diff --git a/apps/api/src/core/config.py b/apps/api/src/core/config.py
index 02457f930..9bad02662 100644
--- a/apps/api/src/core/config.py
+++ b/apps/api/src/core/config.py
@@ -62,6 +62,21 @@ class Settings(BaseSettings):
         description="Phase 24: True=新 AIRouter 路由, False=舊 openclaw.py fallback chain",
     )
 
+    # ==========================================================================
+    # P1-1: KMWriter 統一契約 (2026-04-28 ogt + Claude Sonnet 4.6)
+    # KM_WRITE_AWAIT=true  → 強制 await asyncio.wait_for(timeout=KM_WRITE_TIMEOUT_SECONDS)
+    # KM_WRITE_AWAIT=false → 舊 fire-and-forget 行為（回滾用）
+    # 回滾指令: kubectl set env deployment/awoooi-api KM_WRITE_AWAIT=false
+    # ==========================================================================
+    KM_WRITE_AWAIT: bool = Field(
+        default=True,
+        description="P1-1: True=強制 await KM 寫入（可靠）, False=fire-and-forget（舊行為，回滾用）",
+    )
+    KM_WRITE_TIMEOUT_SECONDS: float = Field(
+        default=5.0,
+        description="P1-1: KMWriter await timeout（秒），超時記錄 warning 但不阻塞主流程",
+    )
+
     # ==========================================================================
     # aider-watch v2 integration (2026-04-20 ADR-091)
     # 整合 Mac aider CLI 監控進 awoooi 飛輪（events → incident → ai_router feedback）
@@ -354,11 +369,13 @@ class Settings(BaseSettings):
 
     # ==========================================================================
     # AI Fallback Strategy (ADR-006 v1.3 + ADR-036)
-    # Order: Ollama (local) -> Gemini (cloud) -> Claude (cloud)
+    # Order: Ollama (local) -> NVIDIA NIM -> Gemini (cloud) -> Claude (cloud)
     # Tool Calling: Nemotron (專用) -> Gemini -> Claude
+    # 2026-04-28 ogt + Claude Opus 4.7: 補 nvidia 對齊 ConfigMap (k8s/awoooi-prod/04-configmap.yaml)
+    # 之前 default 缺 nvidia 與 ConfigMap drift，feedback_ai_fallback_order.md 鐵律
     # ==========================================================================
     AI_FALLBACK_ORDER: list[str] = Field(
-        default=["ollama", "gemini", "claude"],
+        default=["ollama", "nvidia", "gemini", "claude"],
         description="AI provider fallback order",
     )
     GEMINI_API_KEY: str = Field(default="", description="Google Gemini API key")
@@ -663,9 +680,12 @@ class Settings(BaseSettings):
     # ==========================================================================
     # MCP Phase 2b: Prometheus MCP Server (ADR-071, 2026-04-11 Claude Sonnet 4.6)
     # ==========================================================================
+    # 2026-04-29 ogt + Claude Opus 4.7: drift fix — 188 是 Ollama Hub，Prometheus 實際在 110
+    # ConfigMap 04-configmap.yaml 也是 110；governance_agent / SLO check 連 188 會 timeout
+    # 此 drift 是 SPF-4 (governance_agent silently fail) 根因之一
     PROMETHEUS_URL: str = Field(
-        default="http://192.168.0.188:9090",
-        description="Prometheus server URL",
+        default="http://192.168.0.110:9090",
+        description="Prometheus server URL (DevOps 金庫主機)",
     )
     PROMETHEUS_MCP_ENABLED: bool = Field(
         default=True,
@@ -685,8 +705,10 @@ class Settings(BaseSettings):
 
     # MCP Phase 3: ArgoCD MCP Server (2026-04-11 Claude Sonnet 4.6)
     # ==========================================================================
+    # 2026-04-29 ogt + Claude Opus 4.7: drift fix — ArgoCD 實際在 121，125 是舊位置
+    # ConfigMap 04-configmap.yaml 已寫 121；T0 G3 drift 已知，此次正式對齊
     ARGOCD_URL: str = Field(
-        default="https://192.168.0.125:30443",
+        default="https://192.168.0.121:30443",
         description="ArgoCD API Server URL（K3s NodePort HTTPS）",
     )
     ARGOCD_API_TOKEN: str = Field(
diff --git a/apps/api/src/services/governance_agent.py b/apps/api/src/services/governance_agent.py
index 86bafd0c1..815faefbc 100644
--- a/apps/api/src/services/governance_agent.py
+++ b/apps/api/src/services/governance_agent.py
@@ -303,7 +303,24 @@ class GovernanceAgent:
                     data = resp.json()
                     if data.get("status") == "success":
                         result_list = data.get("data", {}).get("result", [])
-                        value = float(result_list[0]["value"][1]) if result_list else 0.0
+                        # 2026-04-28 ogt + Claude Opus 4.7: P0-1 假警報止血
+                        # 空 result = Prometheus 查無資料（metric 未 emit / rule 未部署），不等於 SLO=0
+                        # ADR-100 emitter 全部尚未實作（automation_operation_log_total 等 4 個 counter 零定義）
+                        # 不可 fallback 0.0，否則必觸發 violated=True 噴假警報
+                        if not result_list:
+                            results[name] = {
+                                "error": "no_data",
+                                "skipped": True,
+                                "reason": "prometheus_empty_result_metric_not_emitted",
+                            }
+                            logger.warning(
+                                "governance_slo_no_data",
+                                slo=name,
+                                query=query,
+                                hint="ADR-100 emitter not yet implemented",
+                            )
+                            continue
+                        value = float(result_list[0]["value"][1])
                         threshold = hard_red_lines[name]
                         target = slo_targets[name]
                         violated = value < threshold
diff --git a/apps/api/src/services/telegram_gateway.py b/apps/api/src/services/telegram_gateway.py
index e978291fc..dc815c420 100644
--- a/apps/api/src/services/telegram_gateway.py
+++ b/apps/api/src/services/telegram_gateway.py
@@ -1658,7 +1658,19 @@ class TelegramGateway:
                     await redis.setex(key, 3600, value)
                 logger.debug("llm_button_redis_written", count=len(redis_writes))
             except Exception as exc:
-                logger.warning("llm_button_redis_write_failed", error=str(exc))
+                # 2026-04-28 ogt + Claude Opus 4.7: P0-4 鬼魂按鈕守門
+                # feedback_no_ghost_buttons.md 三缺一鐵律：callback 對應 short_id 找不到 = 鬼魂
+                # Redis 寫入失敗 → LLM 動態按鈕的 callback_data 在 Redis 撈不到 payload → 鬼魂風險
+                # 對策：清空 LLM 動態按鈕，caller (build_keyboard) 1488 行的 first_row 永遠保留
+                #       (✅ 批准 / ❌ 拒絕 用 HMAC nonce，無狀態，不依賴 Redis)
+                # 統帥仍可走核心通道，少了 LLM 推薦的 specific actions（可接受的降級）
+                logger.error(
