From 543d239d8f013a6c7edf23361b216a5a219a56dc Mon Sep 17 00:00:00 2001
From: ogt <ogt@awoooi.local>
Date: Fri, 17 Jul 2026 09:26:21 +0800
Subject: [PATCH] fix(sre): route callback ingress to Host188 executor

---
 apps/api/alert_rules.yaml                     | 24 ++++++++++++++
 apps/api/src/api/v1/webhooks.py               |  6 ++++
 apps/api/src/services/incident_service.py     |  9 ++++++
 .../tests/test_alertmanager_rule_bypass.py    | 31 +++++++++++++++++++
 apps/api/tests/test_classify_alert_early.py   | 11 +++++++
 5 files changed, 81 insertions(+)

diff --git a/apps/api/alert_rules.yaml b/apps/api/alert_rules.yaml
index 8515d74c1..e66d6086f 100644
--- a/apps/api/alert_rules.yaml
+++ b/apps/api/alert_rules.yaml
@@ -614,6 +614,30 @@ rules:
 
   # ── 告警鏈路監控 ────────────────────────────────────────────
 
+  - id: telegram_callback_ingress_host188
+    priority: 108
+    description: Telegram callback ingress requires the exact Host188 OpenClaw overlay
+    match:
+      alertname:
+        - TelegramCallbackIngressUnverified
+        - TelegramCallbackForwarderDrift
+    response:
+      action_title: "受控收斂 Host188 Telegram callback ingress"
+      description: "權威路由: 僅允許 service:openclaw:host188 的 callback-forwarder PlayBook，先 check-mode，再 bounded apply 與獨立 verifier。"
+      suggested_action: CONTROLLED_APPLY
+      kubectl_command: ""
+      estimated_downtime: "0s; additive Compose overlay with rollback"
+      risk: medium
+      execution_risk_independent_of_incident_severity: true
+      responsibility: INFRA
+      responsibility_reasoning: "Telegram callback ingress 唯一 owner 為 Host188 OpenClaw；禁止 generic Kubernetes、跨主機或跨 domain fallback"
+      secondary_teams: [BE]
+      optimization:
+        - type: INDEPENDENT_VERIFIER
+          description: "驗證 Host188 overlay identity、callback receiver health、fresh durable ingress receipt 與 interactive button readiness"
+          command: ""
+      reasoning: "[權威規則] 固定路由 service:openclaw:host188 -> ansible:188-openclaw-callback-forwarder；不允許 LLM 改寫 domain。"
+
   - id: alert_chain_broken_alertmanager
     priority: 109
     description: Alertmanager webhook delivery chain degraded
diff --git a/apps/api/src/api/v1/webhooks.py b/apps/api/src/api/v1/webhooks.py
index f164b095e..c0f4de027 100644
--- a/apps/api/src/api/v1/webhooks.py
+++ b/apps/api/src/api/v1/webhooks.py
@@ -278,6 +278,12 @@ def _should_use_alertmanager_rule_first(
             and suggested_action == "INVESTIGATE"
             and not command
         )
+    if alert_category == "ai_agent":
+        return (
+            rule_id == "telegram_callback_ingress_host188"
+            and suggested_action == "CONTROLLED_APPLY"
+            and not command
+        )
     if alert_category not in {"host_resource", "backup_failure"}:
         return False
     return (
diff --git a/apps/api/src/services/incident_service.py b/apps/api/src/services/incident_service.py
index c9845dbf1..0b9ea02a4 100644
--- a/apps/api/src/services/incident_service.py
+++ b/apps/api/src/services/incident_service.py
@@ -190,6 +190,15 @@ def classify_alert_early(
     ):
         return "alertchain_health", "TYPE-8M"
 
+    # Host188 OpenClaw owns the Telegram callback ingress. Keep this exact
+    # repair signal out of the general/Kubernetes fallback path so the typed
+    # router can select only the bounded Host188 Ansible catalog.
+    if alertname in (
+        "TelegramCallbackIngressUnverified",
+        "TelegramCallbackForwarderDrift",
+    ):
+        return "ai_agent", "TYPE-3"
+
     # 3. 資安告警（高優先，防止被 severity/prefix 規則覆蓋）
     # ADR-075 TYPE-5S (2026-04-12 ogt)
     if any(alertname.startswith(p) for p in (
diff --git a/apps/api/tests/test_alertmanager_rule_bypass.py b/apps/api/tests/test_alertmanager_rule_bypass.py
index 520d574c3..29b9eca35 100644
--- a/apps/api/tests/test_alertmanager_rule_bypass.py
+++ b/apps/api/tests/test_alertmanager_rule_bypass.py
@@ -132,6 +132,37 @@ def test_exact_alertmanager_chain_rule_precedes_llm() -> None:
     ) is True
 
 
+def test_exact_host188_callback_rule_precedes_llm() -> None:
+    rule_response = {
+        "rule_id": "telegram_callback_ingress_host188",
+        "suggested_action": "CONTROLLED_APPLY",
+        "kubectl_command": "",
+    }
+
+    assert _should_use_alertmanager_rule_first(
+        rule_response,
+        "ai_agent",
+    ) is True
+
+
+def test_host188_callback_rule_has_no_generic_kubernetes_action() -> None:
+    rule_response = match_rule(
+        {
+            "labels": {"alertname": "TelegramCallbackIngressUnverified"},
+            "alert_type": "custom",
+            "message": "callback ingress receipt is not fresh",
+            "severity": "warning",
+            "target_resource": "openclaw",
+        }
+    )
+
+    assert rule_response is not None
+    assert rule_response["rule_id"] == "telegram_callback_ingress_host188"
+    assert rule_response["risk_level"] == "medium"
+    assert rule_response["suggested_action"] == "CONTROLLED_APPLY"
+    assert rule_response["kubectl_command"] == ""
+
+
 def test_critical_alertmanager_incident_keeps_bounded_execution_risk_medium() -> None:
     rule_response = match_rule(
         {
diff --git a/apps/api/tests/test_classify_alert_early.py b/apps/api/tests/test_classify_alert_early.py
index 970de6c66..533bf220a 100644
--- a/apps/api/tests/test_classify_alert_early.py
+++ b/apps/api/tests/test_classify_alert_early.py
@@ -15,6 +15,17 @@ import pytest
 
 from src.services.incident_service import classify_alert_early
 
+
+def test_telegram_callback_ingress_routes_to_ai_agent_controlled_lane():
+    for alertname in (
+        "TelegramCallbackIngressUnverified",
+        "TelegramCallbackForwarderDrift",
+    ):
+        assert classify_alert_early(alertname, "warning", {"host": "188"}) == (
+            "ai_agent",
+            "TYPE-3",
+        )
+
 # --------------------------------------------------------------------------- #
 # TYPE-4D: Config Drift
 # --------------------------------------------------------------------------- #