fix(aiops): route backup decisions through ssh

.agents/skills/04-awoooi-devops-commander.md

@@ -1344,6 +1344,43 @@ Architecture Review 發現的安全要求（2026-04-11）：
 
 3. **群組 B 工具需 trust_score >= 0.8**（硬編碼守衛）
 
+### Host/Backup SSH Route Invariants (2026-05-01)
+
+`backup_failure` is a host-layer category. Keep it aligned anywhere
+`host_resource` is routed, especially:
+
+- `DecisionManager`: non-`kubectl` actions must route to SSH MCP before
+  `parse_kubectl_action()`. Otherwise SSH diagnosis strings with shell syntax
+  are blocked as `forbidden_shell_metachar`.
+- `DecisionManager`: `kubectl` actions from `host_resource` or
+  `backup_failure` must be blocked and escalated to emergency intervention.
+- `AutoRepairService`: host/backup incidents must not fall back to K8s
+  rollout Playbooks.
+
+Runtime baseline for host/backup repair:
+
+```bash
+kubectl -n awoooi-prod get secret ssh-mcp-key awoooi-repair-ssh-key awoooi-repair-known-hosts
+
+kubectl -n awoooi-prod exec deploy/awoooi-api -- sh -lc '
+  ls -l /run/secrets/ssh_mcp_key /etc/ssh-mcp/known_hosts \
+        /etc/repair-ssh/id_ed25519 /etc/repair-known-hosts/known_hosts
+'
+
+kubectl -n awoooi-prod exec deploy/awoooi-api -- sh -lc '
+  for h in 192.168.0.110 192.168.0.120 192.168.0.121; do
+    ssh -i /run/secrets/ssh_mcp_key -o BatchMode=yes \
+      -o StrictHostKeyChecking=yes -o ConnectTimeout=5 wooo@$h "echo OK:$h"
+  done
+  ssh -i /run/secrets/ssh_mcp_key -o BatchMode=yes \
+    -o StrictHostKeyChecking=yes -o ConnectTimeout=5 ollama@192.168.0.188 "echo OK:188"
+'
+```
+
+`awoooi-executor` RBAC must include read-only backup evidence:
+`jobs.batch`, `cronjobs.batch`, PVCs, and Velero backup resources. It may patch
+`statefulsets.apps` / `daemonsets.apps` only for safe rollout restart.
+
 ---
 
