feat(adr-075): 修復 Telegram 動態按鈕 4 個斷點 + 新增 7 種告警分類

斷點 A: decision_manager 提取 alert_category/notification_type 傳入 send_approval_card
斷點 B: send_approval_card 新增參數並傳遞至 _build_inline_keyboard
斷點 C: 互動型通知 (TYPE-3/4/4D/8M) 禁止發 SRE 群組，防 nonce 洩漏
斷點 D: _CATEGORY_BUTTONS k8s_workload → kubernetes + 新增 6 類按鈕組

classify_alert_early 新增: alertchain_health, flywheel_health, storage,
devops_tool, external_site, ssl_cert, host_resource (從 infrastructure 分離)
Test: 52 classify + 664 total passed

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

apps/api/src/services/decision_manager.py

@@ -246,6 +246,9 @@ async def _push_decision_to_telegram(
             )
         else:
             # TYPE-2 / TYPE-3 / TYPE-4 都走 send_approval_card（按鈕組合由 alert_category 決定）
+            # 2026-04-12 ogt: ADR-075 斷點 A 修復 — 從 Incident 提取 alert_category/notification_type
+            _alert_category = getattr(incident, "alert_category", "") or ""
+            _notification_type = getattr(incident, "notification_type", "") or _notif_type.value if _notif_type else ""
             tg_result = await gateway.send_approval_card(
                 approval_id=approval_id,
                 risk_level=risk_level,
@@ -265,6 +268,8 @@ async def _push_decision_to_telegram(
                 nemotron_tool_model=nemotron_tool_model,
                 nemotron_tool_backend=nemotron_tool_backend,
                 incident_id=incident.incident_id,
+                alert_category=_alert_category,
+                notification_type=_notification_type,
             )
 
         # 2026-04-09 Claude Sonnet 4.6: 存 message_id → 後續狀態更新在原訊息延續

apps/api/src/services/incident_service.py

@@ -121,20 +121,73 @@ def classify_alert_early(alertname: str, severity: str, labels: dict | None = No
     Returns:
         tuple[str, str]: (alert_category, notification_type)
     """
+    # ADR-075 (2026-04-12): 完整重寫，修正 category 命名對齊 _build_inline_keyboard。
+    # 統帥決議：kubernetes（非 k8s_workload）、host_resource 從 infrastructure 分離。
     alertname_lower = alertname.lower()
+
+    # 1. Config Drift（最高優先，類型特殊不受 severity 影響）
     if alertname in ("ConfigurationDrift", "KubeConfigDrift"):
         return "config_drift", "TYPE-4D"
+
+    # 2. 告警鏈路健康（meta-monitoring，優先於 severity 判斷）
+    if alertname in (
+        "AlertChainBroken_Alertmanager",
+        "AlertChainBroken_Sentry",
+        "NoAlertsReceived2Hours",
+        "AlertChainUnhealthy",
+    ):
+        return "alertchain_health", "TYPE-8M"
+
+    # 3. 飛輪/AI 系統健康（優先於 severity 判斷）
+    if alertname in ("AutoRepairLowSuccessRate", "PermanentFixRequired") or alertname.startswith("Flywheel"):
+        return "flywheel_health", "TYPE-8M"
+
+    # 4. 純資訊
     if severity in ("info", "none"):
         return "info", "TYPE-1"
-    # Watchdog/Heartbeat 永遠是 TYPE-1（Alertmanager 心跳）
-    if "watchdog" in alertname_lower or alertname in ("Heartbeat",):
+
+    # 5. Backup / Heartbeat（Velero 已由 K8s prefix 規則接管）
+    if "watchdog" in alertname_lower or alertname == "Heartbeat":
         return "backup", "TYPE-1"
-    if alertname.startswith(("Docker", "Host")):
+
+    # 6. 主機資源（從 infrastructure 分離，ADR-075 統帥決議）
+    if alertname.startswith("Host"):
+        return "host_resource", "TYPE-3"
+
+    # 7. Docker 容器
+    if alertname.startswith("Docker"):
         return "infrastructure", "TYPE-3"
+
+    # 8. K8s（統帥決議：統一用 kubernetes，不用 k8s_workload）
     if alertname.startswith(("Kube", "Pod", "Deploy", "Node", "Velero", "ArgoCD")):
         return "kubernetes", "TYPE-3"
+
+    # 9. 資料庫
     if alertname.startswith(("Postgres", "Redis")):
         return "database", "TYPE-3"
+
+    # 10. 物件儲存
+    if alertname == "MinIODown":
+        return "storage", "TYPE-3"
+
+    # 11. DevOps 工具（ADR-075 修正：從 general 分離）
+    if alertname in (
+        "OpenClawDown", "SignOzDown", "GiteaDown", "HarborDown",
+        "SentryDown", "AlertmanagerDown", "KaliScannerDown", "GiteaCIPipelineFailed",
+    ):
+        return "devops_tool", "TYPE-3"
+
+    # 12. 外部網站（ADR-075 修正：從 general 分離）
+    if alertname in (
+        "MoWoooWorkDown", "TsenyangWebsiteDown",
+        "StockWoooWorkDown", "BitanWoooWorkDown",
+    ):
+        return "external_site", "TYPE-3"
+
+    # 13. SSL 憑證（ADR-075 修正：從 general 分離）
+    if alertname.startswith(("ExternalSiteSSL", "TLSCert")):
+        return "ssl_cert", "TYPE-3"
+
     return "general", "TYPE-3"
 
