fix(aiops): block k8s playbooks for host repair

apps/api/src/services/auto_repair_service.py

@@ -337,6 +337,23 @@ class AutoRepairService:
                 blocked_by="NOT_APPROVED",
             )
 
+        if self._is_host_or_backup_incident(incident) and self._playbook_has_k8s_steps(best_match.playbook):
+            logger.warning(
+                "auto_repair_blocked_host_backup_k8s_playbook",
+                incident_id=incident.incident_id,
+                playbook_id=best_match.playbook.playbook_id,
+                alert_category=getattr(incident, "alert_category", None),
+            )
+            return AutoRepairDecision(
+                can_auto_repair=False,
+                playbook=best_match.playbook,
+                reason=(
+                    "主機/備份類告警禁止執行 K8s Playbook；"
+                    "需改走 SSH 診斷或緊急介入"
+                ),
+                blocked_by="HOST_BACKUP_K8S_PLAYBOOK",
+            )
+
         # 5. 可以自動修復
         logger.info(
             "auto_repair_approved",
@@ -676,6 +693,29 @@ class AutoRepairService:
 
         return max_risk
 
+    def _is_host_or_backup_incident(self, incident: Incident) -> bool:
+        """主機/備份類事件只能走 SSH/只讀診斷，不允許 K8s rollout 類修復。"""
+
+        category = (getattr(incident, "alert_category", None) or "").lower()
+        if category in {"host_resource", "backup_failure"}:
+            return True
+
+        for signal in incident.signals or []:
+            labels = signal.labels or {}
+            alertname = str(labels.get("alertname") or signal.alert_name or "")
+            if alertname.startswith("HostBackup") or alertname.startswith("Host"):
+                return True
+        return False
+
+    def _playbook_has_k8s_steps(self, playbook: Playbook) -> bool:
+        """檢查 Playbook 是否包含 K8s 指令，避免主機告警誤執行 deployment 操作。"""
+
+        for step in playbook.repair_steps:
+            command = (step.command or "").strip().lower()
+            if step.action_type == ActionType.KUBECTL or command.startswith("kubectl "):
+                return True
+        return False
+
     def _risk_exceeds_threshold(self, risk: RiskLevel) -> bool:
         """檢查風險是否超過自動修復門檻"""
         high_risks = {RiskLevel.HIGH, RiskLevel.CRITICAL}