 ## 🚀 Sprint C — DR 備份與恢復 (2026-04-11) ✅
@@ -1503,4 +1540,3 @@ ssh-mcp-key        ✅ (ssh_mcp_key + known_hosts)
 
 ### Runbook
 `docs/runbooks/ssh-mcp-setup.md`
-

.agents/skills/05-awoooi-sre-qa.md

@@ -786,6 +786,31 @@ kubectl -n awoooi-prod logs -l app=awoooi-api --tail=50 | \
 | `Permission denied (publickey)` | known_hosts 缺少該主機 | Pod exec SSH 看錯誤訊息 |
 | `Load key ... Permission denied` | fsGroup 未設定 | Pod exec `ls -la /etc/repair-ssh/` |
 | `Connection refused/timeout` | NetworkPolicy 封鎖 22 | Pod exec `ssh -v` 看連線過程 |
+| `forbidden_shell_metachar` 且 action 是 `ssh ... '...'` | host/backup category 沒在 DecisionManager kubectl parser 前路由 SSH | 查 `alert_category` 是否為 `backup_failure`，確認 `_is_host_layer_ssh_category()` 覆蓋 |
+
+### Telegram 按鈕 E2E 檢查 (2026-05-01)
+
+告警卡片按鈕不是純 UI。每個按鈕都必須能在
+`callback_action_spec.yaml` 找到 callback pattern，並經
+`callback_dispatcher.py` 路由到實際 handler。
+
+| 卡片/情境 | 必要按鈕 | 預期處理 |
+|-----------|----------|----------|
+| Approval / LLM action | approve, reject, details, ignore | 寫 approval decision、執行或拒絕、查詳情、忽略告警 |
+| Auto repair unavailable / emergency | investigate, escalate/assign, rollback when applicable | 通知人工/AI Agent 介入，不可靜默 |
+| Drift TYPE-4D | view diff, adopt, rollback, ignore | 看 diff、採納變更、回滾、忽略 |
+| Backup / host diagnosis | restart only when rule allows, charts/logs/details, cleanup when safe | 不得提供 K8s-only repair button 當 host/backup 主動作 |
+| Post-verification degraded/failed | rollback proposal, investigate, details | 不自動 rollback，需人工或 emergency AI Agent 接手 |
+
+Regression test target: button callback names emitted by `telegram_gateway.py`
+must stay in sync with `callback_action_spec.yaml`; stale buttons are a
+production bug because Telegram cards can outlive code deploys.
+
+Provider name drift is also a ghost-button bug. `callback_action_spec.yaml`
+may use friendly names (`k8s`, `ssh`), but dispatcher must normalize to actual
+registered MCP providers (`kubernetes`, `ssh_host`) before `get_provider()`.
+`backup_failure` cards must expose read-only diagnostics before any write
+action: host disk, backup jobs, and Velero backup status.
 
 ---
 

apps/api/src/services/callback_action_spec.yaml

@@ -22,7 +22,7 @@
 #     description: <說明>
 
 version: "1.0"
-last_updated: "2026-04-14"
+last_updated: "2026-05-01"
 
 actions:
   # ==========================================================================
@@ -188,6 +188,53 @@ actions:
     timeout_sec: 1
     description: "返回飛輪儀表板 URL"
 
+  backup_check_host_disk:
+    label: "查主機磁碟"
+    emoji: "💾"
+    risk: low
+    callback_format: info
+    category: backup_failure
+    mcp:
+      provider: ssh
+      tool: ssh_get_disk_usage
+      params:
+        host: "{labels.instance}"
+    reply_format: code
+    timeout_sec: 8
+    description: "備份失敗時檢查主機磁碟容量與 Docker 目錄大小"
+
+  backup_check_jobs:
+    label: "查備份 Job"
+    emoji: "📦"
+    risk: low
+    callback_format: info
+    category: backup_failure
+    mcp:
+      provider: k8s
+      tool: kubectl_get
+      params:
+        namespace: "awoooi-prod"
+        resource: "jobs"
+    reply_format: truncated
+    timeout_sec: 8
+    description: "列出 awoooi-prod 內的備份相關 Job 狀態"
+
+  backup_check_velero:
+    label: "查 Velero"
+    emoji: "🧰"
+    risk: low
+    callback_format: info
+    category: backup_failure
+    mcp:
+      provider: k8s
+      tool: kubectl_get
+      params:
+        namespace: "velero"
+        resource: "backups.velero.io"
+    reply_format: truncated
+    timeout_sec: 8
+    description: "列出 Velero backup CR 狀態"
+
   # ==========================================================================
   # 寫類按鈕（有副作用，4-part nonce callback）
   # ==========================================================================

apps/api/src/services/callback_dispatcher.py

@@ -35,6 +35,18 @@ import yaml
 logger = structlog.get_logger(__name__)
 
 
+_PROVIDER_ALIASES = {
+    "k8s": "kubernetes",
+    "ssh": "ssh_host",
+}
+
+
+def _resolve_provider_name(provider_name: str) -> str:
+    """Normalize legacy callback spec provider names to registered MCP providers."""
+
+    return _PROVIDER_ALIASES.get(provider_name, provider_name)
+
+
 # =============================================================================
 # Data Types
 # =============================================================================
@@ -262,14 +274,15 @@ async def dispatch_action(
 
