fix(sensors+drift+repair-card): 全景修復三個節點問題

Fix 1: sensors 7/8 失敗 — SSH host 短名展開 (pre_decision_investigator.py)
  根因: Prometheus instance label 為 "110:9100"，split(":")[0]="110"
        SSH_MCP_ALLOWED_HOSTS 存完整 IP "192.168.0.110" → 7 個 SSH 工具全部失敗
  修復: 加入 _SHORT_HOST_MAP，"110"→"192.168.0.110"，四台主機全覆蓋

Fix 2: Config Drift 誤報 — K8s 預設欄位加入白名單 (drift_detector.py)
  根因: kubectl rollout restart 後 restartedAt annotation 被偵測為 "medium" drift
        restartPolicy/dnsPolicy/terminationGracePeriodSeconds 等 K8s 自動填入欄位未白名單
  修復: _DEFAULT_ALLOWLIST_FIELDS 加入 13 個 K8s 執行時自動填入欄位

Fix 3: 修復請求卡內容垃圾 — fallback 帶入真實 error context (failure_watcher.py)
  根因: LLM 分析失敗時 root_cause = "規則引擎分類: K8S_ERROR"（無任何有用資訊）
  修復: fallback 改為 "[K8S_ERROR] {operation_type} 在 {target_resource} 失敗\n錯誤：{error_message[:200]}"

2026-04-16 ogt + Claude Sonnet 4.6（亞太）

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

apps/api/src/services/drift_detector.py

@@ -35,6 +35,23 @@ _DEFAULT_ALLOWLIST_FIELDS = frozenset([
     "metadata.generation",
     "metadata.uid",
     "status",
+    # K8s 執行時自動填入的欄位（Git manifest 不指定，K8s 注入預設值或運行時資訊）
+    # 2026-04-16 ogt + Claude Sonnet 4.6: 修復 kubectl rollout restart 觸發假 drift 告警
+    # 根因：awoooi-web rollout restart 後 restartedAt annotation 被偵測為 "medium" drift
+    "spec.template.metadata.annotations",          # kubectl.kubernetes.io/restartedAt
+    "spec.template.metadata.creationTimestamp",
+    "spec.template.spec.restartPolicy",            # K8s 預設: Always（Git 不指定）
+    "spec.template.spec.dnsPolicy",                # K8s 預設: ClusterFirst
+    "spec.template.spec.terminationGracePeriodSeconds",  # K8s 預設: 30
+    "spec.template.spec.schedulerName",            # K8s 預設: default-scheduler
+    "spec.strategy.rollingUpdate",                 # K8s 預設: maxSurge=25%
+    "spec.strategy.type",                          # K8s 預設: RollingUpdate
+    "spec.progressDeadlineSeconds",                # K8s 預設: 600
+    "spec.revisionHistoryLimit",                   # K8s 預設: 10
+    "metadata.creationTimestamp",
+    "spec.template.spec.containers[*].terminationMessagePath",
+    "spec.template.spec.containers[*].terminationMessagePolicy",
+    "spec.template.spec.containers[*].imagePullPolicy",  # K8s 預設: IfNotPresent
 ])
 
 # 關鍵欄位（必須立即告警）

apps/api/src/services/failure_watcher.py

@@ -299,12 +299,18 @@ class FailureWatcherService(IFailureWatcher):
             return llm_analysis
 
         # LLM 失敗，使用規則引擎結果
+        # 2026-04-16 ogt + Claude Sonnet 4.6: 修復 root_cause 只顯示 "規則引擎分類: K8S_ERROR"
+        # 根因：LLM 分析失敗時未帶入實際 error_message，用戶看到的卡片無任何有用資訊
+        _error_preview = (error_message[:200] if error_message else "未知錯誤").strip()
         return {
             "classification": classification,
-            "root_cause": f"規則引擎分類: {classification}",
+            "root_cause": (
+                f"[{classification}] {operation_type} 操作在 {target_resource} 失敗\n"
+                f"錯誤：{_error_preview}"
+            ),
             "suggested_repair": self._suggest_repair(classification),
             "risk_level": risk_level,
-            "confidence": 0.5,  # 規則引擎信心度較低
+            "confidence": 0.4,
         }
 
     async def execute_auto_repair(

apps/api/src/services/pre_decision_investigator.py

@@ -382,14 +382,30 @@ def _get_labels(incident: "Incident") -> dict[str, Any]:
     return {}
 
 
+_SHORT_HOST_MAP: dict[str, str] = {
+    "110": "192.168.0.110",
+    "120": "192.168.0.120",
+    "121": "192.168.0.121",
+    "188": "192.168.0.188",
+}
+"""
+Prometheus instance label 使用短主機名（如 "110:9100"），
+SSH_MCP_ALLOWED_HOSTS 使用完整 IP（如 "192.168.0.110"）。
+此映射表做轉換，避免 SSH 工具 "Host 'X' not in SSH_MCP_ALLOWED_HOSTS" 失敗。
+2026-04-16 ogt + Claude Sonnet 4.6: 修復 sensors 7/8 失敗根因
+"""
+
+
 def _build_tool_params(incident: "Incident") -> dict[str, Any]:
     """從 Incident 提取 MCP 工具呼叫所需的公共參數。"""
     labels = _get_labels(incident)
+    raw_host = labels.get("instance", "").split(":")[0] or labels.get("host", "")
+    host = _SHORT_HOST_MAP.get(raw_host, raw_host)  # 短名 → 完整 IP
     return {
         "namespace": labels.get("namespace", "awoooi-prod"),
         "pod_name": labels.get("pod", labels.get("name", "")),
         "deployment": labels.get("deployment", ""),
-        "host": labels.get("instance", "").split(":")[0] or labels.get("host", ""),
+        "host": host,
         "container": labels.get("container", labels.get("name", "")),
         "alertname": labels.get("alertname", ""),
     }