 

apps/api/src/services/telegram_gateway.py

@@ -1355,8 +1355,9 @@ class TelegramGateway:
             notification_type: 通知類型 (TYPE-1/2/3/4/4D)
         """
         # TYPE-3 動態操作按鈕 (ADR-071-E)
+        # ADR-075: 統一用 kubernetes（移除舊 k8s_workload），新增 storage/external_site/alertchain_health/flywheel_health
         _CATEGORY_BUTTONS: dict[str, list[tuple[str, str]]] = {
-            "k8s_workload":  [
+            "kubernetes": [
                 ("🔄 重啟", f"action:restart:{incident_id}"),
                 ("📈 擴容", f"action:scale_up:{incident_id}"),
                 ("📉 縮容", f"action:scale_down:{incident_id}"),
@@ -1379,6 +1380,22 @@ class TelegramGateway:
                 ("🔄 重啟服務", f"action:restart_service:{incident_id}"),
                 ("📋 查 Log", f"action:check_log:{incident_id}"),
             ],
+            "storage": [
+                ("🔄 重啟 MinIO", f"action:restart_service:{incident_id}"),
+                ("📋 查 Log", f"action:check_log:{incident_id}"),
+            ],
+            "external_site": [
+                ("🔍 查健康狀態", f"action:check_health:{incident_id}"),
+                ("📋 查 Log", f"action:check_log:{incident_id}"),
+            ],
+            "alertchain_health": [
+                ("🔍 查 Alertmanager", f"action:check_log:{incident_id}"),
+                ("🔄 重載設定", f"action:reload_config:{incident_id}"),
+            ],
+            "flywheel_health": [
+                ("🔍 查飛輪狀態", f"action:check_health:{incident_id}"),
+                ("📋 查 Log", f"action:check_log:{incident_id}"),
+            ],
             "ai_system": [
                 ("🔀 切換 Provider", f"action:switch_provider:{incident_id}"),
             ],
@@ -1473,6 +1490,9 @@ class TelegramGateway:
         nemotron_tool_backend: str = "",
         # 2026-04-05 Claude Code: incident_id 用於 detail/reanalyze/history 按鈕
         incident_id: str = "",
+        # 2026-04-12 ogt: ADR-075 告警分類與通知類型（斷點 B 修復）
+        alert_category: str = "",
+        notification_type: str = "",
     ) -> dict:
         """
         推送待簽核卡片到 Telegram (v7.0 含 SignOz 整合)
@@ -1553,11 +1573,14 @@ class TelegramGateway:
 
