feat: Turn轮次策略解耦为可插拔 Policy Hook

ga.py

@@ -7,6 +7,7 @@
 sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), '..')))
 
 from agent_loop import BaseHandler, StepOutcome, json_default
+from turn_policy import register_turn_policies
 script_dir = os.path.dirname(os.path.abspath(__file__))
 
 def code_run(code, code_type="python", timeout=60, cwd=None, code_cwd=None, stop_signal=None, maxlen=10000):
@@ -268,6 +269,8 @@ def __init__(self, parent, last_history=None, cwd='./temp'):
         self.history_info = last_history if last_history else []
         self.code_stop_signal = []
         self._done_hooks = []
+        # Turn policy hooks - pluggable strategy chain (from turn_policy.py)
+        register_turn_policies(self)
 
     def _get_abs_path(self, path):
         if not path: return ""
@@ -543,7 +546,7 @@ def _get_anchor_prompt(self, skip=False):
             try: print(prompt)
             except: pass
         return prompt
-    
+
     def turn_end_callback(self, response, tool_calls, tool_results, turn, next_prompt, exit_reason):
         _c = re.sub(r'```.*?```|<thinking>.*?</thinking>', '', response.content, flags=re.DOTALL)
         rsumm = re.search(r"<summary>(.*?)</summary>", _c, re.DOTALL)
@@ -559,15 +562,9 @@ def turn_end_callback(self, response, tool_calls, tool_results, turn, next_promp
         self.history_info.append(f'[Agent] {summary}')
         _plan = self._in_plan_mode()
 
-        if turn % 75 == 0 and (not _plan):
-            next_prompt += f"\n\n[DANGER] 已连续执行第 {turn} 轮。必须总结情况进行ask_user，不允许继续重试。"
-        elif turn % 7 == 0:
-            next_prompt += f"\n\n[DANGER] 已连续执行第 {turn} 轮。禁止无效重试。若无有效进展，必须切换策略：1. 探测物理边界 2. 请求用户协助。如有需要，可调用 update_working_checkpoint 保存关键上下文。"
-        elif turn % 10 == 0: next_prompt += get_global_memory()
-
-        if _plan and turn >= 10 and turn % 5 == 0:
-            next_prompt = f"[Plan Hint] 正在计划模式。必须 file_read({_plan}) 确认当前步骤，回复开头引用：📌 当前步骤：...\n\n" + next_prompt
-        if _plan and turn >= 120: next_prompt += f"\n\n[DANGER] Plan模式已运行 {turn} 轮，已达上限。必须 ask_user 汇报进度并确认是否继续。"
+        # Policy hook chain - each policy returns "" or a string to append
+        for policy in self._turn_policies:
+            next_prompt += policy(turn, _plan, next_prompt) or ""
 
         injkeyinfo = consume_file(self.parent.task_dir, '_keyinfo')
         injprompt = consume_file(self.parent.task_dir, '_intervene')

turn_policy.py

@@ -0,0 +1,61 @@
+
+import os
+
+def policy_danger_ask_user(turn, _plan, next_prompt):
+    """每75轮强制ask_user（非plan模式）"""
+    if turn % 75 == 0 and (not _plan):
+        return (
+            "\n\n[DANGER] 已连续执行第 " + str(turn) + " 轮。"
+            "必须总结情况进行ask_user，不允许继续重试。"
+        )
+    return ""
+
+
+def policy_danger_retry(turn, _plan, next_prompt):
+    """每7轮禁止无效重试"""
+    if turn % 7 == 0:
+        return (
+            "\n\n[DANGER] 已连续执行第 " + str(turn) + " 轮。禁止无效重试。"
+            "若无有效进展，必须切换策略：1. 探测物理边界 "
+            "2. 请求用户协助。如有需要，可调用 update_working_checkpoint "
+            "保存关键上下文。"
+        )
+    return ""
+
+
+def policy_inject_memory(turn, _plan, next_prompt):
+    """每10轮注入全局记忆"""
+    if turn % 10 == 0:
+        from ga import get_global_memory
+        return get_global_memory()
+    return ""
+
+
+def policy_plan_limit(turn, _plan, next_prompt):
+    """Plan模式上限检测"""
+    if _plan and turn >= 10 and turn % 5 == 0 and turn <= 110:
+        return (
+            "[Plan Hint] 正在计划模式。必须 file_read(" + str(_plan) + ") "
+            "确认当前步骤，回复开头引用：当前步骤：...\n\n"
+        )
+    if _plan and turn >= 120:
+        return (
+            "\n\n[DANGER] Plan模式已运行 " + str(turn) + " 轮，已达上限。"
+            "必须 ask_user 汇报进度并确认是否继续。"
+        )
+    return ""
+
+# Default turn policies that ship with the agent
+DEFAULT_TURN_POLICIES = [
+    policy_danger_ask_user,
+    policy_danger_retry,
+    policy_inject_memory,
+    policy_plan_limit,
+]
+
+def register_turn_policies(handler, policies=None):
+    """Register turn policies onto a GenericAgentHandler instance.
+    """
+    if policies is None:
+        policies = DEFAULT_TURN_POLICIES
+    handler._turn_policies = list(policies)