,9 @@ def run(self):
             self.history.append(f"[USER]: {rquery}")
             sys_prompt = get_system_prompt() + '\n'.join(self.extra_sys_prompts) + getattr(self.llmclient.backend, 'extra_sys_prompt', '')
             if self.peer_hint: sys_prompt += f"\n[Peer] 用户提及其他会话/后台任务状态时: temp/model_responses/ (只找近期修改的文件尾部)\n"
-            handler = GenericAgentHandler(self, self.history, os.path.join(script_dir, 'temp'))
+            long_term_tool_enabled = any(t.get('function', {}).get('name') == 'start_long_term_update' for t in TOOLS_SCHEMA)
+            handler = GenericAgentHandler(self, self.history, os.path.join(script_dir, 'temp'),
+                                          long_term_update_pending=long_term_tool_enabled and not is_autonomous)
             if getattr(self, 'no_print', False): handler.print = lambda *a, **k: None
             if self.handler and 'key_info' in self.handler.working: 
                 ki = re.sub(r'\n\[SYSTEM\] 此为.*?工作记忆[。\n]*', '', self.handler.working['key_info'])  # 去旧
diff --git a/assets/tools_schema.json b/assets/tools_schema.json
index 81c52437a..1c8aa1377 100644
--- a/assets/tools_schema.json
+++ b/assets/tools_schema.json
@@ -66,7 +66,7 @@
   }},
   {"type": "function", "function": {
     "name": "start_long_term_update",
-    "description": "After a long task completes, distill long-term memory. Call when discovering info worth remembering (env facts/user prefs/lessons learned). Skip if memory already updated or in autonomous flow. Must call when a task that took 15+ turns is completed.",
+    "description": "After a long task completes, distill long-term memory. Call when discovering info worth remembering (env facts/user prefs/lessons learned). Skip if memory already updated or in autonomous flow. Must call when a task that took 15 or more turns is completed.",
     "parameters": {"type": "object", "properties": {}}}
   }
 ]
diff --git a/assets/tools_schema_cn.json b/assets/tools_schema_cn.json
index 98fbc211b..9a6107045 100644
--- a/assets/tools_schema_cn.json
+++ b/assets/tools_schema_cn.json
@@ -67,7 +67,7 @@
   }},
   {"type": "function", "function": {
     "name": "start_long_term_update",
-    "description": "准备开始提炼记忆。发现值得长期记忆的信息（环境事实/用户偏好/避坑经验）时调用此工具。已记忆更新或在自主流程内时无需调用。超15轮完成的任务必须调用以沉淀经验",
+    "description": "准备开始提炼记忆。发现值得长期记忆的信息（环境事实/用户偏好/避坑经验）时调用此工具。已记忆更新或在自主流程内时无需调用。达到15轮后完成的任务必须调用以沉淀经验",
     "parameters": {"type": "object", "properties": {}}}
   }
-]
\ No newline at end of file
+]
diff --git a/ga.py b/ga.py
index e1ee909f2..13bbf4d29 100644
--- a/ga.py
+++ b/ga.py
@@ -279,10 +279,13 @@ def consume_file(dr, file):
 
 class GenericAgentHandler(BaseHandler):
     '''Generic Agent 工具库，包含多种工具的实现。工具函数自动加上了 do_ 前缀。实际工具名没有前缀。'''
-    def __init__(self, parent, last_history=None, cwd='./temp'):
+    LONG_TERM_UPDATE_MIN_TURN = 15
+
+    def __init__(self, parent, last_history=None, cwd='./temp', long_term_update_pending=True):
         self.parent = parent
         self.working = {}
         self.cwd = cwd;  self.current_turn = 0
+        self.long_term_update_pending = long_term_update_pending
         self.history_info = last_history if last_history else []
         self.code_stop_signal = []
         self._done_hooks = []
@@ -520,12 +523,18 @@ def do_no_tool(self, args, response):
             remaining = self._check_plan_completion()
             if remaining == 0:
                 self._exit_plan_mode(); yield "[Info] Plan完成：plan.md中0个[ ]残留，退出plan模式。\n"
+
+        if (self.current_turn >= self.LONG_TERM_UPDATE_MIN_TURN
+                and self.long_term_update_pending):
+            self.long_term_update_pending = False
+            yield "[Info] Auto-calling start_long_term_update before exit.\n"
+            outcome = yield from self.do_start_long_term_update({}, response)
+            return StepOutcome(None, next_prompt=f'{outcome.data}\n{outcome.next_prompt}')
         
         #yield "[Info] Final response to user.\n"
         return StepOutcome(response, next_prompt=None)
-    
-    def do_start_long_term_update(self, args, response):
-        '''Agent觉得当前任务完成后有重要信息需要记忆时调用此工具。'''
+
+    def _begin_long_term_update(self):
         prompt = '''### [总结提炼经验] 既然你觉得当前任务有重要信息需要记忆，请提取最近一次任务中【事实验证成功且长期有效】的环境事实、用户偏好、重要步骤，更新记忆。
 本工具是标记开启结算过程，若已在更新记忆过程或没有值得记忆的点，忽略本次调用。
 **如果没有经验证的，未来能用上的信息，忽略本次调用！**
@@ -535,13 +544,19 @@ def do_start_long_term_update(self, args, response):
 **禁止**：临时变量、具体推理过程、未验证信息、通用常识、你可以轻松复现的细节、只是做了但没有验证的信息
 **操作**：严格遵循提供的L0的记忆更新SOP。先 `file_read` 看现有 → 判断类型 → 最小化更新 → 无新内容跳过，保证对记忆库最小局部修改。\n
 ''' + get_global_memory()
-        yield "[Info] Start distilling good memory for long-term storage.\n"
         path = './memory/memory_management_sop.md'
         if os.path.exists(path): result = 'This is L0:\n' + file_read(path, show_linenos=False)
         else: result = "Memory Management SOP not found. Do not update memory."
-        if self.current_turn < 10: result, prompt = 'start_long_term_update is only used after completing a long turn task!', '\n'
         return StepOutcome(result, next_prompt=prompt)
 
+    def do_start_long_term_update(self, args, response):
+        '''Agent觉得当前任务完成后有重要信息需要记忆时调用此工具。'''
+        if self.current_turn < 10:
+            return StepOutcome('start_long_term_update is only used after completing a long turn task!', next_prompt='\n')
+        yield "[Info] Start distilling good memory for long-term storage.\n"
+        self.long_term_update_pending = False
+        return self._begin_long_term_update()
+
     def _fold_earlier(self, lines):
         FALLBACK = '直接回答了用户问题'
         parts, cnt, last = [], 0, ''
diff --git a/tests/test_long_term_update_gate.py b/tests/test_long_term_update_gate.py
new file mode 100644
index 000000000..56c141503
--- /dev/null
+++ b/tests/test_long_term_update_gate.py
@@ -0,0 +1,103 @@
+import importlib.util
+import sys
+import unittest
+from pathlib import Path
+from types import SimpleNamespace
+from unittest.mock import patch
+
+
+ROOT = Path(__file__).resolve().parents[1]
+
+
+def load_module(name, path):
+    spec = importlib.util.spec_from_file_location(name, path)
+    module = importlib.util.module_from_spec(spec)
+    sys.modules[name] = module
+    spec.loader.exec_module(module)
+    return module
+
+
+previous_agent_loop = sys.modules.get('agent_loop')
+real_agent_loop = load_module('_long_term_gate_agent_loop', ROOT / 'agent_loop.py')
+sys.modules['agent_loop'] = real_agent_loop
+try:
+    ga = load_module('_long_term_gate_ga', ROOT / 'ga.py')
+finally:
+    if previous_agent_loop is None:
+        del sys.modules['agent_loop']
+    else:
+        sys.modules['agent_loop'] = previous_agent_loop
+
+GenericAgentHandler = ga.GenericAgentHandler
+
+
+class Parent:
+    def get_ctx_multiplier(self):
+        return 1
+
+
+def exhaust(generator):
+    try:
+        while True:
+            next(generator)
+    except StopIteration as exc:
+        return exc.value
+
+
+class LongTermUpdateGateTests(unittest.TestCase):
+    def handler(self, enabled=True):
+        return GenericAgentHandler(Parent(), cwd='./temp', long_term_update_pending=enabled)
+
+    def finish(self, handler, turn):
+        handler.current_turn = turn
+        response = SimpleNamespace(content='Task complete.', thinking='')
+        return exhaust(handler.do_no_tool({}, response))
+
+    def test_turn_before_threshold_exits_normally(self):
+        handler = self.handler()
+        outcome = self.finish(handler, 14)
+        self.assertIsNone(outcome.next_prompt)
+        self.assertTrue(handler.long_term_update_pending)
+
+    @patch.object(ga, 'file_read', return_value='memory SOP')
+    @patch.object(ga, 'get_global_memory', return_value='memory index')
+    @patch.object(ga.os.path, 'exists', return_value=True)
+    def test_threshold_starts_evaluation_once(self, _exists, _memory, _read):
+        handler = self.handler()
+        outcome = self.finish(handler, 15)
+        self.assertIn('memory SOP', outcome.next_prompt)
+        self.assertIn('总结提炼经验', outcome.next_prompt)
+        self.assertFalse(handler.long_term_update_pending)
+
+        second = self.finish(handler, 16)
+        self.assertIsNone(second.next_prompt)
+
+    def test_disabled_gate_exits_normally(self):
+        handler = self.handler(enabled=False)
+        outcome = self.finish(handler, 20)
+        self.assertIsNone(outcome.next_prompt)
+        self.assertFalse(handler.long_term_update_pending)
+
+    @patch.object(ga, 'file_read', return_value='memory SOP')
+    @patch.object(ga, 'get_global_memory', return_value='memory index')
+    @patch.object(ga.os.path, 'exists', return_value=True)
+    def test_explicit_update_consumes_completion_gate(self, _exists, _memory, _read):
+        handler = self.handler()
+        handler.current_turn = 10
+        exhaust(handler.do_start_long_term_update({}, SimpleNamespace()))
+        self.assertFalse(handler.long_term_update_pending)
+
+        outcome = self.finish(handler, 20)
+        self.assertIsNone(outcome.next_prompt)
+        self.assertFalse(handler.long_term_update_pending)
+
+    def test_rejected_early_call_keeps_completion_gate_pending(self):
+        handler = self.handler()
+        handler.current_turn = 9
+        outcome = exhaust(handler.do_start_long_term_update({}, SimpleNamespace()))
+        self.assertEqual('\n', outcome.next_prompt)
+        self.assertTrue(handler.long_term_update_pending)
+
+
+if __name__ == '__main__':
+    unittest.main()