diff --git a/apps/api/src/services/openclaw.py b/apps/api/src/services/openclaw.py
index 9705beda..3ecaa7b7 100644
--- a/apps/api/src/services/openclaw.py
+++ b/apps/api/src/services/openclaw.py
@@ -1684,7 +1684,8 @@ Focus on:
                     import asyncio
                     await asyncio.sleep(2)  # 重試前等 2 秒
 
-        # 重試全部失敗 — 仍然標記 enabled 並顯示失敗狀態（不隱藏）
+        # 重試全部失敗 — fallback 到 Gemini 模擬 tool calling
+        # 2026-04-08 ogt: NIM 完全不可用時，改用 Gemini 產生執行方案（不可跳過）
         if last_error is not None:
             logger.error(
                 "nemotron_collaboration_exhausted",
@@ -1692,10 +1693,18 @@ Focus on:
                 error=str(last_error),
                 retries=max_retries,
             )
-            proposal["nemotron_enabled"] = True  # 🔴 仍然顯示區塊，讓統帥知道失敗了
-            proposal["nemotron_tools"] = []
-            proposal["nemotron_validation"] = f"❌ {max_retries}次重試均失敗"
-            proposal["nemotron_latency_ms"] = 0.0
+            logger.info("nemotron_fallback_gemini_start", incident_id=incident_id)
+            gemini_fallback_result = await self._call_nemotron_tools_via_gemini(
+                incident_id=incident_id,
+                reasoning=proposal.get("reasoning", ""),
+                target_resource=proposal.get("target_resource", ""),
+                suggested_action=proposal.get("action", ""),
+                namespace=proposal.get("namespace", "awoooi-prod"),
+            )
+            proposal["nemotron_enabled"] = True
+            proposal["nemotron_tools"] = gemini_fallback_result.get("tools", [])
+            proposal["nemotron_validation"] = gemini_fallback_result.get("validation", "⚠️ Gemini 代理")
+            proposal["nemotron_latency_ms"] = gemini_fallback_result.get("latency_ms", 0.0)
 
         return proposal, provider, True
 
@@ -1860,6 +1869,78 @@ Focus on:
             )
             raise
 
+    async def _call_nemotron_tools_via_gemini(
+        self,
+        incident_id: str,
+        reasoning: str,
+        target_resource: str,
+        suggested_action: str,
+        namespace: str = "awoooi-prod",
+    ) -> dict:
+        """
+        NIM 完全不可用時，由 Gemini 代理產生 tool calling 執行方案。
+        2026-04-08 ogt: NIM timeout 後的唯一 fallback，不可跳過。
+
+        Returns: {"tools": [...], "validation": str, "latency_ms": float}
+        """
+        import time as _time
+        start_time = _time.time()
+
+        prompt = f"""你是 K8s SRE 專家。根據以下分析，輸出對應的 kubectl 操作指令（JSON 格式）。
+
+Incident ID: {incident_id}
+目標資源: {target_resource}
+Namespace: {namespace}
+建議操作: {suggested_action}
+分析摘要: {reasoning[:300]}
+
+請輸出以下 JSON 格式（只輸出 JSON，不要其他文字）：
+{{
+  "tool_name": "restart_deployment 或 scale_deployment 或 no_action",
+  "deployment_name": "部署名稱",
+  "namespace": "{namespace}",
+  "reason": "一句話說明原因"
+}}"""
+
+        try:
+            text, success, _, _ = await self._call_gemini(prompt)
+            latency_ms = (_time.time() - start_time) * 1000
+
+            if not success:
+                logger.warning("nemotron_gemini_fallback_failed", incident_id=incident_id, error=text)
+                return {"tools": [], "validation": "❌ NIM + Gemini 均不可用", "latency_ms": latency_ms}
+
+            import json as _json
+            data = _json.loads(text)
+            tool_name = data.get("tool_name", "no_action")
+            tools = []
+            if tool_name != "no_action":
+                tools = [{
+                    "tool": tool_name,
+                    "args": {
+                        "deployment_name": data.get("deployment_name", target_resource),
+                        "namespace": data.get("namespace", namespace),
+                    },
+                    "valid": True,
+                }]
+
+            logger.info(
+                "nemotron_gemini_fallback_success",
+                incident_id=incident_id,
+                tool=tool_name,
+                latency_ms=latency_ms,
+            )
+            return {
+                "tools": tools,
+                "validation": "✅ Gemini 代理驗證通過",
+                "latency_ms": latency_ms,
+            }
+
+        except Exception as e:
+            latency_ms = (_time.time() - start_time) * 1000
+            logger.error("nemotron_gemini_fallback_error", incident_id=incident_id, error=str(e))
+            return {"tools": [], "validation": f"❌ Gemini 代理失敗: {str(e)[:50]}", "latency_ms": latency_ms}
+
     # =========================================================================
     # Shadow Mode Auto-Tuning
     # =========================================================================
diff --git a/scripts/ops/docker-health-monitor.sh b/scripts/ops/docker-health-monitor.sh
index 4c2f42a2..f0a739d4 100755
--- a/scripts/ops/docker-health-monitor.sh
+++ b/scripts/ops/docker-health-monitor.sh
@@ -25,6 +25,9 @@ fi
 # 冷卻期：避免同一容器在短時間內重複發送 webhook（去重，非修復冷卻）
 : "${SEND_COOLDOWN_SECONDS:=300}"
 : "${COOLDOWN_DIR:=/tmp/docker-health-monitor-cooldown}"
+# 排除清單：逗號分隔，支援 glob（如 signoz-*）
+# 用途：init containers、永久停用服務、已知 exited 但不需告警的容器
+: "${EXCLUDE_CONTAINERS:=signoz-telemetrystore-migrator,signoz-clickhouse,signoz-init-clickhouse}"
 
 mkdir -p "$COOLDOWN_DIR"
 
@@ -135,6 +138,21 @@ check_containers() {
         # 跳過 header 或空行
         [[ -z "$container_name" ]] && continue
 
+        # 排除清單檢查（EXCLUDE_CONTAINERS 逗號分隔）
+        local excluded=false
+        IFS=',' read -ra EXCLUDES <<< "$EXCLUDE_CONTAINERS"
+        for pattern in "${EXCLUDES[@]}"; do
+            pattern="${pattern// /}"  # trim spaces
+            [[ -z "$pattern" ]] && continue
+            # shellcheck disable=SC2254
+            case "$container_name" in
+                $pattern) excluded=true; break ;;
+            esac
+        done
+        if $excluded; then
+            continue
+        fi
+
         local needs_alert=false
         local detected_status=""