         # 建立按鈕 (含自動調優)
         # 2026-04-05 Claude Code: 傳入 incident_id 以啟用 detail/reanalyze/history 按鈕
+        # ADR-075: 傳入 alert_category/notification_type 以啟用分類動態按鈕（斷點 B 修復）
         keyboard = self._build_inline_keyboard(
             approval_id=approval_id,
             include_auto_tuning=bool(auto_tuning_command),
             auto_tuning_command=auto_tuning_command,
             incident_id=incident_id,
+            alert_category=alert_category,
+            notification_type=notification_type,
         )
 
         # 發送訊息
@@ -1597,8 +1620,14 @@ class TelegramGateway:
 
         # 2026-04-03 ogt: 發到 SRE 群組並觸發 AI 雙向討論 (Triumvirate ADR-053)
         # 2026-04-05 ogt: 升級為完整 v7.0 格式，與個人 chat 一致
-        # 非同步執行，失敗不影響告警主流程
-        if settings.SRE_GROUP_CHAT_ID:
+        # 2026-04-12 ogt: ADR-075 斷點 C 修復 — 含按鈕的互動型通知禁止發群組（nonce 洩漏）
+        #   TYPE-1/TYPE-2 → 可發群組（純資訊，無 callback 按鈕）
+        #   TYPE-3/TYPE-4/TYPE-4D/TYPE-8M → 僅 DM，不發群組
+        _interactive_types = {"TYPE-3", "TYPE-4", "TYPE-4D", "TYPE-8M"}
+        _is_interactive = notification_type in _interactive_types or (
+            not notification_type and alert_category  # 有分類但無明確型別 → 視為互動型
+        )
+        if settings.SRE_GROUP_CHAT_ID and not _is_interactive:
             asyncio.create_task(
                 self._send_approval_card_to_group(
                     approval_id=approval_id,

apps/api/tests/test_classify_alert_early.py

@@ -76,17 +76,144 @@ class TestInfrastructure:
     def test_docker_prefix(self):
         ac, nt = classify_alert_early("DockerContainerOOM", "critical", {})
         assert nt == "TYPE-3"
-        assert ac == "infrastructure"
-
-    def test_host_prefix(self):
-        ac, nt = classify_alert_early("HostHighCpuLoad", "warning", {})
-        assert nt == "TYPE-3"
-        assert ac == "infrastructure"
+        assert ac == "infrastructure"  # Docker → 保留 infrastructure
 
     def test_docker_restart(self):
         ac, nt = classify_alert_early("DockerContainerRestarting", "warning", {})
         assert ac == "infrastructure"
 