         # MCP registry dispatch
         from src.plugins.mcp.registry import get_provider
-        provider = get_provider(spec.mcp_provider)
+        provider_name = _resolve_provider_name(spec.mcp_provider)
+        provider = get_provider(provider_name)
         if not provider:
             duration = (time.perf_counter() - start) * 1000
             return DispatchResult(
                 success=False, action=action_name, incident_id=incident_id,
                 user_id=user_id,
-                result_text=f"{spec.emoji} {spec.label} 失敗：MCP provider '{spec.mcp_provider}' 未註冊",
+                result_text=f"{spec.emoji} {spec.label} 失敗：MCP provider '{provider_name}' 未註冊",
-                error=f"provider_not_found: {spec.mcp_provider}",
+                error=f"provider_not_found: {provider_name}",
                 duration_ms=duration,
             )
 

apps/api/src/services/decision_manager.py

@@ -85,6 +85,22 @@ def _should_escalate_auto_approve_rejection(reason: Any) -> bool:
     }
 
 
+_HOST_LAYER_SSH_CATEGORIES = {"infrastructure", "host_resource", "backup_failure"}
+_NON_K8S_HOST_CATEGORIES = {"host_resource", "backup_failure"}
+
+
+def _is_host_layer_ssh_category(category: str | None) -> bool:
+    """Return True when DecisionManager must route non-kubectl actions to SSH."""
+
+    return (category or "") in _HOST_LAYER_SSH_CATEGORIES
+
+
+def _is_non_k8s_host_category(category: str | None) -> bool:
+    """Return True for host/backup alerts that must not auto-run kubectl."""
+
+    return (category or "") in _NON_K8S_HOST_CATEGORIES
+
+
 async def _escalate_decision_auto_repair_unavailable(
     *,
     incident: Incident,
@@ -1990,36 +2006,36 @@ class DecisionManager:
         except Exception as _rescue_err:
             logger.debug("target_rescue_skipped", error=str(_rescue_err))
 
-        # ADR-073 Phase 3-2: infrastructure 告警 (Docker/Host) → SSH MCP routing (2026-04-12 ogt)
+        # ADR-073 Phase 3-2: infrastructure/host/backup 告警 → SSH MCP routing.
-        # alert_category = "infrastructure" 表示 Docker 告警，非 kubectl action → SSH
+        # alert_category = "backup_failure" uses the same host-layer path as AutoRepairService.
         # P1-1 fix 2026-04-12: 必須在 kubectl safety guard 之前 routing，否則 docker 指令被 _action_safe=False 攔截
         _alert_category = getattr(incident, "alert_category", None) or ""
-        if _alert_category in {"infrastructure", "host_resource"} and action and not action.startswith("kubectl"):
+        if _is_host_layer_ssh_category(_alert_category) and action and not action.startswith("kubectl"):
             await self._ssh_execute(incident, token, action, _target)
             return
 
-        # 2026-04-15 ogt: host_resource 告警（HostHighCpuLoad 等）不是 K8s workload 問題
+        # 2026-04-15 ogt: host_resource/backup_failure 告警不是 K8s workload 問題
         # 不得執行 kubectl 操作，改降級人工審核
         # 根因：原本只擋了 infrastructure，忘記 host_resource 也不走 K8s
-        if _alert_category == "host_resource" and action and action.startswith("kubectl"):
+        if _is_non_k8s_host_category(_alert_category) and action and action.startswith("kubectl"):
             logger.warning(
-                "auto_execute_blocked_host_resource_no_k8s",
+                "auto_execute_blocked_host_layer_no_k8s",
                 incident_id=incident.incident_id,
                 alert_category=_alert_category,
                 action=action[:80],
-                reason="host_resource 告警不應執行 K8s kubectl 操作，降級人工審核",
+                reason="host/backup 告警不應執行 K8s kubectl 操作，降級人工審核",
             )
             token.state = DecisionState.READY
             token.proposal_data["auto_executed"] = False
             token.proposal_data["mcp_all_failed"] = True
-            token.proposal_data["blocked_reason"] = "host_resource 告警禁止 K8s kubectl，請人工排查主機"
+            token.proposal_data["blocked_reason"] = f"{_alert_category} 告警禁止 K8s kubectl，請人工排查主機/備份"
             await self._save_token(token)
             _fire_and_forget(
                 _escalate_decision_auto_repair_unavailable(
                     incident=incident,
                     token=token,
                     failure_reason=token.proposal_data["blocked_reason"],
-                    attempted_actions="auto_execute -> host_resource_k8s_block -> emergency_intervention",
+                    attempted_actions=f"auto_execute -> {_alert_category}_k8s_block -> emergency_intervention",
                 )
             )
             _fire_and_forget(_push_decision_to_telegram(incident, token.proposal_data))

