fix(cr): Code Review P0/P1 全修補 — 積木化+SSH路由+安全守衛順序

P0-1: classify_alert_early 移至 incident_service (Service層)，webhooks.py import 修正
P0-2: _ssh_execute() 改用 self._ssh，移除冗餘 SSHProvider() 實例化
P1-1: infrastructure SSH routing 移至 kubectl safety guard 之前，docker指令不再被攔截
P1-2: alert_rule_engine 新增 get_risk_for_alertname() public API
P1-3: classify_notification() docstring 修正 ORM→Pydantic

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

apps/api/src/api/v1/webhooks.py

@@ -55,6 +55,7 @@ from src.services.approval_db import get_approval_service
 # Phase 17 P0: Service 層 (消除 Router 直接存取 Redis)
 # C2 修正 (首席架構師審查 2026-04-10): create_incident_for_approval + extract_affected_services 已移入 Service 層
 from src.services.incident_service import (
+    classify_alert_early,
     create_incident_for_approval,
     extract_affected_services,
     get_incident_service,
@@ -970,30 +971,6 @@ class AlertmanagerPayload(BaseModel):
     alerts: list[AlertmanagerAlert]
 
 
-def classify_alert_early(alertname: str, severity: str, _labels: dict) -> tuple[str, str]:
-    """
-    ADR-073 Phase 2-2: 早期分診，在 LLM 分析前決定 alert_category + notification_type。
-    2026-04-12 ogt: 防止 HostBackupFailed 等被誤路由到 K8s executor。
-
-    Returns:
-        tuple[str, str]: (alert_category, notification_type)
-    """
-    alertname_lower = alertname.lower()
-    if alertname in ("ConfigurationDrift", "KubeConfigDrift"):
-        return "config_drift", "TYPE-4D"
-    if severity in ("info", "none"):
-        return "info", "TYPE-1"
-    if any(kw in alertname_lower for kw in ("backup", "heartbeat")):
-        return "backup", "TYPE-1"
-    if alertname.startswith(("Docker", "Host")):
-        return "infrastructure", "TYPE-3"
-    if alertname.startswith(("Kube", "Pod", "Deploy", "Node", "Velero", "ArgoCD")):
-        return "kubernetes", "TYPE-3"
-    if alertname.startswith(("Postgres", "Redis")):
-        return "database", "TYPE-3"
-    return "general", "TYPE-3"
-
-
 def is_internal_ip(client_ip: str) -> bool:
     """檢查是否為內網 IP"""
     import ipaddress

apps/api/src/services/alert_rule_engine.py

@@ -145,6 +145,20 @@ def _is_generic(rule: dict) -> bool:
 # ── 公開 API ────────────────────────────────────────────────
 
 
+def get_risk_for_alertname(alertname: str) -> str | None:
+    """
+    根據 alertname 從 alert_rules.yaml 取得 risk 等級。
+    ADR-073 Phase 3-6: 供 decision_manager 做 YAML risk override。
+
+    Returns:
+        risk 字串 ("low"/"medium"/"critical") 或 None（無匹配規則）
+    """
+    for rule in _load_rules():
+        if _matches(rule, alertname, "", "", ""):
+            return rule.get("response", {}).get("risk")
+    return None
+
+
 def match_rule(alert_context: dict) -> dict[str, Any] | None:
     """
     根據 alert_context 匹配規則，回傳填充後的 response dict。

apps/api/src/services/decision_manager.py

@@ -1246,6 +1246,15 @@ class DecisionManager:
         action = _re.sub(r"<deployment_name>", _target, action)
         action = _re.sub(r"<[^>]+>", _target, action)
 
+        # ADR-073 Phase 3-2: infrastructure 告警 (Docker/Host) → SSH MCP routing (2026-04-12 ogt)
+        # alert_category = "infrastructure" 表示 Docker/Host 告警，不走 K8s executor
+        # action 格式應為 "docker restart <container>" 或 "systemctl restart <service>"
+        # 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 == "infrastructure" and action and not action.startswith("kubectl"):
+            await self._ssh_execute(incident, token, action, _target)
+            return
+
         # 安全守衛: 替換後仍含 "unknown" 或未替換的 <...>/{...} → 拒絕執行
         # 另外：若 target 等於 alertname，代表 LLM 把告警名稱填入 deployment_name，也拒絕
         _alertname = incident.signals[0].labels.get("alertname", "") if incident.signals else ""
@@ -1274,14 +1283,6 @@ class DecisionManager:
             )
             return
 