+                    "llm_button_redis_write_failed_fallback_to_static",
+                    error=str(exc),
+                    dropped_count=len(btn_list),
+                    hint="user will see only first_row (approve/reject), LLM-recommended actions dropped",
+                )
+                btn_list.clear()
 
         # 每排最多 2 個
         rows: list[list[dict]] = [btn_list[i:i+2] for i in range(0, len(btn_list), 2)]
diff --git a/ops/alertmanager/alertmanager.yml b/ops/alertmanager/alertmanager.yml
index efa254c06..826070817 100644
--- a/ops/alertmanager/alertmanager.yml
+++ b/ops/alertmanager/alertmanager.yml
@@ -7,32 +7,43 @@
 # 架構: awoooi-webhook (主路徑) + telegram-direct (fallback，獨立路由)
 # 當 AWOOOI API 無法回應時，critical 告警直接送 Telegram Bot API
 # ⚠️  bot_token/chat_id 部署時由 secrets 替換，此檔為模板
+#
+# 2026-04-29 ogt + Claude Opus 4.7: P1-4 新版語法升級 + 因果抑制擴展
+# 改動：
+#   1. match/match_re → matchers (Alertmanager v0.27+ deprecated 警告)
+#   2. source_match/target_match/target_match_re → source_matchers/target_matchers
+#   3. group_by 加 team label（防 4 條 SLO 同秒爆，依 web-researcher 文件）
+#   4. PostgreSQLDown / RedisDown inhibit 補 equal: ['instance']（防全 ns 爆炸抑制）
+#   5. 新增 OllamaInstanceDown / KMConverterDown / SLO FastBurn 三組因果抑制
+#      根因：本次 4 SLO 雪崩證實 Ollama 111 掛 → AI 推理鏈斷 → SLO 級聯爆炸無守門
+#   6. 命名鐵律 feedback_telegram_alert_format.md 對齊（label team=ai/component/auto_repair）
 
 global:
   resolve_timeout: 5m
 
 route:
   receiver: 'awoooi-webhook'
-  group_by: ['alertname', 'severity']
+  # 2026-04-29: 加 team — SLO/AI PrometheusRule 含 team=ai 時可獨立分組合併
+  group_by: ['team', 'alertname', 'severity']
   group_wait: 30s
   group_interval: 5m
   repeat_interval: 4h
   routes:
-    - match:
-        severity: critical
+    - matchers:
+        - severity="critical"
       receiver: 'awoooi-webhook'
       group_wait: 10s
       # continue:true 讓 critical 同時送 telegram-direct (fallback)
       continue: true
-    - match:
-        severity: critical
+    - matchers:
+        - severity="critical"
       receiver: 'telegram-direct'
       group_wait: 10s
-    - match:
-        severity: warning
+    - matchers:
+        - severity="warning"
       receiver: 'awoooi-webhook'
-    - match_re:
-        alertname: 'Zombie.*|Container.*'
+    - matchers:
+        - alertname=~"Zombie.*|Container.*"
       receiver: 'awoooi-webhook'
       group_wait: 1m
 
@@ -41,6 +52,8 @@ receivers:
   # 2026-04-16 ogt + Claude Sonnet 4.6: 改指向 VIP 192.168.0.125
   #   根因: 121:32334 Connection Refused，120:32334 也 Refused
   #   只有 VIP 125:32334 可連通（kube-proxy NodePort 路由正常）
+  # ⚠️ SPF-1 風險：VIP 125 為單點，VIP host 整機 down → 主鏈斷
+  #    緩解計畫見 critic SPF 治理設計（中度方案：webhook_configs 多 url round-robin）
   - name: 'awoooi-webhook'
     webhook_configs:
       - url: 'http://192.168.0.125:32334/api/v1/webhooks/alertmanager'
@@ -48,6 +61,8 @@ receivers:
 
   # Fallback 路徑: AWOOOI API 掛掉時，critical 告警直接送 Telegram
   # 只有 critical severity 走此路徑（避免 warning 雙重通知）
+  # ⚠️ bot_token / chat_id 由 CD pipeline 在 deploy 時用 K8s Secret 注入
+  #    feedback_telegram_secrets_injection.md 鐵律：禁止 PLACEHOLDER 上線
   - name: 'telegram-direct'
     telegram_configs:
       - bot_token: 'TELEGRAM_BOT_TOKEN_PLACEHOLDER'
@@ -65,26 +80,60 @@ receivers:
         send_resolved: false
 
 inhibit_rules:
-  - source_match:
-      severity: critical
-    target_match:
-      severity: warning
+  # === 基礎因果抑制（原有規則，新語法重寫）===
+  - source_matchers:
+      - severity="critical"
+    target_matchers:
+      - severity="warning"
     equal: ['alertname', 'instance']
-  - source_match:
-      alertname: HostDown
-    target_match_re:
-      alertname: 'HostHighCpuLoad|HostOutOfMemory|HostOutOfDiskSpace'
+
+  - source_matchers:
+      - alertname="HostDown"
+    target_matchers:
+      - alertname=~"HostHighCpuLoad|HostOutOfMemory|HostOutOfDiskSpace"
     equal: ['host']
-  - source_match:
-      alertname: KubeNodeNotReady
-    target_match_re:
-      alertname: 'KubePodCrashLooping|KubePodNotReady|KubeDeploymentReplicasMismatch'
+
+  - source_matchers:
+      - alertname="KubeNodeNotReady"
+    target_matchers:
+      - alertname=~"KubePodCrashLooping|KubePodNotReady|KubeDeploymentReplicasMismatch"
     equal: ['node']
-  - source_match:
-      alertname: PostgreSQLDown
-    target_match:
-      alertname: PostgreSQLHighConnections
-  - source_match:
-      alertname: RedisDown
-    target_match:
-      alertname: RedisMemoryHigh
+
+  # 2026-04-29: 補 equal: ['instance'] — 原本缺，PG 在 instance A down
+  # 不該抑制 instance B 的 HighConnections（爆炸抑制 bug）
+  - source_matchers:
+      - alertname="PostgreSQLDown"
+    target_matchers:
+      - alertname="PostgreSQLHighConnections"
+    equal: ['instance']
+
+  - source_matchers:
+      - alertname="RedisDown"
+    target_matchers:
+      - alertname="RedisMemoryHigh"
+    equal: ['instance']
+
+  # === 新增：AI 鏈因果抑制（2026-04-29 ADR-035 因果抑制擴展）===
+  # 根因：本次 4 SLO 雪崩證實 Ollama 111 掛 → AI 推理鏈斷 → 4 SLO 同秒爆
+  #       無此抑制 → 假警報淹沒真警報（Ollama down 本身才是真信號）
+
+  # Ollama 任一實例掛 → 抑制所有 AI/SLO 告警 30 分鐘
+  - source_matchers:
+      - alertname="OllamaInstanceDown"
+    target_matchers:
+      - alertname=~"SLO_.*|AI_.*"
+    equal: []
+
+  # KM converter 掛 → 抑制 KM Growth Rate SLO（避免 KM 寫入失敗本身觸發 SLO）
+  - source_matchers:
+      - alertname="KMConverterDown"
+    target_matchers:
+      - alertname=~"SLO_KMGrowthRate.*"
+    equal: []
+
+  # 同 SLO 較嚴重抑制較輕（FastBurn 抑制 Medium/Slow Burn）
+  - source_matchers:
+      - alertname=~"SLO_.+_FastBurn"
+    target_matchers:
+      - alertname=~"SLO_.+_(Medium|Slow)Burn"
+    equal: ['alertname']
diff --git a/scripts/check_config_drift.py b/scripts/check_config_drift.py
new file mode 100755
index 000000000..9aa29fe70
--- /dev/null
+++ b/scripts/check_config_drift.py
@@ -0,0 +1,109 @@
+#!/usr/bin/env python3
+# 2026-04-28 ogt + Claude Opus 4.7: P2-1 ConfigMap vs code default drift checker
+# 來源：tool-expert 統一治理方案
+# 目的：CI / pre-commit 階段驗證 k8s ConfigMap 與 apps/api/src/core/config.py default 一致
+# 違反「事實驅動」紅線案例：AI_FALLBACK_ORDER、ARGOCD_URL 都曾發生 drift
+"""
+ConfigMap vs Code Default Drift Checker
+
+用法：
+    python3 scripts/check_config_drift.py
+退出碼：
+    0 = 全部對齊
+    1 = 至少一項 drift（CI 應 fail）
+
+可加進 .pre-commit-config.yaml：
+- repo: local
+  hooks:
+    - id: config-drift-check
+      name: ConfigMap vs code default drift
+      entry: python3 scripts/check_config_drift.py
+      language: python
+      pass_filenames: false
+      additional_dependencies: [pyyaml]