+    # ADR-075: Host* 從 infrastructure 分離為 host_resource
+    def test_host_prefix_is_host_resource(self):
+        ac, nt = classify_alert_early("HostHighCpuLoad", "warning", {})
+        assert nt == "TYPE-3"
+        assert ac == "host_resource"
+
+    def test_host_down(self):
+        ac, nt = classify_alert_early("HostDown", "critical", {})
+        assert ac == "host_resource"
+
+    def test_host_memory(self):
+        ac, nt = classify_alert_early("HostOutOfMemory", "warning", {})
+        assert ac == "host_resource"
+
+    def test_host_disk(self):
+        ac, nt = classify_alert_early("HostOutOfDiskSpace", "warning", {})
+        assert ac == "host_resource"
+
+
+# --------------------------------------------------------------------------- #
+# ADR-075: alertchain_health (TYPE-8M)
+# --------------------------------------------------------------------------- #
+
+class TestAlertchainHealth:
+    @pytest.mark.parametrize("alertname", [
+        "AlertChainBroken_Alertmanager",
+        "AlertChainBroken_Sentry",
+        "NoAlertsReceived2Hours",
+        "AlertChainUnhealthy",
+    ])
+    def test_alertchain_alerts(self, alertname):
+        ac, nt = classify_alert_early(alertname, "critical", {})
+        assert ac == "alertchain_health"
+        assert nt == "TYPE-8M"
+
+    def test_alertchain_beats_severity_info(self):
+        # 即使 severity=info，AlertChainBroken 也必須是 alertchain_health
+        ac, nt = classify_alert_early("AlertChainBroken_Alertmanager", "info", {})
+        assert ac == "alertchain_health"
+        assert nt == "TYPE-8M"
+
+
+# --------------------------------------------------------------------------- #
+# ADR-075: flywheel_health (TYPE-8M)
+# --------------------------------------------------------------------------- #
+
+class TestFlywheelHealth:
+    def test_auto_repair_low_success(self):
+        ac, nt = classify_alert_early("AutoRepairLowSuccessRate", "warning", {})
+        assert ac == "flywheel_health"
+        assert nt == "TYPE-8M"
+
+    def test_permanent_fix_required(self):
+        ac, nt = classify_alert_early("PermanentFixRequired", "warning", {})
+        assert ac == "flywheel_health"
+        assert nt == "TYPE-8M"
+
+    def test_flywheel_prefix(self):
+        ac, nt = classify_alert_early("FlywheelPlaybookZero", "critical", {})
+        assert ac == "flywheel_health"
+        assert nt == "TYPE-8M"
+
+    def test_flywheel_beats_severity_info(self):
+        ac, nt = classify_alert_early("AutoRepairLowSuccessRate", "info", {})
+        assert ac == "flywheel_health"
+        assert nt == "TYPE-8M"
+
+
+# --------------------------------------------------------------------------- #
+# ADR-075: storage (TYPE-3)
+# --------------------------------------------------------------------------- #
+
+class TestStorage:
+    def test_minio_down(self):
+        ac, nt = classify_alert_early("MinIODown", "critical", {})
+        assert ac == "storage"
+        assert nt == "TYPE-3"
+
+
+# --------------------------------------------------------------------------- #
+# ADR-075: devops_tool (TYPE-3)
+# --------------------------------------------------------------------------- #
+
+class TestDevopsTool:
+    @pytest.mark.parametrize("alertname", [
+        "OpenClawDown",
+        "SignOzDown",
+        "GiteaDown",
+        "HarborDown",
+        "SentryDown",
+        "AlertmanagerDown",
+        "KaliScannerDown",
+        "GiteaCIPipelineFailed",
+    ])
+    def test_devops_tools(self, alertname):
+        ac, nt = classify_alert_early(alertname, "critical", {})
+        assert ac == "devops_tool"
+        assert nt == "TYPE-3"
+
+
+# --------------------------------------------------------------------------- #
+# ADR-075: external_site (TYPE-3)
+# --------------------------------------------------------------------------- #
+
+class TestExternalSite:
+    @pytest.mark.parametrize("alertname", [
+        "MoWoooWorkDown",
+        "TsenyangWebsiteDown",
+        "StockWoooWorkDown",
+        "BitanWoooWorkDown",
+    ])
+    def test_external_sites(self, alertname):
+        ac, nt = classify_alert_early(alertname, "critical", {})
+        assert ac == "external_site"
+        assert nt == "TYPE-3"
+
+
+# --------------------------------------------------------------------------- #
+# ADR-075: ssl_cert (TYPE-3)
+# --------------------------------------------------------------------------- #
+
+class TestSslCert:
+    def test_external_site_ssl(self):
+        ac, nt = classify_alert_early("ExternalSiteSSLExpiringSoon", "warning", {})
+        assert ac == "ssl_cert"
+        assert nt == "TYPE-3"
+
+    def test_tls_cert(self):
+        ac, nt = classify_alert_early("TLSCertExpiryCritical", "critical", {})
+        assert ac == "ssl_cert"
+        assert nt == "TYPE-3"
+
 
 # --------------------------------------------------------------------------- #
 # TYPE-3: Kubernetes