apps/api/tests/test_alertmanager_rule_bypass.py

@@ -4,7 +4,11 @@ from src.api.v1.webhooks import (
     _should_bypass_alertmanager_llm,
     _should_use_alertmanager_rule_first,
 )
-from src.services.decision_manager import _should_escalate_auto_approve_rejection
+from src.services.decision_manager import (
+    _is_host_layer_ssh_category,
+    _is_non_k8s_host_category,
+    _should_escalate_auto_approve_rejection,
+)
 from src.services.telegram_gateway import _format_resolved_guard_stamp
 
 
@@ -84,6 +88,18 @@ def test_manual_gate_reasons_escalate_to_emergency_intervention():
     assert _should_escalate_auto_approve_rejection("critical_operation") is False
 
 
+def test_backup_failure_routes_to_decision_ssh_before_kubectl_parser():
+    assert _is_host_layer_ssh_category("backup_failure") is True
+    assert _is_host_layer_ssh_category("host_resource") is True
+    assert _is_host_layer_ssh_category("kubernetes") is False
+
+
+def test_backup_failure_blocks_k8s_auto_execute():
+    assert _is_non_k8s_host_category("backup_failure") is True
+    assert _is_non_k8s_host_category("host_resource") is True
+    assert _is_non_k8s_host_category("infrastructure") is False
+
+
 def test_resolved_guard_stamp_without_timestamp_is_clean():
     assert _format_resolved_guard_stamp(None) == "✅ 此事件已解決"
 

apps/api/tests/test_callback_dispatcher.py

@@ -21,6 +21,7 @@ from src.services.callback_dispatcher import (
     list_actions_for_category,
     load_action_registry,
     _lookup_context,
+    _resolve_provider_name,
     _resolve_template,
 )
 
@@ -68,6 +69,11 @@ class TestRegistryLoading:
             assert spec and spec.callback_format == "info", \
                 f"{qa} should use info format"
 
+    def test_legacy_provider_aliases_resolve_to_registered_names(self):
+        assert _resolve_provider_name("k8s") == "kubernetes"
+        assert _resolve_provider_name("ssh") == "ssh_host"
+        assert _resolve_provider_name("prometheus") == "prometheus"
+
 
 # =============================================================================
 # Category filtering
@@ -91,6 +97,16 @@ class TestCategoryFiltering:
         assert any(a.callback_format == "info" for a in actions), "需至少 1 個查類"
         assert any(a.callback_format == "nonce" for a in actions), "需至少 1 個寫類"
 
+    def test_backup_failure_has_read_only_diagnostics(self):
+        actions = list_actions_for_category("backup_failure")
+        names = {a.name for a in actions}
+        assert {
+            "backup_check_host_disk",
+            "backup_check_jobs",
+            "backup_check_velero",
+        }.issubset(names)
+        assert all(a.callback_format == "info" for a in actions)
+
 
 # =============================================================================
 # Template variable resolution

docs/LOGBOOK.md

@@ -12,19 +12,31 @@ Live e2e 用 `HostBackupFailed` 打 Alertmanager 後發現 aged backup 告警會
 
 ### 完成
 - `_should_use_alertmanager_rule_first()` / `_should_bypass_alertmanager_llm()` 納入 `backup_failure`，備份失敗 YAML `SSH_DIAGNOSE` 不再被 LLM 覆蓋成 K8s 動作。
+- `DecisionManager` SSH route 與 `AutoRepairService` 分類對齊：`backup_failure` 非 kubectl action 先走 SSH MCP，不再落入 `parse_kubectl_action()` 後被 `forbidden_shell_metachar` 擋下。
+- `DecisionManager` host/backup K8s block 納入 `backup_failure`，若 LLM 或 Playbook 產生 kubectl 動作，直接走 emergency escalation，而不是對備份告警誤做 K8s 修復。
 - `AutoRepairService` 追加 host/backup Playbook guard：主機/備份 incident 若匹配到 K8s rollout 類 Playbook，阻擋為 `HOST_BACKUP_K8S_PLAYBOOK`，改走緊急介入。
 - `AutoRepairService` post-verification rollback guard：host/backup 或非 K8s Playbook 驗證失敗時，不再合成 `kubectl rollout restart deployment/{target}`，改走 emergency escalation，且不自動 resolve incident。
 - `EmergencyEscalationService` 沿用既有 `APPROVAL_ESCALATED` DB enum 寫 AOL，避免緊急通道因新 enum 未 migration 而留痕失敗。
 - 補 `phase25_knowledge_enum_names.sql`，讓 `AUTO_RUNBOOK` / `ANTI_PATTERN` enum name 可寫入 PG，修復 auto runbook KM 沉澱失敗。
 - `NodeExporterDown` Prometheus rule `auto_repair` 改為 `true`，與 YAML rule catalog 的 exporter restart 策略一致。
+- `awoooi-executor` RBAC 補 backup/DR 診斷權限：PVC、Jobs/CronJobs、Velero resources read-only，以及 StatefulSet/DaemonSet safe rollout patch。
+- NetworkPolicy 補 K3s master/worker `22/tcp` egress，讓 SSH MCP 可以覆蓋 120/121，不只 110/188。
+- Telegram category buttons 補 provider alias 正規化：`k8s` → `kubernetes`、`ssh` → `ssh_host`，避免按鈕畫出來後 dispatcher 找不到 MCP provider。
+- `backup_failure` 補三個 read-only 診斷按鈕：查主機磁碟、查備份 Job、查 Velero；備份告警不再只有通用批准/拒絕/詳情。
 - 補 `backup_failure` NO_ACTION / SSH_DIAGNOSE 單元測試。
 
 ### 驗證
 - `python3 -m py_compile apps/api/src/api/v1/webhooks.py` 通過。
+- `python3 -m py_compile apps/api/src/services/decision_manager.py apps/api/src/services/callback_dispatcher.py` 通過。
 - `cd apps/api && pytest tests/test_alertmanager_rule_bypass.py tests/test_telegram_ai_automation_block.py tests/test_ai_router_diagnose_fallback.py -q` → 24 passed。
 - `cd apps/api && pytest tests/test_auto_repair_service.py tests/test_alertmanager_rule_bypass.py -q` → 27 passed。
 - `cd apps/api && pytest tests/test_auto_repair_service.py tests/test_alertmanager_rule_bypass.py -q` → 29 passed。
+- `cd apps/api && pytest tests/test_alertmanager_rule_bypass.py tests/test_callback_dispatcher.py tests/test_telegram_button_consistency.py -q` → 56 passed。
 - YAML parse `ops/monitoring/alerts-unified.yml`、`apps/api/alert_rules.yaml` 通過。
+- YAML parse `callback_action_spec.yaml`、`07-rbac.yaml`、`02-network-policy.yaml` 通過。
+- Live Secret/mount 檢查：`ssh-mcp-key`、`awoooi-repair-ssh-key`、`awoooi-repair-known-hosts` 存在且掛載可讀。
+- Live SSH MCP key 檢查：`wooo@192.168.0.110`、`ollama@192.168.0.188` OK；`wooo@192.168.0.120/121` 已通過 host key，但 remote `authorized_keys` 尚未納入該公鑰，回 `Permission denied (publickey,password)`。
+- Live RBAC apply 被 Argo 依 Git 狀態拉回；`07-rbac.yaml` 需推上 Gitea 由 Argo 同步後再驗 `can-i`。
 
 ## 2026-04-30 | ADR-104 Playbook 版本化 lineage
 

docs/adr/ADR-058-host-auto-repair-ssh-whitelist.md

@@ -140,6 +140,48 @@ MoWoooWorkDown → Jaccard 匹配 momo-app-down-repair → SSH ollama@192.168.0.
 
 ---
 
+## Appendix B — Backup Failure Route Parity (2026-05-01)
+
+`HostBackupFailed` / backup 類告警的 `alert_category` 是 `backup_failure`。它必須在所有 host-layer 自動化路徑與 `host_resource` 同級處理：
+
+| Layer | 必須行為 |
+|-------|----------|
+| Alertmanager rule-first | YAML `SSH_DIAGNOSE` / `NO_ACTION` 不進 LLM 覆寫 |
+| AutoRepairService | `backup_failure` 視為 host-layer，拒絕 K8s Playbook fallback |
+| DecisionManager | 非 `kubectl` 動作在 kubectl parser 前路由 SSH MCP |
+| DecisionManager K8s guard | `backup_failure` 產生 `kubectl` 時降級 emergency escalation |
+| Telegram buttons | `backup_failure` 顯示只讀診斷按鈕：主機磁碟、備份 Job、Velero 狀態 |
+
+2026-05-01 根因：DecisionManager SSH route 只含 `infrastructure` / `host_resource`，漏掉 `backup_failure`，導致 `ssh 192.168.0.110 '...;...'` 類只讀診斷動作掉進 `parse_kubectl_action()`，被 `forbidden_shell_metachar` 擋下。
+
+同日按鈕審計也發現 category button 的 friendly provider name 會漂移：`callback_action_spec.yaml` 使用 `k8s` / `ssh`，但 MCP registry 實際名稱是 `kubernetes` / `ssh_host`。Dispatcher 必須正規化 provider alias，否則卡片按鈕會顯示但執行時變成 `provider_not_found`。
+
+### Runtime 權限基準
+
+- K8s Secret:
+  - `awoooi-repair-ssh-key` 掛載 `/etc/repair-ssh/`
+  - `awoooi-repair-known-hosts` 掛載 `/etc/repair-known-hosts/`
+  - `ssh-mcp-key` 掛載 `/run/secrets/ssh_mcp_key` 與 `/etc/ssh-mcp/known_hosts`
+- Remote `authorized_keys`:
+  - `wooo@192.168.0.110`
+  - `wooo@192.168.0.120`
+  - `wooo@192.168.0.121`
+  - `ollama@192.168.0.188`
+- NetworkPolicy egress:
+  - `192.168.0.110:22`
+  - `192.168.0.120:22`
+  - `192.168.0.121:22`
+  - `192.168.0.188:22`
+- `awoooi-executor` RBAC:
+  - read `jobs.batch`, `cronjobs.batch`
+  - read `persistentvolumeclaims`
+  - read Velero `backups`, `backupstoragelocations`, `backuprepositories`, `podvolumebackups`, `podvolumerestores`, `restores`, `schedules`
+  - patch `statefulsets.apps` / `daemonsets.apps` only for safe rollout restart
+
+If SSH MCP fails, the incident must not silently become a manual approval card; it must raise the emergency intervention path with the exact SSH failure reason when available.
+
+---
+
 ## 首席架構師 Review 記錄 (2026-04-05)
 
 評分：**72/100 → 修正後 88/100**

k8s/awoooi-prod/02-network-policy.yaml

@@ -1,8 +1,9 @@
 # AWOOOI 正式環境零信任網路策略
 # 負責人: CIO
-# 版本: v1.5
+# 版本: v1.6
-# 日期: 2026-04-14
+# 日期: 2026-05-01
 # 變更:
+#   - v1.6: 新增 K3s node 120/121 SSH egress，供 SSH MCP 主機診斷/修復
 #   - v1.5: 新增 keepalived VIP 192.168.0.125/32 ArgoCD NodePort 30443 egress（修復 heartbeat probe）
 #   - v1.4: 新增 ArgoCD MCP egress（argocd namespace port 80/443）
 #   - v1.3: 新增 192.168.0.111 Ollama 主機 (M1 Pro)，移除 188 的 Ollama port
@@ -168,7 +169,7 @@ spec:
         - protocol: TCP
           port: 8080
 
-    # 允許訪問 K8s API (Executor 執行 kubectl 指令)
+    # 允許訪問 K8s API + K3s master SSH (Executor 執行 kubectl/host diagnosis)
     # 2026-03-23 修復: Y 按鈕執行超時
     # 重要: ClusterIP (10.43.0.1:443) 會路由到實際端點 (192.168.0.120:6443)
     #       必須同時允許兩者，否則流量會被 192.168.0.0/16 排除規則阻擋
@@ -180,8 +181,11 @@ spec:
           port: 443
     - to:
         - ipBlock:
-            cidr: 192.168.0.120/32  # K3s Master 實際 API Server 端點 + ArgoCD NodePort
+            cidr: 192.168.0.120/32  # K3s Master 實際 API Server 端點 + ArgoCD NodePort + SSH MCP
       ports:
+        # SSH MCP — K3s master host diagnosis/repair path
+        - protocol: TCP
+          port: 22
         - protocol: TCP
           port: 6443
         # ArgoCD MCP NodePort (2026-04-11): ClusterIP DNAT 跨 namespace 不穩定，改用 NodePort
@@ -221,6 +225,9 @@ spec:
         - ipBlock:
             cidr: 192.168.0.121/32
       ports:
+        # SSH MCP — K3s worker host diagnosis/repair path
+        - protocol: TCP
+          port: 22
         - protocol: TCP
           port: 6443
         - protocol: TCP

k8s/awoooi-prod/07-rbac.yaml

@@ -62,6 +62,11 @@ rules:
     resources: ["services", "configmaps"]
     verbs: ["get", "list", "watch"]
 
+  # 2026-05-01: backup/disk diagnostics need PVC visibility; read-only only.
+  - apiGroups: [""]
+    resources: ["persistentvolumeclaims"]
+    verbs: ["get", "list"]
+
   - apiGroups: ["networking.k8s.io"]
     resources: ["ingresses"]
     verbs: ["get", "list", "watch"]
@@ -89,6 +94,23 @@ rules:
     resources: ["statefulsets", "daemonsets"]
     verbs: ["get", "list", "watch"]
 
+  # 2026-05-01: HostBackupFailed / VeleroBackupFailed diagnosis needs backup job status.
+  - apiGroups: ["batch"]
+    resources: ["jobs", "cronjobs"]
+    verbs: ["get", "list", "watch"]
+
+  # 2026-05-01: Velero backup status is read-only evidence for backup_failure alerts.
+  - apiGroups: ["velero.io"]
+    resources:
+      - backups
+      - backupstoragelocations
+      - backuprepositories
+      - podvolumebackups
+      - podvolumerestores
+      - restores
+      - schedules
+    verbs: ["get", "list", "watch"]
+
   # ============================================================================
   # 寫入權限 (Write) - 僅限故障排除操作
   # ============================================================================
@@ -104,6 +126,11 @@ rules:
     resources: ["deployments"]
     verbs: ["patch"]
 
+  # 2026-05-01: allow the same safe rollout restart primitive on controller types.
+  - apiGroups: ["apps"]
+    resources: ["statefulsets", "daemonsets"]
+    verbs: ["patch"]
+
   # Scale Deployments (擴縮容)
   - apiGroups: ["apps"]
     resources: ["deployments/scale"]