-        # ADR-073 Phase 3-2: infrastructure 告警 (Docker/Host) → SSH MCP routing (2026-04-12 ogt)
-        # alert_category = "infrastructure" 表示 Docker/Host 告警，不走 K8s executor
-        # action 格式應為 "docker restart <container>" 或 "systemctl restart <service>"
-        _alert_category = getattr(incident, "alert_category", None) or ""
-        if _alert_category == "infrastructure" and action and not action.startswith("kubectl"):
-            await self._ssh_execute(incident, token, action, _target)
-            return
-
         # BUG-003 修復 2026-04-11: 加入 K8s deployment 存在性驗證，
         # 避免 LLM 產生的無效 deployment name（<placeholder>/alertname/unknown）通過 safety guard
         # 但仍對 K8s 發出錯誤指令
@@ -1596,24 +1597,21 @@ class DecisionManager:
                 # ADR-073 Phase 3-6: YAML rule risk_level 優先於 LLM 輸出 (2026-04-12 ogt)
                 # LLM 有時把 critical 告警估為 medium，YAML 規則是由人工審閱過的，優先採用
                 try:
-                    from src.services.alert_rule_engine import _load_rules, _matches
+                    from src.services.alert_rule_engine import get_risk_for_alertname
                     _alertname_for_risk = (
                         incident.signals[0].labels.get("alertname", "")
                         if incident.signals else ""
                     )
                     if _alertname_for_risk:
-                        for _rule in _load_rules():
-                            if _matches(_rule, _alertname_for_risk, "", "", ""):
-                                _yaml_risk = _rule.get("response", {}).get("risk")
-                                if _yaml_risk and _yaml_risk != result.get("risk_level"):
-                                    logger.info(
-                                        "risk_level_yaml_override",
-                                        incident_id=incident.incident_id,
-                                        llm_risk=result.get("risk_level"),
-                                        yaml_risk=_yaml_risk,
-                                    )
-                                    result["risk_level"] = _yaml_risk
-                                break
+                        _yaml_risk = get_risk_for_alertname(_alertname_for_risk)
+                        if _yaml_risk and _yaml_risk != result.get("risk_level"):
+                            logger.info(
+                                "risk_level_yaml_override",
+                                incident_id=incident.incident_id,
+                                llm_risk=result.get("risk_level"),
+                                yaml_risk=_yaml_risk,
+                            )
+                            result["risk_level"] = _yaml_risk
                 except Exception as _re:
                     logger.debug("risk_level_yaml_override_failed", error=str(_re))
 
@@ -2030,7 +2028,6 @@ class DecisionManager:
         - systemctl restart <service>
         - docker rm -f <container> (含 docker start)
         """
-        from src.plugins.mcp.providers.ssh_provider import SSHProvider
         import os as _os
 
         # 取得主機 — 從 instance label 或 SSH_MCP_ALLOWED_HOSTS 第一台
@@ -2076,7 +2073,6 @@ class DecisionManager:
             _fire_and_forget(_push_decision_to_telegram(incident, token.proposal_data))
             return
 
-        ssh = SSHProvider()
         params: dict = {"host": _host}
         if _tool == "docker_restart":
             params["container"] = _container
@@ -2084,7 +2080,7 @@ class DecisionManager:
             params["service"] = _service
 
         try:
-            result = await ssh.execute(tool_name=_tool, parameters=params)
+            result = await self._ssh.execute(tool_name=_tool, parameters=params)
             success = result.success
 
             logger.info(

apps/api/src/services/incident_service.py

@@ -101,6 +101,42 @@ def extract_affected_services(labels: dict, target_resource: str) -> list[str]:
     return []
 
 
+def classify_alert_early(alertname: str, severity: str, _labels: dict) -> tuple[str, str]:
+    """
+    ADR-073 Phase 2-2: 早期分診，在 LLM 分析前決定 alert_category + notification_type。
+    防止 HostBackupFailed 等告警被誤路由到 K8s executor。
+
+    規則優先順序（由高到低）:
+    1. ConfigurationDrift / KubeConfigDrift → TYPE-4D (Config Drift 卡片)
+    2. severity=info/none → TYPE-1 (純資訊，無按鈕)
+    3. backup/heartbeat 關鍵字 → TYPE-1
+    4. Docker/Host 前綴 → infrastructure TYPE-3
+    5. Kube/Pod/Deploy/Node/Velero/ArgoCD 前綴 → kubernetes TYPE-3
+    6. Postgres/Redis 前綴 → database TYPE-3
+    7. 預設 → general TYPE-3
+
+    C3 修正 (首席架構師 CR 2026-04-13): 從 Router 層 (webhooks.py) 移入 Service 層
+    原違規: 業務邏輯函數定義在 api/v1/webhooks.py
+
+    Returns:
+        tuple[str, str]: (alert_category, notification_type)
+    """
+    alertname_lower = alertname.lower()
+    if alertname in ("ConfigurationDrift", "KubeConfigDrift"):
+        return "config_drift", "TYPE-4D"
+    if severity in ("info", "none"):
+        return "info", "TYPE-1"
+    if any(kw in alertname_lower for kw in ("backup", "heartbeat")):
+        return "backup", "TYPE-1"
+    if alertname.startswith(("Docker", "Host")):
+        return "infrastructure", "TYPE-3"
+    if alertname.startswith(("Kube", "Pod", "Deploy", "Node", "Velero", "ArgoCD")):
+        return "kubernetes", "TYPE-3"
+    if alertname.startswith(("Postgres", "Redis")):
+        return "database", "TYPE-3"
+    return "general", "TYPE-3"
+
+
 async def create_incident_for_approval(
     approval_id: str,
     risk_level: str,

apps/api/src/services/telegram_gateway.py

@@ -1127,7 +1127,7 @@ def classify_notification(
       TYPE-4D > TYPE-1 > TYPE-2 > TYPE-4 > TYPE-3(預設)
 
     Args:
-        incident: Incident ORM 物件（需要 signals[].labels + title）
+        incident: Incident Pydantic 模型（需要 signals[].labels + title）
         confidence: AI 決策信心值 (0.0~1.0)
         auto_executed: 是否已自動修復執行完成
         mcp_all_failed: 所有 MCP provider 是否全失敗