+"""
+from __future__ import annotations
+
+import json
+import re
+import sys
+from pathlib import Path
+
+import yaml  # noqa: F401  pre-commit 會經 additional_dependencies 安裝
+
+ROOT = Path(__file__).resolve().parent.parent
+CONFIGMAP_PATH = ROOT / "k8s" / "awoooi-prod" / "04-configmap.yaml"
+CONFIG_PY_PATH = ROOT / "apps" / "api" / "src" / "core" / "config.py"
+
+# 需要比對的欄位
+# code_default_pattern: 在 config.py 找 default=... 用的 regex（DOTALL）
+CHECK_FIELDS: dict[str, dict[str, str]] = {
+    "AI_FALLBACK_ORDER": {
+        "configmap_key": "AI_FALLBACK_ORDER",
+        "code_pattern": r"AI_FALLBACK_ORDER:\s*list\[str\]\s*=\s*Field\([^)]*?default=(\[[^\]]+\])",
+    },
+    "ARGOCD_URL": {
+        "configmap_key": "ARGOCD_URL",
+        "code_pattern": r"ARGOCD_URL[^\n]*?\n[^)]*?default=[\"']([^\"']+)[\"']",
+    },
+    "PROMETHEUS_URL": {
+        "configmap_key": "PROMETHEUS_URL",
+        "code_pattern": r"PROMETHEUS_URL[^\n]*?\n[^)]*?default=[\"']([^\"']+)[\"']",
+    },
+    "OLLAMA_URL": {
+        "configmap_key": "OLLAMA_URL",
+        "code_pattern": r"OLLAMA_URL[^\n]*?\n[^)]*?default=[\"']([^\"']+)[\"']",
+    },
+}
+
+
+def _normalize(raw: str) -> object:
+    """嘗試把字串解析成 list/dict，失敗就回原字串。"""
+    raw_strip = raw.strip().strip("'\"")
+    if raw_strip.startswith("["):
+        try:
+            return json.loads(raw_strip.replace("'", '"'))
+        except json.JSONDecodeError:
+            return raw_strip
+    return raw_strip
+
+
+def main() -> int:
+    if not CONFIGMAP_PATH.exists():
+        print(f"[ERROR] ConfigMap not found: {CONFIGMAP_PATH}")
+        return 2
+    if not CONFIG_PY_PATH.exists():
+        print(f"[ERROR] config.py not found: {CONFIG_PY_PATH}")
+        return 2
+
+    with CONFIGMAP_PATH.open() as fh:
+        cm_data: dict = yaml.safe_load(fh).get("data", {}) or {}
+    py_src = CONFIG_PY_PATH.read_text()
+
+    exit_code = 0
+    print("=== ConfigMap ↔ code.default Drift Check ===")
+    for field, spec in CHECK_FIELDS.items():
+        cm_raw = cm_data.get(spec["configmap_key"], "<MISSING_IN_CONFIGMAP>")
+        m = re.search(spec["code_pattern"], py_src, re.DOTALL)
+        py_raw = m.group(1) if m else "<NOT_FOUND_IN_CONFIG_PY>"
+
+        cm_val = _normalize(cm_raw)
+        py_val = _normalize(py_raw)
+
+        if cm_val == py_val:
+            print(f"[OK]    {field}: {cm_val}")
+        else:
+            print(f"[DRIFT] {field}:")
+            print(f"        ConfigMap   = {cm_val}")
+            print(f"        config.py   = {py_val}")
+            exit_code = 1
+
+    if exit_code == 0:
+        print("=== All drift-check fields aligned ===")
+    else:
+        print("=== DRIFT detected, fix the inconsistency ===")
+    return exit_code
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/scripts/health_check_session.sh b/scripts/health_check_session.sh
new file mode 100755
index 000000000..d93924cb2
--- /dev/null
+++ b/scripts/health_check_session.sh
@@ -0,0 +1,110 @@
+#!/usr/bin/env bash
+# 2026-04-28 ogt + Claude Opus 4.7: P2-2 Session 啟動健康驗證
+# 來源：tool-expert 統一治理方案
+# 目的：每次 Claude session 啟動時快速確認 5 主機 + 關鍵服務可達
+# 純 read-only（curl + ssh -o BatchMode），不修改任何狀態
+#
+# 用法：
+#   bash scripts/health_check_session.sh
+#   或加 alias: alias awoooi-health='bash ~/awoooi/scripts/health_check_session.sh'
+
+set -uo pipefail  # 不要 -e，個別 check 失敗不阻擋全部
+
+RED='\033[0;31m'
+GREEN='\033[0;32m'
+YELLOW='\033[0;33m'
+NC='\033[0m'
+
+ok() { printf "${GREEN}[OK]${NC}    %s\n" "$1"; }
+fail() { printf "${RED}[FAIL]${NC}  %s\n" "$1"; }
+warn() { printf "${YELLOW}[WARN]${NC}  %s\n" "$1"; }
+
+check_url() {
+    local name=$1 url=$2
+    local code
+    code=$(curl -sk --max-time 3 -o /dev/null -w "%{http_code}" "$url" 2>/dev/null || echo "000")
+    if [[ "$code" =~ ^[23] ]]; then
+        ok "$name → $url ($code)"
+        return 0
+    elif [[ "$code" =~ ^[45] ]]; then
+        warn "$name → $url ($code, 服務有回應但非 2xx/3xx)"
+        return 0
+    else
+        fail "$name → $url (unreachable)"
+        return 1
+    fi
+}
+
+check_ssh() {
+    local name=$1 host=$2
+    if ssh -o ConnectTimeout=3 -o BatchMode=yes -o StrictHostKeyChecking=no \
+            "$host" "echo ok" 2>/dev/null | grep -q ok; then
+        ok "SSH $name ($host)"
+        return 0
+    else
+        fail "SSH $name ($host) — 無法連線（timeout / 認證失敗 / 主機不可達）"
+        return 1
+    fi
+}
+
+ROOT="$(cd "$(dirname "$0")/.." && pwd)"
+TS="$(date '+%Y-%m-%d %H:%M %Z')"
+
+echo "=========================================="
+echo "AWOOOI Session Health Check $TS"
+echo "=========================================="
+
+echo ""
+echo "--- K8s 控制平面 ---"
+check_url "K3s VIP API"     "https://192.168.0.125:6443/healthz"
+check_url "ArgoCD (121)"    "https://192.168.0.121:30443"
+
+echo ""
+echo "--- AI 推理層 ---"
+check_url "Ollama 111 GPU"  "http://192.168.0.111:11434/api/tags"
+check_url "Ollama 188 Hub"  "http://192.168.0.188:11434/api/tags"
+
+echo ""
+echo "--- 觀測層 ---"
+check_url "Prometheus 110"  "http://192.168.0.110:9090/-/healthy"
+check_url "Alertmanager 110" "http://192.168.0.110:9093/-/healthy"
+check_url "Gitea 110"       "http://192.168.0.110:3001"
+check_url "Langfuse 110"    "http://192.168.0.110:3100"
+
+echo ""
+echo "--- AWOOOI 核心服務 (prod NodePort) ---"
+check_url "AWOOOI API (125)" "http://192.168.0.125:32334/api/v1/health"
+
+echo ""
+echo "--- SSH 連通 ---"
+check_ssh "awoooi-devops (110)" "wooo@192.168.0.110"
+check_ssh "k3s-1 (120)" "wooo@192.168.0.120"
+check_ssh "k3s-2 (121)" "wooo@192.168.0.121"
+check_ssh "ollama-111-gpu (ProxyJump 110)" "ollama-111-gpu"
+
+echo ""
+echo "--- Config Drift Check ---"
+if [ -x "$ROOT/scripts/check_config_drift.py" ]; then
+    python3 "$ROOT/scripts/check_config_drift.py" || warn "config drift detected (見上方 [DRIFT] 行)"
+else
+    warn "drift checker 不存在 ($ROOT/scripts/check_config_drift.py)"
+fi
+
+echo ""
+echo "--- Git 狀態 ---"
+if [ -d "$ROOT/.git" ]; then
+    cd "$ROOT" || exit
+    branch=$(git branch --show-current 2>/dev/null || echo "<detached>")
+    upstream_diff=$(git rev-list --count "@{u}..HEAD" 2>/dev/null || echo "?")
+    echo "  分支: $branch (本地超前上游 $upstream_diff 個 commit)"
+    if ! git diff --quiet 2>/dev/null; then
+        warn "  有未 commit 的變更（git status 自查）"
+    else
+        ok "  工作目錄 clean"
+    fi
+fi
+
+echo ""
+echo "=========================================="
+echo "Session Health Check 結束"
+echo "=========================================="