goal 异常退出记 interrupted，可断点续跑

长 goal 实弹（2026-10-28）：宿主机内存挤兑致回合挂起 59 分钟，恢复时 goal 进程异常退出、无结论行，云端旧语义（returncode 非 0 一律 failed）把可续跑的断点判成终局。

修法：host.py run_goal_loop 回合异常统一记 interrupted 并打印 goal interrupted after rounds 标记行；cloud/sandbox.py 四处 goal 解析同步新语义（标记或无结论非 0 退出 -> interrupted，有结论的非 0 退出仍 failed）；worker 终局审计名与状态对齐。单测：sandbox 新测 5 个、host 新测 1 个、更新旧测 1 个；三套件 cloud 1017、host 61、deploy 29 全绿。只本地提交，未推未部署。

tests/test_host.py

@@ -345,6 +345,42 @@ class RetryAndBudgetTest(unittest.TestCase):
             self.assertIn("budget-exceeded",
                           Path(notify_log).read_text())
 
+    def test_goal_turn_exception_is_interrupted_not_failed(self):
+        # 2026-10-28 长 goal 实弹：回合抛错必须记 interrupted +
+        # 打印标记行（云端据此可断点续跑），绝不能向上传播致
+        # 进程非 0 退出、无结论行（旧语义云端判 terminal failed）。
+        import contextlib
+        import io
+
+        class FakeServer:
+            def run_turn(self, *args, **kwargs):
+                raise RuntimeError("relay boom")
+
+        with tempfile.TemporaryDirectory() as tmp:
+            state_path = os.path.join(tmp, "host-state.json")
+            state = {"tasks": {}, "automations": [], "usage": {}}
+            task = {"name": "g", "thread_id": "x", "cwd": tmp,
+                    "sandbox": None, "max_retries": 0, "retry_count": 0,
+                    "turns": []}
+            state["tasks"]["g"] = task
+            buf = io.StringIO()
+            with contextlib.redirect_stdout(buf):
+                result = host.run_goal_loop(
+                    FakeServer(), state, state_path,
+                    task, "make it", "true", 3, "m",
+                    "low", None, timeout=600)
+            self.assertTrue(result["interrupted"])
+            self.assertFalse(result["ok"])
+            self.assertEqual(task["status"], "interrupted")
+            self.assertEqual(task["goal_rounds_done"], 0)
+            self.assertIn("relay boom",
+                          task["goal_interrupted_reason"])
+            self.assertIn("goal interrupted after rounds",
+                          buf.getvalue())
+            saved = json.loads(Path(state_path).read_text())
+            self.assertEqual(saved["tasks"]["g"]["status"],
+                             "interrupted")
+
 
 class TurnErrorFromHistoryTest(unittest.TestCase):
     def _server_with_history(self, tmp, error_json):
@@ -934,16 +970,21 @@ class GoalLoopLongTaskTest(unittest.TestCase):
             state = {"tasks": {}, "automations": []}
             task = self._task(tmp)
             state["tasks"]["g"] = task
-            with self.assertRaises(RuntimeError):
-                host.run_goal_loop(FakeServer(), state, state_path,
-                                   task, "make it", "true", 3, "m", "low",
-                                   None, cfg={}, timeout=600)
+            # 2026-10-28 起：回合抛错不再向上传播（旧断言），而
+            # 是记 interrupted + 标记行（云端可断点续跑）；首轮
+            # 前落盘的字段仍必须都在。
+            result = host.run_goal_loop(FakeServer(), state, state_path,
+                                        task, "make it", "true", 3, "m",
+                                        "low", None, cfg={}, timeout=600)
+            self.assertTrue(result["interrupted"])
             saved = host.load_state(state_path)
             saved_task = saved["tasks"]["g"]
             self.assertEqual(saved_task["goal"], "make it")
             self.assertEqual(saved_task["verify_command"], "true")
             self.assertEqual(saved_task["goal_rounds_done"], 0)
-            self.assertEqual(saved_task["status"], "goal-running")
+            self.assertEqual(saved_task["status"], "interrupted")
+            self.assertIn("killed during turn",
+                          saved_task["goal_interrupted_reason"])
 
     def test_goal_budget_stop_is_interrupted_not_unmet(self):
         class FakeServer: