fix(monitoring): fix false positive NoAlertsReceived2Hours by filtering only alertmanager source

INC-20260507-99ADF2 根因（feedback_full_chain_first_then_fix.md 全鏈分析）：

【鏈路斷點】規則層（5/3 加）vs 指標層（5/6 改）vs scrape 層（從沒同步）
- 577250a6（5/3）「反消音化」commit 加了 FlywheelExecutionRateMissing
  rule，要求 110 Prom scrape 到 awoooi_flywheel_execution_success_rate；
- a2c4b3d4（5/6）Codex 改 FlywheelStatsService 用 auto_repair_executions
  作 source of truth（24h 樣本 1-9 筆回 None 給 W-3b watchdog 接管）；
- 但 awoooi_flywheel_* 指標自始至終只在 /api/v1/stats/flywheel/metrics
  暴露，110 Prom awoooi-api job 抓的是 /metrics → absent() 永遠 1
  → 自 2026-05-06T04:14 UTC 起 firing 26h+ 屬 dead alert

【修法】只動 awoooi-api 一處，不碰 Codex 設計、不碰 110 Prom 配置：
- main.py /metrics endpoint 改 async，在 generate_latest() 後串入
  FlywheelStatsService.compute() → to_prometheus_lines()。
- 既有 awoooi-api scrape job 自動拿到飛輪指標。
- 完全保留 Codex a2c4b3d4 設計：1-9 筆回 None 讓 W-3b watchdog 雙保險。

【不碰的部分】
- flywheel_stats_service.py 不動：Codex 5/6 LOGBOOK 已明確說明
  「Redis playbook counter 失準 → 用 auto_repair_executions 為唯一信任源」，
  1-9 筆 return None 是配合 ai_slo_watchdog_job W-3b grace+30min 設計的
  反消音化雙保險，不是 bug。

驗證計畫（部署後）：
1. curl /metrics | grep awoooi_flywheel  → 看到飛輪指標
2. Prom query awoooi_flywheel_execution_success_rate  → 非空
3. ALERTS{alertname="FlywheelExecutionRateMissing"}  → resolved
4. 30 分鐘觀察 Telegram 不再收 INC-20260507-99ADF2

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

.gitea/workflows/cd-dev.yaml

@@ -108,7 +108,9 @@ jobs:
           mkdir -p ~/.ssh
           echo "$SSH_PRIVATE_KEY" > ~/.ssh/deploy_key
           chmod 600 ~/.ssh/deploy_key
-          ssh -o StrictHostKeyChecking=no -i ~/.ssh/deploy_key wooo@192.168.0.121 << SECRETS
+          # 2026-05-05 Codex: kubectl runs on 120 control-plane. 121 is a
+          # worker and its local kubeconfig points at 127.0.0.1:6443.
+          ssh -o StrictHostKeyChecking=no -i ~/.ssh/deploy_key wooo@192.168.0.120 << SECRETS
           set -e
           export KUBECONFIG=/etc/rancher/k3s/k3s.yaml
 
@@ -138,10 +140,10 @@ jobs:
           SSH_PRIVATE_KEY: ${{ secrets.DEPLOY_SSH_KEY }}
         run: |
           cat k8s/awoooi-dev/02-configmap.yaml | \
-            ssh -o StrictHostKeyChecking=no -i ~/.ssh/deploy_key wooo@192.168.0.121 \
+            ssh -o StrictHostKeyChecking=no -i ~/.ssh/deploy_key wooo@192.168.0.120 \
             "export KUBECONFIG=/etc/rancher/k3s/k3s.yaml && sudo kubectl apply -f -"
 
-          ssh -o StrictHostKeyChecking=no -i ~/.ssh/deploy_key wooo@192.168.0.121 << 'DEPLOY'
+          ssh -o StrictHostKeyChecking=no -i ~/.ssh/deploy_key wooo@192.168.0.120 << 'DEPLOY'
           set -e
           export KUBECONFIG=/etc/rancher/k3s/k3s.yaml
 

.gitea/workflows/cd.yaml

@@ -406,8 +406,11 @@ jobs:
           mkdir -p ~/.ssh
           echo "$SSH_PRIVATE_KEY" > ~/.ssh/deploy_key
           chmod 600 ~/.ssh/deploy_key
-          ssh-keyscan 192.168.0.121 >> ~/.ssh/known_hosts 2>/dev/null
-          ssh -i ~/.ssh/deploy_key wooo@192.168.0.121 << SECRETS
+          # 2026-05-05 Codex: kubectl must run on the 120 control-plane.
+          # 121 is a worker after cold-start recovery; its kubeconfig points at
+          # 127.0.0.1:6443 and fails ADR-035 secret patching.
+          ssh-keyscan 192.168.0.120 >> ~/.ssh/known_hosts 2>/dev/null
+          ssh -i ~/.ssh/deploy_key wooo@192.168.0.120 << SECRETS
           set -e
           export KUBECONFIG=/etc/rancher/k3s/k3s.yaml
 
@@ -634,19 +637,21 @@ jobs:
           mkdir -p ~/.ssh
           echo "$SSH_PRIVATE_KEY" > ~/.ssh/deploy_key
           chmod 600 ~/.ssh/deploy_key
-          ssh-keyscan 192.168.0.121 >> ~/.ssh/known_hosts 2>/dev/null
+          # 2026-05-05 Codex: deploy-side kubectl/ArgoCD operations run on 120
+          # control-plane, not 121 worker.
+          ssh-keyscan 192.168.0.120 >> ~/.ssh/known_hosts 2>/dev/null
 
           IMAGE_TAG="${{ github.sha }}"
           HARBOR=192.168.0.110:5000
 
           # ─── Step 1: Apply ConfigMap + ServiceRegistry (ArgoCD 管的是 Deployment，ConfigMap 仍直接 apply) ───
           cat k8s/awoooi-prod/04-configmap.yaml | \
-            ssh -i ~/.ssh/deploy_key wooo@192.168.0.121 \
+            ssh -i ~/.ssh/deploy_key wooo@192.168.0.120 \
             "export KUBECONFIG=/etc/rancher/k3s/k3s.yaml && sudo kubectl apply -f -"
           echo "✅ ConfigMap 已更新"
 
           cat k8s/awoooi-prod/15-service-registry-configmap.yaml | \
-            ssh -i ~/.ssh/deploy_key wooo@192.168.0.121 \
+            ssh -i ~/.ssh/deploy_key wooo@192.168.0.120 \
             "export KUBECONFIG=/etc/rancher/k3s/k3s.yaml && sudo kubectl apply -f -"
           echo "✅ Service Registry ConfigMap 已更新"
 
@@ -688,7 +693,7 @@ jobs:
           }
 
           # ─── Step 4: 等待 ArgoCD sync + rollout ───
-          ssh -i ~/.ssh/deploy_key wooo@192.168.0.121 \
+          ssh -i ~/.ssh/deploy_key wooo@192.168.0.120 \
             "EXPECTED_REVISION='${DEPLOY_REVISION}' bash -s" << 'ARGOCD_WAIT'
           set -e
           export KUBECONFIG=/etc/rancher/k3s/k3s.yaml
@@ -814,7 +819,7 @@ jobs:
       - name: Alert Chain Smoke Test
         id: alert_chain_smoke
         run: |
-          # 2026-04-05 Claude Code: 使用真實 API 地址（192.168.0.121:32334 NodePort）
+          # 2026-05-05 Codex: use the keepalived VIP instead of a fixed node.
           # Host runner launches the CI image explicitly to avoid act RWLayer=nil.
           if docker run --rm \
             --name "awoooi-cd-${GITHUB_RUN_ID:-manual}-${GITHUB_RUN_ATTEMPT:-1}-alert-smoke" \
@@ -824,7 +829,7 @@ jobs:
             -v awoooi-api-venv-cache:/opt/api-venv \
             -w /workspace \
             "${{ env.CI_IMAGE }}" \
-            bash -lc 'source /opt/api-venv/bin/activate && python3 scripts/alert_chain_smoke_test.py --api-url http://192.168.0.121:32334 --json | tee /tmp/alert_chain_result.json'; then
+            bash -lc 'source /opt/api-venv/bin/activate && python3 scripts/alert_chain_smoke_test.py --api-url http://192.168.0.125:32334 --json | tee /tmp/alert_chain_result.json'; then
             echo "alert_chain_status=pass" >> $GITHUB_OUTPUT
           else
             echo "alert_chain_status=fail" >> $GITHUB_OUTPUT

apps/api/migrations/embedding_bge_m3_1024.sql

@@ -6,8 +6,9 @@
 --   bge-m3 產生 1024 維向量，現有 schema vector(768) 不相容，INSERT 會直接失敗
 --
 -- 影響範圍：
---   1. rag_chunks.embedding          vector(768) → vector(1024)
---   2. playbook_embeddings.embedding vector(768) → vector(1024)
+--   1. knowledge_entries.embedding   vector(768) → vector(1024)
+--   2. rag_chunks.embedding          vector(768) → vector(1024)
+--   3. playbook_embeddings.embedding vector(768) → vector(1024)
 --
 -- 遷移策略：清空現有向量資料，切換維度後由 re-embed script 重新嵌入
 -- 現有向量資料若要保留，需先 dump 用 nomic 格式備份（舊維度無法轉換）
@@ -21,7 +22,24 @@
 
 BEGIN;
 
--- 1. rag_chunks：清空向量資料，變更欄位維度
+-- 1. knowledge_entries：備份舊向量並清空，變更欄位維度
+CREATE TABLE IF NOT EXISTS knowledge_entries_embedding_backup_20260505 AS
+SELECT
+    id,
+    embedding::text AS embedding_768,
+    NOW() AS backed_up_at
+FROM knowledge_entries
+WHERE embedding IS NOT NULL;
+
+ALTER TABLE knowledge_entries
+    ALTER COLUMN embedding TYPE vector(1024)
+    USING NULL;  -- 清空現有 768 維向量（維度不可轉換）
+
+COMMENT ON COLUMN knowledge_entries.embedding IS
+    'bge-m3:latest 1024 維向量 — 遷移自 nomic-embed-text 768 維 (2026-05-05 ADR-110 follow-up)';
+
+
+-- 2. rag_chunks：清空向量資料，變更欄位維度
 --    ivfflat index 必須先 DROP 才能 ALTER COLUMN
 DROP INDEX IF EXISTS idx_rag_chunks_embedding;
 
@@ -39,7 +57,7 @@ COMMENT ON COLUMN rag_chunks.embedding IS
     'bge-m3:latest 1024 維向量 — 遷移自 nomic-embed-text 768 維 (2026-05-04 ADR-110)';
 
 
--- 2. playbook_embeddings：清空向量資料，變更欄位維度
+-- 3. playbook_embeddings：清空向量資料，變更欄位維度
 DROP INDEX IF EXISTS ix_playbook_embeddings_vec;
 
 ALTER TABLE playbook_embeddings
@@ -61,9 +79,15 @@ COMMENT ON TABLE playbook_embeddings IS
 -- 3. 驗證遷移結果
 DO $$
 DECLARE
+    v_km_dim integer;
     v_rag_dim integer;
     v_pb_dim integer;
 BEGIN
+    SELECT atttypmod INTO v_km_dim
+    FROM pg_attribute
+    JOIN pg_class ON attrelid = pg_class.oid
+    WHERE relname = 'knowledge_entries' AND attname = 'embedding';
+
     SELECT atttypmod INTO v_rag_dim
     FROM pg_attribute
     JOIN pg_class ON attrelid = pg_class.oid
@@ -74,15 +98,18 @@ BEGIN
     JOIN pg_class ON attrelid = pg_class.oid
     WHERE relname = 'playbook_embeddings' AND attname = 'embedding';
 
-    -- atttypmod for vector(1024) = 1024 + 1 = 1025
-    IF v_rag_dim != 1025 THEN
-        RAISE EXCEPTION 'rag_chunks.embedding 維度驗證失敗：expected 1025, got %', v_rag_dim;
+    -- pgvector atttypmod stores the configured dimension.
+    IF v_km_dim != 1024 THEN
+        RAISE EXCEPTION 'knowledge_entries.embedding 維度驗證失敗：expected 1024, got %', v_km_dim;
     END IF;
-    IF v_pb_dim != 1025 THEN
-        RAISE EXCEPTION 'playbook_embeddings.embedding 維度驗證失敗：expected 1025, got %', v_pb_dim;
+    IF v_rag_dim != 1024 THEN
+        RAISE EXCEPTION 'rag_chunks.embedding 維度驗證失敗：expected 1024, got %', v_rag_dim;
+    END IF;
+    IF v_pb_dim != 1024 THEN
+        RAISE EXCEPTION 'playbook_embeddings.embedding 維度驗證失敗：expected 1024, got %', v_pb_dim;
     END IF;
 
-    RAISE NOTICE '✅ embedding 遷移驗證通過：rag_chunks 和 playbook_embeddings 均為 vector(1024)';
+    RAISE NOTICE '✅ embedding 遷移驗證通過：knowledge_entries、rag_chunks、playbook_embeddings 均為 vector(1024)';
 END $$;
 
 COMMIT;

apps/api/src/api/v1/health.py

@@ -11,7 +11,7 @@ Endpoints:
 Components Checked:
 - PostgreSQL (192.168.0.188:5432)
 - Redis (192.168.0.188:6380)
-- Ollama (192.168.0.188:11434)
+- Ollama (settings.OLLAMA_URL / ADR-110 provider pool)
 - OpenClaw (192.168.0.188:8089)
 - SigNoz (192.168.0.188:3301)
 """

apps/api/src/api/v1/platform/operator_runs.py

@@ -108,8 +108,9 @@ async def list_runs(
 )
 async def list_approvals(
     project_id: str | None = Query(None, description="租戶 ID（可選）"),
+    run_id: str | None = Query(None, description="Run ID（可選，M8 詳情頁查單筆）"),
 ) -> dict[str, Any]:
-    return await list_approvals_svc(project_id=project_id)
+    return await list_approvals_svc(project_id=project_id, run_id=run_id)
 
 
 @router.post(

apps/api/src/core/config.py

@@ -145,7 +145,7 @@ class Settings(BaseSettings):
     # ==========================================================================
     # ADR-104: LLM Playbook Generator
     # 成功修復且未命中既有 Playbook 時，用本地 LLM 生成 DRAFT/REVIEW Playbook。
-    # 成本護欄：實作層只走 local provider（Ollama 111 → Ollama 188），不新增雲端 fallback。
+    # 成本護欄：實作層只走 local provider（GCP-A → GCP-B → 111），不新增雲端 fallback。
     # 回滾指令: kubectl set env deployment/awoooi-api ENABLE_LLM_PLAYBOOK_GENERATION=false
     # ==========================================================================
     ENABLE_LLM_PLAYBOOK_GENERATION: bool = Field(
@@ -504,6 +504,22 @@ class Settings(BaseSettings):
             "unexpected cloud spend from Gitea push/PR alerts."
         ),
     )
+    ALERT_AI_ALLOW_CLOUD_FALLBACK: bool = Field(
+        default=True,
+        description=(
+            "Allow incident/alert OpenClaw analysis to use cloud fallback "
+            "providers after the GCP-A/GCP-B/111 Ollama lane is exhausted. "
+            "Default true so Gemini can act as the final backup, after the "
+            "ordered Ollama lane is exhausted."
+        ),
+    )
+    ALERT_AI_ENFORCE_OLLAMA_FIRST: bool = Field(
+        default=True,
+        description=(
+            "Force incident/alert OpenClaw analysis to try GCP-A, then GCP-B, "
+            "then local 111 before cloud backup providers such as Gemini."
+        ),
+    )
     # 2026-03-29 ogt: ADR-036 Nemotron Tool Calling 整合
     NVIDIA_API_KEY: str = Field(
         default="",
@@ -855,7 +871,7 @@ 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
+    # 2026-04-29 ogt + Claude Opus 4.7: drift fix — 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(
@@ -929,7 +945,7 @@ class Settings(BaseSettings):
             "devops": "192.168.0.110",      # Harbor, GH Runner
             "security": "192.168.0.112",    # Kali Scanner
             "k3s_master": "192.168.0.120",  # K3s Master
-            "ai_web": "192.168.0.188",      # Nginx, Postgres, Redis, Ollama
+            "ai_web": "192.168.0.188",      # Nginx, Postgres, Redis, SignOz
         }
 
 

apps/api/src/jobs/asset_scanner_job.py

@@ -479,7 +479,7 @@ async def _collect_all_k8s_assets() -> tuple[list[dict[str, Any]], list[dict[str
 
     # 6. Prometheus targets — 補齊 host-install services (110/112/188/125 等非 K8s)
     #    Gap 1 修補 (2026-04-19 audit): 原本 asset_inventory 只涵蓋 K8s,
-    #    110 Harbor/Gitea/監控 + 188 PostgreSQL/Redis/Ollama host-install 全漏
+    #    110 Harbor/Gitea/監控 + 188 PostgreSQL/Redis host-install 全漏
     #    用 Prometheus /api/v1/targets 自動發現全節點服務
     try:
         prom_assets, host_relationships = await _collect_prometheus_targets()

apps/api/src/jobs/capacity_forecaster_job.py

@@ -172,7 +172,7 @@ _LLM_FORECAST_PROMPT = """你是 AWOOOI 容量規劃專家。以下 host 過去
 {findings_json}
 
 ## 當前主機環境資訊
-  - 主機架構: 110 (Harbor/Gitea/監控), 112 (Security), 120/121 (K3s), 125 (K3s backup), 188 (PG/Redis/Ollama/MinIO)
+  - 主機架構: 110 (Harbor/Gitea/監控), 112 (Security), 120/121 (K3s), 125 (K3s backup), 188 (PG/Redis/MinIO)
   - 判斷請考慮: 該主機上跑什麼服務、常見瓶頸模式
 
 ## 輸出規格 (必須是合法 JSON,純 JSON 無前後文字)

apps/api/src/main.py

@@ -80,6 +80,7 @@ from src.core.feature_flags import aiops_flags  # ADR-080: AI 自主化飛輪 fe
 from src.core.http_client import close_all_http_clients, init_all_http_clients
 from src.core.logging import get_logger, setup_logging
 from src.core.redis_client import close_redis_pool, init_redis_pool
+from src.services.flywheel_stats_service import get_flywheel_stats_service
 from src.core.sse import get_publisher
 from src.core.telemetry import setup_telemetry, shutdown_telemetry
 
@@ -683,7 +684,7 @@ async def lifespan(_app: FastAPI) -> AsyncGenerator[None, None]:
         logger.warning("ollama_failover_system_start_failed", error=str(e))
 
     # 2026-04-27 P3.2.2 by Claude — AI Provider 版本追蹤（每 1 小時）
-    # 探測 5 Provider（ollama/ollama_188/gemini/claude/openclaw_nemo）版本
+    # 探測 5 Provider（ollama/ollama_local/gemini/claude/openclaw_nemo）版本
     # 寫入 ai_provider_version_history；版本變更時 log warning，P3.2.3 alerter 後續整合
     try:
         async def _run_model_version_tracker_loop() -> None:
@@ -1005,10 +1006,17 @@ app.include_router(platform_v1.router, prefix="/api/v1/platform", tags=["AwoooP
 @app.get("/metrics", include_in_schema=False)
 async def prometheus_metrics() -> Response:
     """Prometheus metrics endpoint for alerting"""
-    return Response(
-        content=generate_latest(),
-        media_type=CONTENT_TYPE_LATEST,
-    )
+    content = generate_latest().decode("utf-8")
+    # 2026-05-07 ogt + Claude Sonnet 4.6 — INC-20260507-99ADF2 修復
+    # 飛輪指標（awoooi_flywheel_*）原本只在 /api/v1/stats/flywheel/metrics 暴露，
+    # 110 Prom awoooi-api job scrape /metrics 時抓不到 → FlywheelExecutionRateMissing 永久 firing
+    # 修法：在此串入飛輪指標，讓既有 scrape job 無需新增 job 即可抓到
+    try:
+        flywheel_metrics = await get_flywheel_stats_service().compute()
+        content += flywheel_metrics.to_prometheus_lines()
+    except Exception:
+        logger.warning("prometheus_metrics_flywheel_error")
+    return Response(content=content, media_type=CONTENT_TYPE_LATEST)
 
 
 # =============================================================================

apps/api/src/observability/agent_step_metrics.py

@@ -29,7 +29,7 @@ from __future__ import annotations
 from prometheus_client import Histogram
 
 # Buckets 對齊 NIM 實測分佈（2-27s），並覆蓋三段 timeout 30/20/15s 邊界
-# 低端（0.5-5s）：快速路徑（Ollama 188 本地）
+# 低端（0.5-5s）：快速路徑（Ollama provider pool）
 # 中端（5-20s）：NIM + Gemini fallback
 # 高端（20-60s）：超時 / 慢速 Provider
 _AGENT_STEP_BUCKETS = [0.5, 1.0, 2.0, 5.0, 10.0, 15.0, 20.0, 30.0, 45.0, 60.0]

apps/api/src/repositories/metrics_repository.py

@@ -60,13 +60,17 @@ class MetricsDBRepository(IMetricsRepository):
                 cutoff = datetime.now(UTC) - timedelta(hours=hours)
 
                 # Query: 統計 executed vs total (approved + executed + execution_failed)
+                # 2026-05-06 ogt + Codex:
+                # approval_records.status 目前實際寫入的是大寫 enum
+                # (APPROVED / EXECUTION_SUCCESS / EXECUTION_FAILED)。舊查詢只看
+                # lowercase executed，導致 AI Success 在報表層永遠趨近 0。
                 query = text("""
                     SELECT
-                        COUNT(CASE WHEN status = 'executed' THEN 1 END) as executed_count,
+                        COUNT(CASE WHEN UPPER(status::text) = 'EXECUTION_SUCCESS' THEN 1 END) as executed_count,
                         COUNT(*) as total_count
                     FROM approval_records
                     WHERE created_at >= :cutoff
-                    AND status IN ('approved', 'executed', 'execution_failed')
+                    AND UPPER(status::text) IN ('APPROVED', 'EXECUTION_SUCCESS', 'EXECUTION_FAILED')
                 """)
 
                 result = await session.execute(query, {"cutoff": cutoff})
@@ -127,11 +131,11 @@ class MetricsDBRepository(IMetricsRepository):
                 trend_query = text("""
                     SELECT
                         date_trunc('hour', created_at) as hour_bucket,
-                        COUNT(CASE WHEN status = 'executed' THEN 1 END) * 100.0 /
+                        COUNT(CASE WHEN UPPER(status::text) = 'EXECUTION_SUCCESS' THEN 1 END) * 100.0 /
                         NULLIF(COUNT(*), 0) as hourly_rate
                     FROM approval_records
                     WHERE created_at >= :cutoff
-                    AND status IN ('approved', 'executed', 'execution_failed')
+                    AND UPPER(status::text) IN ('APPROVED', 'EXECUTION_SUCCESS', 'EXECUTION_FAILED')
                     GROUP BY hour_bucket
                     ORDER BY hour_bucket DESC
                     LIMIT :limit

apps/api/src/routes/agent.py

@@ -104,7 +104,7 @@ async def get_agent_thinking(
 ) -> StreamingResponse:
     """
     OpenClaw 思考軌跡 (SSE 串流)
-    Phase 1.2: 真實串接 Ollama at 192.168.0.188:11434
+    Phase 1.2: 真實串接設定中的 Ollama provider pool
     """
 
     async def generate_thinking_stream():

apps/api/src/services/ai_providers/ollama.py

@@ -1,10 +1,10 @@
 """
 Ollama Provider - Phase 24 ADR-052
 ====================================
-本地 LLM 推理 (192.168.0.188 VMware VM, CPU-only)
+本地 / 私有 LLM 推理 Provider。
 
 搬移自: openclaw.py _call_ollama (L349-409)
-特性: 免費、隱私安全 (local)、但 CPU 慢 (~97s/30tokens for qwen2.5:7b)
+特性: 免費、隱私安全 (local)、可依 ADR-110 指向 GCP-A/GCP-B/111。
 
 2026-04-02 ogt: Phase 24-A 從 openclaw.py 抽出
 """
@@ -268,33 +268,27 @@ class OllamaProvider:
             self._http_client = None
 
 
-# 2026-04-26 Wave5 B1-fix by Claude Engineer-A4 — OLLAMA_188 provider 註冊
-class Ollama188Provider(OllamaProvider):
+# 2026-05-06 Codex — 188 不再作為 Ollama Provider；本地備援統一命名為 ollama_local。
+class OllamaLocalProvider(OllamaProvider):
     """
-    Ollama 188 CPU-only 備援 Provider
+    Ollama Local fallback Provider
 
-    繼承 OllamaProvider，但使用 OLLAMA_FALLBACK_URL（192.168.0.188:11434）
-    作為推理端點，模型預設 OLLAMA_HEALTH_CHECK_MODEL（qwen2.5:7b-instruct）。
-
-    B1 修復：原本 _init_registry 未登錄此 provider，導致
-    executor.execute() 遇到 "ollama_188" → not_registered → 跳過，
-    188 從未被打到。此類別補全登錄鏈路。
-
-    2026-04-26 Wave5 B1-fix by Claude Engineer-A4
+    使用 OLLAMA_FALLBACK_URL 作為本地最後防線端點。
+    ADR-110 目前設定為 110 nginx proxy → 111 Ollama；188 不得再作為 Ollama provider。
     """
 
     @property
     def name(self) -> str:
-        return "ollama_188"
+        return "ollama_local"
 
     @property
     def is_enabled(self) -> bool:
         import os
-        # 優先查 ENABLE_OLLAMA_188；若未設定（預設 true）則看 OLLAMA_FALLBACK_URL 是否有值
-        env_override = os.getenv("ENABLE_OLLAMA_188", "true").lower() == "true"
+        # 優先查 ENABLE_OLLAMA_LOCAL；若未設定（預設 true）則看 OLLAMA_FALLBACK_URL 是否有值。
+        env_override = os.getenv("ENABLE_OLLAMA_LOCAL", "true").lower() == "true"
         if not env_override:
             return False
-        # OLLAMA_FALLBACK_URL 空字串 → 未設定 188 節點 → 停用
+        # OLLAMA_FALLBACK_URL 空字串 → 未設定本地節點 → 停用。
         return bool(getattr(settings, "OLLAMA_FALLBACK_URL", ""))
 
     def _endpoint_url(self) -> str:
@@ -319,18 +313,18 @@ class Ollama188Provider(OllamaProvider):
             client = await self._get_client()
 
             registry = get_model_registry()
-            # 嘗試取 ollama_188 專屬設定，fallback 到 ollama 預設
+            # 嘗試取本地 fallback 專屬設定，fallback 到 ollama 預設。
             try:
-                model_name = registry.get_model("ollama_188", "rca")
+                model_name = registry.get_model("ollama_local", "rca")
             except Exception:
                 model_name = getattr(settings, "OLLAMA_HEALTH_CHECK_MODEL", "qwen2.5:7b-instruct")
 
             try:
-                options = registry.get_provider_options("ollama_188")
+                options = registry.get_provider_options("ollama_local")
             except Exception:
                 options = registry.get_provider_options("ollama")
 
-            # CPU-only 備援：固定使用較長 timeout（CPU 推理慢）
+            # 本地備援：固定使用較長 timeout，避免 111 模型載入時被過早判死。
             task_type = (context or {}).get("task_type", "")
             if task_type in ("diagnose", "force_local"):
                 read_timeout = float(getattr(settings, "OLLAMA_DIAGNOSE_TIMEOUT_SECONDS", 200))
@@ -359,7 +353,7 @@ class Ollama188Provider(OllamaProvider):
             latency = (time.perf_counter() - start) * 1000
 
             logger.info(
-                "ollama_188_provider_success",
+                "ollama_local_provider_success",
                 response_length=len(result),
                 tokens=tokens,
                 latency_ms=round(latency, 1),
@@ -375,12 +369,12 @@ class Ollama188Provider(OllamaProvider):
 
         except httpx.TimeoutException as e:
             latency = (time.perf_counter() - start) * 1000
-            logger.warning("ollama_188_provider_timeout", error=str(e), latency_ms=round(latency, 1))
+            logger.warning("ollama_local_provider_timeout", error=str(e), latency_ms=round(latency, 1))
             return AIResult(raw_response="", success=False, provider=self.name, latency_ms=latency, error=f"Timeout: {e}")
 
         except Exception as e:
             latency = (time.perf_counter() - start) * 1000
-            logger.warning("ollama_188_provider_failed", error=str(e), latency_ms=round(latency, 1))
+            logger.warning("ollama_local_provider_failed", error=str(e), latency_ms=round(latency, 1))
             return AIResult(raw_response="", success=False, provider=self.name, latency_ms=latency, error=str(e))
 
     async def health_check(self) -> bool:

apps/api/src/services/ai_router.py

@@ -73,10 +73,6 @@ class AIProviderEnum(str, Enum):
     """AI 提供者"""
 
     OLLAMA = "ollama"
-    # 2026-04-25 critic-fix Part2 B2 by Claude Engineer-C2
-    # P1.1b OllamaFailoverManager 使用 provider_name="ollama_188"，
-    # 但 AIProviderEnum 沒有此值 → P1.2 整合時 lookup 失敗
-    OLLAMA_188 = "ollama_188"  # 188 CPU-only 備援節點（P1.1b）
     # 2026-05-04 ogt + Claude Sonnet 4.6: ADR-110 GCP 三層容災
     # OllamaFailoverManager 回傳 provider_name="ollama_gcp_a"/"ollama_gcp_b"/"ollama_local"
     # 缺少 enum 值 → AIProviderEnum(primary_str) 拋 ValueError → fallback chain 清空 → 直跳 Gemini
@@ -96,8 +92,6 @@ class AIProviderEnum(str, Enum):
 # Provider 對應延遲預算 (ms)
 PROVIDER_LATENCY_BUDGET: dict[AIProviderEnum, int] = {
     AIProviderEnum.OLLAMA: 60000,  # 本地，允許較長處理時間
-    # 2026-04-25 critic-fix Part2 B2 by Claude Engineer-C2 — 188 CPU-only 推理較慢
-    AIProviderEnum.OLLAMA_188: 120000,  # 120s budget for CPU inference
     # 2026-05-04 ogt: ADR-110 GCP 三層容災 — GCP NVMe SSD 推理快，60s 足夠
     AIProviderEnum.OLLAMA_GCP_A: 60000,
     AIProviderEnum.OLLAMA_GCP_B: 60000,
@@ -432,7 +426,7 @@ class AIRouter:
                     model = failover_result.primary.model
                     reason = f"{reason} [failover→{primary_str}]"
                 except ValueError:
-                    # provider_name 無法對應已知 enum（理論上不應發生，OLLAMA_188 已加）
+                    # provider_name 無法對應已知 enum；避免未知 provider 靜默進入執行層。
                     logger.warning(
                         "ai_router_unknown_failover_provider",
                         provider=primary_str,
@@ -1078,11 +1072,46 @@ class AIRouterExecutor:
             cached = await redis.get(cache_key)
             if cached:
                 data = _json.loads(cached)
+                cached_provider = data.get("provider", "cache")
+                provider_allowed = cached_provider in provider_order
+                ollama_first_required = (
+                    bool(context)
+                    and any(
+                        key in context
+                        for key in (
+                            "alert_type",
+                            "alertname",
+                            "alert_name",
+                            "fingerprint",
+                            "incident_id",
+                            "severity",
+                            "target_resource",
+                        )
+                    )
+                    and bool(provider_order)
+                    and provider_order[0].startswith("ollama")
+                )
+                if (
+                    cached_provider == "ollama"
+                    and any(provider.startswith("ollama") for provider in provider_order)
+                ):
+                    provider_allowed = True
+                if ollama_first_required and not cached_provider.startswith("ollama"):
+                    provider_allowed = False
+                if not provider_allowed:
+                    logger.info(
+                        "ai_router_cache_provider_mismatch_skip",
+                        cache_key=cache_key[:30],
+                        cached_provider=cached_provider,
+                        provider_order=provider_order,
+                        ollama_first_required=ollama_first_required,
+                    )
+                    raise ValueError("cached provider not allowed by current provider_order")
                 logger.info("ai_router_cache_hit", cache_key=cache_key[:30])
                 return AIResult(
                     raw_response=data.get("response", ""),
                     success=True,
-                    provider=data.get("provider", "cache"),
+                    provider=cached_provider,
                     from_cache=True,
                 )
         except Exception as e:
@@ -1306,7 +1335,7 @@ def _init_registry() -> AIProviderRegistry:
     """初始化 Provider Registry (首次呼叫時自動註冊所有 Provider)"""
     from src.services.ai_providers.ollama import (
         OllamaProvider,
-        Ollama188Provider,
+        OllamaLocalProvider,
         OllamaGcpBProvider,  # 2026-05-04 ADR-110 GCP-B
     )
     from src.services.ai_providers.gemini import GeminiProvider
@@ -1327,8 +1356,9 @@ def _init_registry() -> AIProviderRegistry:
     from src.services.ai_providers.nemotron import NemotronProvider
     registry.register(NemotronProvider())
 
-    # 2026-04-26 Wave5 B1-fix by Claude Engineer-A4 — 補登 OLLAMA_188 備援 provider
-    ollama_local = Ollama188Provider()
+    # 2026-05-06 Codex: 188 不再作為 Ollama provider。
+    # Local fallback 統一命名為 ollama_local，端點由 OLLAMA_FALLBACK_URL 指向 111/110 proxy。
+    ollama_local = OllamaLocalProvider()
     registry.register(ollama_local)
 
     # 2026-05-04 ogt + Claude Sonnet 4.6: ADR-110 GCP 三層容災修復
@@ -1337,7 +1367,7 @@ def _init_registry() -> AIProviderRegistry:
     # 修復：
     #   "ollama_gcp_a" alias → 同 OllamaProvider（OLLAMA_URL = GCP-A）
     #   "ollama_gcp_b" → 新 OllamaGcpBProvider（OLLAMA_SECONDARY_URL = GCP-B）
-    #   "ollama_local" alias → 同 Ollama188Provider（OLLAMA_FALLBACK_URL = 111）
+    #   "ollama_local" → OllamaLocalProvider（OLLAMA_FALLBACK_URL = 111 / 110:11437）
     registry._providers["ollama_gcp_a"] = ollama_gcp_a
     registry.register(OllamaGcpBProvider())
     registry._providers["ollama_local"] = ollama_local

apps/api/src/services/auto_repair_service.py

@@ -457,6 +457,8 @@ class AutoRepairService:
             except Exception as _db_e:
                 logger.error("auto_repair_db_write_failed", error=str(_db_e))
 
+            self._record_auto_repair_metric(playbook, success=True)
+
             # 2026-04-07 Claude Code: Sprint 4 B1/B2 — 記錄處置類型
             # P0-1 Fix: 統一使用 AnomalyCounter.hash_signature()
             try:
@@ -630,6 +632,8 @@ class AutoRepairService:
             except Exception as _db_e:
                 logger.error("auto_repair_db_write_failed", error=str(_db_e))
 
+            self._record_auto_repair_metric(playbook, success=False)
+
             # 2026-04-04 Claude Code: Phase 25 P1 — 失敗修復後 fire-and-forget 生成 ANTI_PATTERN
             # 2026-04-05 Claude Code: I1 修正 — 補齊 _pending_tasks GC 防護（對稱化）
             try:
@@ -700,6 +704,35 @@ class AutoRepairService:
 
         return max_risk
 
+    def _record_auto_repair_metric(self, playbook: Playbook, success: bool) -> None:
+        """把實際 auto-repair 執行寫入 Prometheus 指標。
+
+        2026-05-06 ogt + Codex：DB 已有 auto_repair_executions，但
+        core.metrics.record_auto_repair() 長期零 caller，導致治理/心跳用
+        Prometheus 看起來像「飛輪沒做事」。label 使用 action_type，避免
+        playbook_id 造成高基數。
+        """
+        try:
+            from src.core.metrics import record_auto_repair
+
+            first_step = playbook.repair_steps[0] if playbook.repair_steps else None
+            action = first_step.action_type.value if first_step else "unknown"
+            max_risk = self._get_max_risk_level(playbook)
+            tier = {
+                RiskLevel.LOW: 1,
+                RiskLevel.MEDIUM: 2,
+                RiskLevel.HIGH: 3,
+                RiskLevel.CRITICAL: 4,
+            }.get(max_risk, 0)
+            record_auto_repair(action=action, tier=tier, success=success)
+        except Exception as e:
+            logger.warning(
+                "auto_repair_metric_record_failed",
+                playbook_id=playbook.playbook_id,
+                success=success,
+                error=str(e),
+            )
+
     def _is_host_or_backup_incident(self, incident: Incident) -> bool:
         """主機/備份類事件只能走 SSH/只讀診斷，不允許 K8s rollout 類修復。"""
 

apps/api/src/services/decision_manager.py

@@ -607,7 +607,7 @@ async def _nemoclaw_second_opinion(incident: "Incident", primary_result: dict) -
     """
     MCP Phase 4a: NemoClaw second opinion — 信心 < 0.7 時觸發
     ============================================================
-    用 deepseek-r1:14b (Ollama 188) 對同一份資料做獨立推理，
+    用 deepseek-r1:14b (設定的 Ollama primary) 對同一份資料做獨立推理，
     輸出純文字 advisory_note，不執行任何操作。
 
     2026-04-11 Claude Sonnet 4.6 Asia/Taipei
@@ -666,7 +666,7 @@ async def _generate_playbook_draft_if_new(incident: "Incident") -> None:
     MCP Phase 4c: Playbook 無命中時，自動生成 AI 草稿 Playbook 寫入 KM
     =====================================================================
     - 僅在 KM 中不存在同 alertname 的 Playbook 時觸發（避免重複）
-    - 用 qwen2.5:7b-instruct (Ollama 188) 生成結構化 Playbook 草稿
+    - 用 qwen2.5:7b-instruct (設定的 Ollama primary) 生成結構化 Playbook 草稿
     - 寫入 KnowledgeEntry，status=DRAFT，需人工審核後升為 APPROVED
     - 寫入 AlertOperationLog PLAYBOOK_DRAFT_CREATED 事件
 

apps/api/src/services/flywheel_stats_service.py

@@ -237,6 +237,31 @@ class FlywheelStatsService:
                 except (json.JSONDecodeError, KeyError):
                     continue
 
+            # 2026-05-06 ogt + Codex:
+            # 執行成功率的 source of truth 是 auto_repair_executions。
+            # Redis playbook success_count/failure_count 會因回寫鏈路中斷而落後，
+            # 造成 governance / heartbeat 判定「飛輪沒有執行」。
+            try:
+                async with get_db_context() as db:
+                    row = await db.execute(
+                        text("""
+                            SELECT
+                                COUNT(*) FILTER (WHERE success IS TRUE) AS success,
+                                COUNT(*) AS total
+                            FROM auto_repair_executions
+                            WHERE created_at >= NOW() - interval '24 hours'
+                        """)
+                    )
+                    repair_stats = row.one()
+                    db_total_exec = int(repair_stats.total or 0)
+                    if db_total_exec >= FLYWHEEL_MIN_SAMPLE:
+                        db_total_success = int(repair_stats.success or 0)
+                        return count, db_total_success / db_total_exec
+                    if db_total_exec > 0:
+                        return count, None
+            except Exception:
+                logger.warning("flywheel_stats_auto_repair_execution_query_failed")
+
             if total_exec < FLYWHEEL_MIN_SAMPLE:
                 # 樣本不足（含 Redis 空），回 None 通知呼叫方跳過 W-3 告警判斷
                 return count, None

apps/api/src/services/heartbeat_report_service.py

@@ -15,7 +15,7 @@ HeartbeatReportService — ADR-073 心跳監控重構
 import asyncio
 import html
 from dataclasses import dataclass, field
-from datetime import datetime, timedelta
+from datetime import datetime
 from typing import Optional
 
 import httpx
@@ -420,8 +420,8 @@ class HeartbeatReportService:
         try:
             # KM 向量化率（DB 查詢）
             from src.db.base import get_db_context
-            from src.db.models import IncidentRecord, KnowledgeEntryRecord
-            from sqlalchemy import func, select
+            from src.db.models import KnowledgeEntryRecord
+            from sqlalchemy import func, select, text as sa_text
             async with get_db_context() as db:
                 # KM 總數
                 km_total = await db.scalar(select(func.count()).select_from(KnowledgeEntryRecord))
@@ -436,20 +436,22 @@ class HeartbeatReportService:
                 stats.km_vectorized = vec_result.scalar() or 0
 
                 # 24h 修復統計
-                since = datetime.utcnow() - timedelta(hours=24)
-                outcomes = await db.execute(
-                    select(IncidentRecord.outcome).where(
-                        IncidentRecord.created_at >= since,
-                        IncidentRecord.outcome.isnot(None),
-                    )
-                )
-                outcome_list = [r[0] for r in outcomes.all() if r[0]]
-                stats.attempt_24h = len(outcome_list)
-                stats.success_24h = sum(
-                    1 for o in outcome_list
-                    if isinstance(o, dict) and o.get("execution_success")
-                    or isinstance(o, str) and "success" in o.lower()
+                # 2026-05-06 ogt + Codex:
+                # incidents.outcome 已不是自動修復 source of truth。實際執行紀錄
+                # 寫在 auto_repair_executions；舊查詢會讓心跳報告顯示 0/15，
+                # 造成「全系統正常」但飛輪 KPI 失真的假象。
+                repair_result = await db.execute(
+                    sa_text("""
+                        SELECT
+                            COUNT(*) FILTER (WHERE success IS TRUE) AS success,
+                            COUNT(*) AS total
+                        FROM auto_repair_executions
+                        WHERE created_at >= NOW() - interval '24 hours'
+                    """)
                 )
+                repair_row = repair_result.one()
+                stats.success_24h = int(repair_row.success or 0)
+                stats.attempt_24h = int(repair_row.total or 0)
 
                 # 最後學習活動
                 last_km = await db.scalar(
@@ -865,9 +867,10 @@ def report_to_telegram_html(report: HeartbeatReport) -> str:
         lines.append("☸️ <b>Kubernetes Pods</b>")
         for i, pod in enumerate(report.pods):
             prefix = "└─" if i == len(report.pods) - 1 else "├─"
-            ready_icon = "✅" if pod.ready else "❌"
+            ready_icon = "✅" if pod.ready or pod.status in ("Succeeded", "Completed") else "❌"
             restart_str = f" (重啟×{pod.restarts})" if pod.restarts > 0 else ""
-            lines.append(f"{prefix} {ready_icon} {html.escape(pod.name[:35])}{restart_str}")
+            status_str = "" if pod.ready else f" <code>{html.escape(pod.status)}</code>"
+            lines.append(f"{prefix} {ready_icon} {html.escape(pod.name[:35])}{restart_str}{status_str}")
 
     # --- Scanner 狀態 ---
     if report.scanners.last_runs:

apps/api/src/services/host_aggregator.py

@@ -7,7 +7,7 @@ Hosts:
 - 192.168.0.110: DevOps 金庫 (Harbor, GH Runner)
 - 192.168.0.112: Kali Security (Scanner API)
 - 192.168.0.120: K3s Master (awoooi-prod namespace)
-- 192.168.0.188: AI+Web 中心 (Nginx, PostgreSQL, Redis, Ollama, OpenClaw, SigNoz)
+- 192.168.0.188: AI+Web 中心 (Nginx, PostgreSQL, Redis, OpenClaw, SigNoz)
 
 Features:
 - asyncio.gather for parallel fetching

apps/api/src/services/k3s_monitor_service.py

@@ -34,8 +34,12 @@ logger = structlog.get_logger(__name__)
 # 台北時區
 TZ_TAIPEI = ZoneInfo("Asia/Taipei")
 
-# Prometheus 端點
-PROMETHEUS_URL = "http://192.168.0.121:30090"
+# Prometheus endpoint.
+#
+# 2026-05-05 Codex: do not pin this report job to a K3s worker NodePort.
+# Production already injects PROMETHEUS_URL from ConfigMap, currently the
+# Docker Prometheus on 110. This keeps reboot recovery independent of 121.
+PROMETHEUS_URL = settings.PROMETHEUS_URL.rstrip("/")
 
 # kube-state-metrics 查詢
 PROM_QUERIES = {
@@ -215,7 +219,7 @@ class K3sMonitorService:
 
             # 發送訊息
             formatted = status.format()
-            result = await gateway.send_message(formatted)
+            result = await gateway.send_text(formatted)
 
             if result:
                 logger.info("k3s_daily_report_sent", date=status.report_date)

apps/api/src/services/model_version_probe.py

@@ -5,7 +5,7 @@ AI Provider 版本探測 — 為每個 Provider 提供 get_version()
 
 Provider:
   - ollama        : 34.143.170.20 GCP-A Ollama (primary) — 2026-05-03 ogt: ADR-110 GCP-A Primary
-  - ollama_188    : 192.168.0.188 Ollama (fallback)
+  - ollama_local  : 192.168.0.111 / 110 proxy Ollama (local fallback)
   - gemini        : Google Gemini API (版本 = model name)
   - claude        : Anthropic Claude (版本 = model name)
   - openclaw_nemo : OpenClaw NemoTron (版本 = OPENCLAW_DEFAULT_MODEL)
@@ -31,7 +31,7 @@ TAIPEI_TZ = timezone(timedelta(hours=8))
 class ProviderVersionInfo:
     """AI Provider 版本快照"""
 
-    provider: str  # "ollama" / "ollama_188" / "gemini" / "claude" / "openclaw_nemo"
+    provider: str  # "ollama" / "ollama_local" / "gemini" / "claude" / "openclaw_nemo"
     model: str
     version: str  # version string 或 tag（Ollama 用 modified_at，其他用 model name）
     digest: str | None = None  # SHA256 digest（僅 Ollama 有）
@@ -43,7 +43,7 @@ class ProviderVersionInfo:
 # =============================================================================
 
 async def probe_ollama_version(url: str, model: str) -> ProviderVersionInfo:
-    """探測 Ollama（GCP-A 或 188）：GET /api/tags 取 model digest + modified_at
+    """探測 Ollama（GCP-A/GCP-B 或本地 111）：GET /api/tags 取 model digest + modified_at
 
     Args:
         url:   Ollama base URL，例如 "http://34.143.170.20:11434"（GCP-A Primary）
@@ -58,15 +58,12 @@ async def probe_ollama_version(url: str, model: str) -> ProviderVersionInfo:
     """
     import httpx
 
-    # 2026-05-03 ogt: ADR-110 GCP-A Primary — 擴展 provider 判斷邏輯支援 GCP 三層容災
-    # 188 保留 ollama_188 命名（CPU-only 主機，雖移出 routing chain 但仍可被 probe）
+    # 2026-05-06 Codex: 188 不再作為 Ollama provider；local fallback 一律標示 ollama_local。
     _GCP_OLLAMA_IPS = {"34.143.170.20", "34.21.145.224"}
     if any(ip in url for ip in _GCP_OLLAMA_IPS):
         provider_name = "ollama"
-    elif "192.168.0.111" in url:
+    elif "192.168.0.111" in url or "192.168.0.110:11437" in url:
         provider_name = "ollama_local"
-    elif "192.168.0.188" in url:
-        provider_name = "ollama_188"
     else:
         provider_name = "ollama_remote"
 
@@ -179,7 +176,7 @@ async def probe_claude_version() -> ProviderVersionInfo:
 async def probe_openclaw_nemo_version() -> ProviderVersionInfo:
     """OpenClaw NemoTron：版本字串從 settings.OPENCLAW_DEFAULT_MODEL 讀取
 
-    NemoTron 運行在 OpenClaw 188 節點（使用 Ollama 推理），
+    NemoTron 運行在 OpenClaw 節點，
     透過 OPENCLAW_URL /api/tags 探測，模型名稱即版本識別。
 
     Returns:
@@ -195,18 +192,18 @@ async def probe_openclaw_nemo_version() -> ProviderVersionInfo:
 
     # OpenClaw 底層是 Ollama，使用 OPENCLAW_URL 的 host:port 加上 Ollama port
     # OPENCLAW_URL 是 8088（OpenClaw API），Ollama 通常在 11434
-    # 188 的 Ollama URL 若有設定則直接用 OLLAMA_FALLBACK_URL
-    ollama_188_url = settings.OLLAMA_FALLBACK_URL
-    if not ollama_188_url:
+    # OpenClaw 底層 tags 來源優先使用本地 fallback Ollama URL。
+    ollama_local_url = settings.OLLAMA_FALLBACK_URL
+    if not ollama_local_url:
         # fallback：從 OPENCLAW_URL host 構建 Ollama URL
         from urllib.parse import urlparse
         parsed = urlparse(settings.OPENCLAW_URL)
-        ollama_188_url = f"{parsed.scheme}://{parsed.hostname}:11434"
+        ollama_local_url = f"{parsed.scheme}://{parsed.hostname}:11434"
 
     import httpx
 
     async with httpx.AsyncClient(timeout=5.0) as client:
-        resp = await client.get(f"{ollama_188_url}/api/tags")
+        resp = await client.get(f"{ollama_local_url}/api/tags")
         resp.raise_for_status()
         models = resp.json().get("models", [])
 
@@ -220,7 +217,7 @@ async def probe_openclaw_nemo_version() -> ProviderVersionInfo:
             )
 
     # model 不在清單時：version 用 model name，digest=None
-    logger.warning("openclaw_nemo_model_not_in_tags", model=model, url=ollama_188_url)
+    logger.warning("openclaw_nemo_model_not_in_tags", model=model, url=ollama_local_url)
     return ProviderVersionInfo(
         provider="openclaw_nemo",
         model=model,
@@ -257,7 +254,7 @@ async def probe_all_providers() -> list[ProviderVersionInfo]:
     raw = await asyncio.gather(*tasks, return_exceptions=True)
 
     results: list[ProviderVersionInfo] = []
-    provider_labels = ["ollama", "ollama_188", "gemini", "claude", "openclaw_nemo"]
+    provider_labels = ["ollama", "ollama_local", "gemini", "claude", "openclaw_nemo"]
     for label, outcome in zip(provider_labels, raw, strict=True):
         if isinstance(outcome, ProviderVersionInfo):
             results.append(outcome)

apps/api/src/services/openclaw.py

@@ -5,7 +5,7 @@ Phase 5: OpenClaw 實體化升級 (2026-03-21)
 統帥校正: SignOz 為唯一全能視力中心
 
 Features:
-- 真實 LLM SDK 整合 (Ollama → Gemini → Claude)
+- 真實 LLM SDK 整合 (告警預設 Ollama GCP-A → GCP-B → 111 → Gemini)
 - SignOz Gold Metrics 即時擷取 (P99/Error/RPS)
 - AIOps Agent 專業人格 (K8s 維運 + SRE RCA 專精)
 - 強制結構化 JSON 輸出 (符合 API 契約)
@@ -144,8 +144,8 @@ class OpenClawService:
     """
     OpenClaw AI 決策服務 - True LLM + SignOz Integration
 
-    實作 AI_FALLBACK_ORDER 備援機制:
-    Ollama → Gemini → Claude → Mock
+    實作 AI_FALLBACK_ORDER 備援機制。
+    告警/incident 上下文預設套用成本防線，只允許 Ollama GCP-A → GCP-B → 111。
 
     新增 SignOz 整合:
     - 自動擷取 Gold Metrics
@@ -176,6 +176,89 @@ class OpenClawService:
             await self._http_client.aclose()
             self._http_client = None
 
+    def _is_incident_alert_context(self, alert_context: dict | None) -> bool:
+        """Return true when a request came from the alert/incident automation path."""
+        if not alert_context:
+            return False
+        alert_keys = {
+            "alert_type",
+            "alertname",
+            "alert_name",
+            "fingerprint",
+            "incident_id",
+            "severity",
+            "signals",
+            "target_resource",
+        }
+        return any(key in alert_context for key in alert_keys)
+
+    def _cloud_fallback_allowed_for_alert(self, alert_context: dict | None) -> bool:
+        """Cloud fallback is allowed after the ordered Ollama lane for alerts."""
+        if not self._is_incident_alert_context(alert_context):
+            return True
+        return bool(getattr(settings, "ALERT_AI_ALLOW_CLOUD_FALLBACK", True))
+
+    def _alert_enforces_ollama_first(self, alert_context: dict | None) -> bool:
+        """Alert cards must try GCP-A/GCP-B/111 before Gemini backup."""
+        return (
+            self._is_incident_alert_context(alert_context)
+            and bool(getattr(settings, "ALERT_AI_ENFORCE_OLLAMA_FIRST", True))
+        )
+
+    async def _resolve_alert_provider_order(
+        self,
+        task_type: str = "diagnose",
+        alert_context: dict | None = None,
+        cloud_provider_order: list[str] | None = None,
+    ) -> list[str]:
+        """Resolve GCP-A/GCP-B/111, then Gemini backup, for alert analysis."""
+        provider_order: list[str] = []
+        try:
+            route = await get_ollama_failover_manager().select_provider(task_type=task_type)
+            provider_order = [
+                endpoint.provider_name
+                for endpoint in route.all_endpoints_in_order()
+                if endpoint.provider_name.startswith("ollama")
+            ]
+        except Exception as route_error:
+            logger.warning(
+                "alert_ollama_route_lookup_failed",
+                error=str(route_error),
+                task_type=task_type,
+            )
+
+        if not provider_order:
+            provider_order = ["ollama_gcp_a", "ollama_gcp_b", "ollama_local"]
+
+        deduped: list[str] = []
+        for provider_name in provider_order:
+            if provider_name and provider_name not in deduped:
+                deduped.append(provider_name)
+
+        if not self._alert_enforces_ollama_first(alert_context):
+            return deduped
+
+        ollama_order = {"ollama_gcp_a": 0, "ollama_gcp_b": 1, "ollama_local": 2}
+        ordered_ollama = [
+            provider_name
+            for provider_name in deduped
+            if provider_name in ollama_order
+        ]
+        ordered_ollama.sort(key=lambda provider_name: ollama_order[provider_name])
+        if not ordered_ollama:
+            ordered_ollama = ["ollama_gcp_a", "ollama_gcp_b", "ollama_local"]
+
+        if not self._cloud_fallback_allowed_for_alert(alert_context):
+            return ordered_ollama
+
+        cloud_candidates = cloud_provider_order or []
+        cloud_backup: list[str] = []
+        for provider_name in [*cloud_candidates, "gemini"]:
+            if provider_name == "gemini" and provider_name not in cloud_backup:
+                cloud_backup.append(provider_name)
+
+        return ordered_ollama + cloud_backup
+
     # =========================================================================
     # SignOz Integration
     # =========================================================================
@@ -437,13 +520,13 @@ class OpenClawService:
     # 完整移除時機: Phase 24 完整驗收後 (ADR-052 D11)
     # =========================================================================
 
-    async def _call_ollama(self, prompt: str) -> tuple[str, bool]:
+    async def _call_ollama(self, prompt: str, *, ollama_only: bool = False) -> tuple[str, bool]:
         """
         呼叫 Ollama (支援 JSON Mode)。
 
         USE_AI_ROUTER=true 正常會走 AIRouterExecutor；這裡是 legacy safety-net。
         2026-05-05 Codex: safety-net 也必須遵守 ADR-110 三層 Ollama
-        路由，不能只打 OLLAMA_URL 後直接掉 Gemini。
+        路由，告警路徑預設只允許 GCP-A/GCP-B/111，不能只打 OLLAMA_URL 後直接掉 Gemini。
         """
         try:
             client = await self._get_client()
@@ -484,6 +567,26 @@ class OpenClawService:
                         endpoints.append((provider_name, endpoint_url))
                         seen_urls.add(endpoint_url)
 
+            if ollama_only:
+                allowed_provider_order = {"ollama_gcp_a": 0, "ollama_gcp_b": 1, "ollama_local": 2}
+                endpoints = [
+                    (provider_name, endpoint_url)
+                    for provider_name, endpoint_url in endpoints
+                    if provider_name in allowed_provider_order
+                ]
+                endpoints.sort(key=lambda item: allowed_provider_order[item[0]])
+                if not endpoints:
+                    endpoints = [
+                        ("ollama_gcp_a", settings.OLLAMA_URL),
+                        ("ollama_gcp_b", getattr(settings, "OLLAMA_SECONDARY_URL", "")),
+                        ("ollama_local", getattr(settings, "OLLAMA_FALLBACK_URL", "")),
+                    ]
+                    endpoints = [
+                        (provider_name, endpoint_url)
+                        for provider_name, endpoint_url in endpoints
+                        if endpoint_url
+                    ]
+
             last_error = ""
             for provider_name, endpoint_url in endpoints:
                 try:
@@ -973,7 +1076,11 @@ class OpenClawService:
             try:
                 # 2026-04-02 ogt: C2 修復 — 呼叫 AIRouter.route() 智慧路由 (非靜態 order)
                 # D1 意圖分類路由、D7 隱私保護 (DIAGNOSE/CODE_REVIEW 強制 local) 生效
-                from src.services.ai_router import get_ai_router, get_ai_executor, IntentType
+                from src.services.ai_router import (
+                    IntentType,
+                    get_ai_executor,
+                    get_ai_router,
+                )
                 router = get_ai_router()
                 executor = get_ai_executor()
 
@@ -987,7 +1094,10 @@ class OpenClawService:
                     if p.value != decision.selected_provider.value
                 ]
                 try:
-                    from src.services.ai_control import get_primary_provider, is_provider_disabled
+                    from src.services.ai_control import (
+                        get_primary_provider,
+                        is_provider_disabled,
+                    )
                     _primary = await get_primary_provider()
                     if _primary and _primary != decision.selected_provider.value:
                         # 把 primary 移到首位 (保留原始 fallback)
@@ -1003,6 +1113,20 @@ class OpenClawService:
                 except Exception as _e:
                     logger.warning("ai_control_override_failed", error=str(_e))
 
+                if self._alert_enforces_ollama_first(alert_context):
+                    original_provider_order = list(provider_order)
+                    provider_order = await self._resolve_alert_provider_order(
+                        task_type=decision.intent.value if decision.intent else "diagnose",
+                        alert_context=alert_context,
+                        cloud_provider_order=original_provider_order,
+                    )
+                    logger.info(
+                        "alert_ollama_first_provider_order",
+                        original_provider_order=original_provider_order,
+                        provider_order=provider_order,
+                        cloud_fallback_allowed=self._cloud_fallback_allowed_for_alert(alert_context),
+                    )
+
                 # Step 3: D7 隱私 — CODE_REVIEW 強制 local
                 # 2026-04-15 ogt: DIAGNOSE 移除 require_local（v4.3 決策：NIM 為主力，無隱私問題）
                 # ai_router.py v4.3 已明確：「NIM 從 Phase 22 起就是主力，無隱私問題」
@@ -1045,13 +1169,18 @@ class OpenClawService:
             _mock_json, _rule_id = self._generate_mock_response(alert_context or {}, signoz_metrics)
             if _rule_id == "generic_fallback":
                 import asyncio
+
                 from src.services.alert_rule_engine import auto_generate_rule
                 try:
                     asyncio.create_task(auto_generate_rule(
                         alert_context or {},
                         ollama_url=settings.OLLAMA_URL,
                         model=settings.OPENCLAW_DEFAULT_MODEL,
-                        gemini_api_key=getattr(settings, "GEMINI_API_KEY", ""),
+                        gemini_api_key=(
+                            getattr(settings, "GEMINI_API_KEY", "")
+                            if self._cloud_fallback_allowed_for_alert(alert_context)
+                            else ""
+                        ),
                     ))
                 except Exception as _e:
                     logger.warning("auto_rule_trigger_failed", error=str(_e))
@@ -1086,7 +1215,18 @@ class OpenClawService:
             from src.services.ai_rate_limiter import get_ai_rate_limiter
             rate_limiter = get_ai_rate_limiter()
 
-            for provider in settings.AI_FALLBACK_ORDER:
+            legacy_provider_order = list(settings.AI_FALLBACK_ORDER)
+            if self._alert_enforces_ollama_first(alert_context):
+                legacy_provider_order = ["ollama"]
+                if self._cloud_fallback_allowed_for_alert(alert_context):
+                    legacy_provider_order.append("gemini")
+                logger.info(
+                    "legacy_alert_ollama_first_provider_order",
+                    provider_order=legacy_provider_order,
+                    cloud_fallback_allowed=self._cloud_fallback_allowed_for_alert(alert_context),
+                )
+
+            for provider in legacy_provider_order:
                 # Rate Limit 檢查 (nvidia/gemini/claude 需檢查，ollama 不限)
                 # 2026-03-30 ogt: 加入 nvidia (RPM=5 限制)
                 if provider in ("nvidia", "gemini", "claude"):
@@ -1109,7 +1249,10 @@ class OpenClawService:
                 cost_usd = 0.0
 
                 if provider == "ollama":
-                    response, success = await self._call_ollama(prompt)
+                    response, success = await self._call_ollama(
+                        prompt,
+                        ollama_only=self._alert_enforces_ollama_first(alert_context),
+                    )
                 elif provider == "gemini":
                     response, success, total_tokens, cost_usd = await self._call_gemini(prompt)
                 elif provider == "nvidia":
@@ -1165,13 +1308,18 @@ class OpenClawService:
             _mock_json, _rule_id = self._generate_mock_response(alert_context or {}, signoz_metrics)
             if _rule_id == "generic_fallback":
                 import asyncio
+
                 from src.services.alert_rule_engine import auto_generate_rule
                 try:
                     asyncio.create_task(auto_generate_rule(
                         alert_context or {},
                         ollama_url=settings.OLLAMA_URL,
                         model=settings.OPENCLAW_DEFAULT_MODEL,
-                        gemini_api_key=getattr(settings, "GEMINI_API_KEY", ""),
+                        gemini_api_key=(
+                            getattr(settings, "GEMINI_API_KEY", "")
+                            if self._cloud_fallback_allowed_for_alert(alert_context)
+                            else ""
+                        ),
                     ))
                 except Exception as _e:
                     logger.warning("auto_rule_trigger_failed", error=str(_e))
@@ -1218,14 +1366,14 @@ class OpenClawService:
                 except json.JSONDecodeError:
                     # 3. 啟發式修補: 如果結尾缺少括號，嘗試補齊
                     if candidate.startswith("{") and not candidate.endswith("}"):
-                        for i in range(1, 5): # 嘗試補 1-5 個括號/引號
-                             try:
-                                 repaired = candidate + '"' * (i-1) + "}" * i
-                                 json.loads(repaired)
-                                 logger.info("json_repaired_heuristically", level=i)
-                                 return repaired
-                             except:
-                                 continue
+                        for i in range(1, 5):  # 嘗試補 1-5 個括號/引號
+                            try:
+                                repaired = candidate + '"' * (i - 1) + "}" * i
+                                json.loads(repaired)
+                                logger.info("json_repaired_heuristically", level=i)
+                                return repaired
+                            except json.JSONDecodeError:
+                                continue
                     continue
 
         # 4. 極端情況: 找出最後一個有效 key
@@ -1235,11 +1383,11 @@ class OpenClawService:
             # 暴力去除非法尾綴 (如 \t\t...)
             candidate = re.sub(r"[ \t\r\n]+$", "", candidate)
             if not candidate.endswith("}"):
-                candidate += '"}' # 嘗試最簡單的閉合
+                candidate += '"}'  # 嘗試最簡單的閉合
             try:
                 json.loads(candidate)
                 return candidate
-            except:
+            except json.JSONDecodeError:
                 pass
 
         return None
@@ -1791,7 +1939,7 @@ Focus on:
             from src.services.ai_router import get_ai_registry
 
             ai_registry = get_ai_registry()
-            provider = ai_registry.get("ollama") or ai_registry.get("ollama_188")
+            provider = ai_registry.get("ollama") or ai_registry.get("ollama_local")
             if provider is None or not hasattr(provider, "analyze_with_tools"):
                 logger.warning(
                     "openclaw_agent_loop_shadow_skipped",
@@ -2200,6 +2348,7 @@ Expert context: {json.dumps(expert_context or {}, ensure_ascii=False, default=st
             }
         """
         import asyncio
+
         from src.services.nvidia_provider import get_nvidia_provider
 
         nvidia = get_nvidia_provider()
@@ -2334,7 +2483,7 @@ Expert context: {json.dumps(expert_context or {}, ensure_ascii=False, default=st
                 "latency_ms": latency_ms,
             }
 
-        except asyncio.TimeoutError:
+        except TimeoutError:
             latency_ms = (time.time() - start_time) * 1000
             logger.error(
                 "nemotron_tool_call_timeout",
@@ -2528,6 +2677,7 @@ async def _fetch_k8s_inventory_for_openclaw(
         "awoooi-api, awoooi-web, ..." 格式字串，失敗時返回 ""
     """
     import asyncio as _asyncio
+
     import structlog as _structlog
     _logger = _structlog.get_logger(__name__)
     try:
@@ -2542,7 +2692,7 @@ async def _fetch_k8s_inventory_for_openclaw(
         )
         try:
             stdout, _ = await _asyncio.wait_for(proc.communicate(), timeout=timeout_sec)
-        except _asyncio.TimeoutError:
+        except TimeoutError:
             proc.kill()
             _logger.warning("k8s_inventory_timeout_openclaw", namespace=namespace)
             return ""

apps/api/src/services/platform_operator_service.py

@@ -9,8 +9,6 @@ ADR-106（AwoooP Agent Platform）
 from __future__ import annotations
 
 import uuid
-from datetime import datetime
-from decimal import Decimal
 from typing import Any
 from uuid import UUID
 
@@ -153,8 +151,21 @@ async def list_runs(
 # Approvals
 # =============================================================================
 
-async def list_approvals(project_id: str | None) -> dict[str, Any]:
-    """列出所有 waiting_approval 狀態的 runs。"""
+async def list_approvals(
+    project_id: str | None,
+    run_id: str | None = None,
+) -> dict[str, Any]:
+    """列出 waiting_approval runs，可依 project_id 或 run_id 篩選。"""
+    run_uuid: UUID | None = None
+    if run_id:
+        try:
+            run_uuid = uuid.UUID(run_id)
+        except ValueError as exc:
+            raise HTTPException(
+                status_code=status.HTTP_422_UNPROCESSABLE_ENTITY,
+                detail=f"run_id 格式錯誤: {exc}",
+            ) from exc
+
     async with get_db_context("awoooi") as db:
         stmt = (
             select(AwoooPRunState)
@@ -163,6 +174,8 @@ async def list_approvals(project_id: str | None) -> dict[str, Any]:
         )
         if project_id is not None:
             stmt = stmt.where(AwoooPRunState.project_id == project_id)
+        if run_uuid is not None:
+            stmt = stmt.where(AwoooPRunState.run_id == run_uuid)
 
         count_stmt = select(func.count()).select_from(stmt.subquery())
         total_result = await db.execute(count_stmt)

apps/api/src/services/playbook_generator.py

@@ -4,7 +4,7 @@ LLM Playbook Generator - ADR-104 T1/T2/T6
 從成功修復案例生成可治理的 Playbook 草稿。
 
 設計重點:
-- 只用 local provider 順序（Ollama 111 -> Ollama 188），避免新增雲端成本。
+- 只用 local/provider pool 順序（GCP-A -> 111 local），避免新增雲端成本。
 - LLM 產出必須經 Pydantic + action_parser 安全收斂。
 - 不直接 APPROVED；先 DRAFT/REVIEW，再交治理 job 晉級。
 """
@@ -30,7 +30,6 @@ from src.models.playbook import (
     RiskLevel,
     SymptomPattern,
 )
-from src.services.action_parser import is_safe_kubectl_action
 from src.services.action_parser import kubectl_safety_reason
 
 logger = structlog.get_logger(__name__)
@@ -218,7 +217,7 @@ class LLMPlaybookGenerator:
             executor = get_ai_executor()
             result = await executor.execute(
                 prompt=prompt,
-                provider_order=["ollama", "ollama_188"],
+                provider_order=["ollama", "ollama_local"],
                 context=context,
                 cache_ttl=86400,
                 require_local=True,

apps/api/src/services/weekly_report_service.py

@@ -244,7 +244,7 @@ class WeeklyReportService:
 
             # 發送訊息
             formatted = report.format()
-            result = await gateway.send_message(formatted)
+            result = await gateway.send_text(formatted)
 
             if result:
                 logger.info("weekly_report_sent", week=report.week_range)

apps/api/tests/test_ai_router_cache_provider_policy.py

@@ -0,0 +1,90 @@
+from __future__ import annotations
+
+import json
+from typing import Any
+
+import pytest
+
+from src.services import ai_router as ai_router_module
+from src.services.ai_providers.interfaces import AIResult
+from src.services.ai_router import AIProviderRegistry, AIRouterExecutor
+
+
+class _FakeRedis:
+    def __init__(self, cached_provider: str) -> None:
+        self.cached_provider = cached_provider
+        self.set_calls: list[tuple[str, str, int | None]] = []
+
+    async def get(self, key: str) -> str:
+        return json.dumps({
+            "response": '{"provider":"stale"}',
+            "provider": self.cached_provider,
+        })
+
+    async def set(self, key: str, value: str, ex: int | None = None) -> None:
+        self.set_calls.append((key, value, ex))
+
+
+class _FakeProvider:
+    name = "ollama_gcp_a"
+    privacy_level = "local"
+    is_enabled = True
+    capabilities = {"rca", "chat"}
+
+    def __init__(self) -> None:
+        self.calls = 0
+
+    async def analyze(self, prompt: str, context: dict[str, Any] | None = None) -> AIResult:
+        self.calls += 1
+        return AIResult(
+            raw_response='{"provider":"fresh_ollama"}',
+            success=True,
+            provider=self.name,
+        )
+
+
+@pytest.mark.asyncio
+async def test_executor_skips_cached_cloud_provider_when_ollama_lane_is_required(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    fake_redis = _FakeRedis(cached_provider="gemini")
+    fake_provider = _FakeProvider()
+    registry = AIProviderRegistry()
+    registry.register(fake_provider)
+
+    monkeypatch.setattr(ai_router_module._settings, "MOCK_MODE", False)
+    monkeypatch.setattr("src.core.redis_client.get_redis", lambda: fake_redis)
+
+    result = await AIRouterExecutor(registry).execute(
+        prompt="diagnose alert",
+        provider_order=["ollama_gcp_a", "ollama_gcp_b", "ollama_local", "gemini"],
+        context={"intent_hint": "diagnose", "alert_type": "HostHighCpuLoad"},
+    )
+
+    assert result.provider == "ollama_gcp_a"
+    assert result.raw_response == '{"provider":"fresh_ollama"}'
+    assert fake_provider.calls == 1
+    assert fake_redis.set_calls
+
+
+@pytest.mark.asyncio
+async def test_executor_allows_cached_ollama_provider_for_ollama_lane(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    fake_redis = _FakeRedis(cached_provider="ollama")
+    fake_provider = _FakeProvider()
+    registry = AIProviderRegistry()
+    registry.register(fake_provider)
+
+    monkeypatch.setattr(ai_router_module._settings, "MOCK_MODE", False)
+    monkeypatch.setattr("src.core.redis_client.get_redis", lambda: fake_redis)
+
+    result = await AIRouterExecutor(registry).execute(
+        prompt="diagnose alert",
+        provider_order=["ollama_gcp_a", "ollama_gcp_b", "ollama_local"],
+        context={"intent_hint": "diagnose", "alert_type": "HostHighCpuLoad"},
+    )
+
+    assert result.provider == "ollama"
+    assert result.from_cache is True
+    assert fake_provider.calls == 0

apps/api/tests/test_ai_router_diagnose_fallback.py

@@ -124,8 +124,9 @@ def test_diagnose_fallback_chain_ollama_primary():
     assert AIProviderEnum.OPENCLAW_NEMO in providers_in_chain
     assert AIProviderEnum.GEMINI in providers_in_chain
     assert AIProviderEnum.CLAUDE in providers_in_chain
-    # OLLAMA_188 (CPU-only 備援) 仍排除（M1 Pro 111 才是 GPU 主推理）
-    assert AIProviderEnum.OLLAMA_188 not in providers_in_chain
+    # 188 不得作為 Ollama provider；本地備援只允許 ollama_local。
+    provider_values = {p.value for p in providers_in_chain}
+    assert "ollama_188" not in provider_values
 
 
 def test_diagnose_fallback_chain_contains_cloud_providers():
@@ -159,7 +160,7 @@ async def test_diagnose_route_primary_is_ollama():
 
     # 雲端 fallback 仍在（OpenClaw / Gemini / Claude 救命備援）
     fb_providers = [p for p, _ in decision.fallback_chain]
-    # ollama_failover_manager 可能轉到 ollama_188，但 ollama variant 必須有
+    # ollama_failover_manager 可能轉到 GCP-B / ollama_local，但雲端救命備援仍必須存在。
     has_cloud_fallback = (
         AIProviderEnum.GEMINI in fb_providers or AIProviderEnum.CLAUDE in fb_providers
     )

apps/api/tests/test_ai_router_failover_integration.py

@@ -83,7 +83,7 @@ async def test_router_uses_failover_when_ollama_initial_provider():
         return_value=_make_failover_result(
             primary_provider="gemini",
             primary_model="gemini-1.5-flash",
-            fallback=[("ollama_188", "qwen2.5:7b-instruct"), ("nemotron", "nvidia/nemotron-mini-4b-instruct")],
+            fallback=[("ollama_local", "qwen2.5:7b-instruct"), ("nemotron", "nvidia/nemotron-mini-4b-instruct")],
         )
     )
 
@@ -109,14 +109,14 @@ async def test_router_uses_failover_when_ollama_initial_provider():
 
 @pytest.mark.asyncio
 async def test_router_failover_fallback_chain_converted():
-    """failover_manager 回傳 fallback_chain → decision.fallback_chain 包含 OLLAMA_188"""
+    """failover_manager 回傳 fallback_chain → decision.fallback_chain 包含 OLLAMA_LOCAL"""
     mock_fm = MagicMock()
     mock_fm.select_provider = AsyncMock(
         return_value=_make_failover_result(
             primary_provider="gemini",
             primary_model="gemini-1.5-flash",
             fallback=[
-                ("ollama_188", "qwen2.5:7b-instruct"),
+                ("ollama_local", "qwen2.5:7b-instruct"),
                 ("nemotron", "nvidia/nemotron-mini-4b-instruct"),
                 ("claude", "claude-haiku-4-5-20251001"),
             ],
@@ -134,8 +134,8 @@ async def test_router_failover_fallback_chain_converted():
         decision = await router.route("test alert message")
 
     fb_providers = [p for p, _ in decision.fallback_chain]
-    assert AIProviderEnum.OLLAMA_188 in fb_providers, (
-        f"OLLAMA_188 not in fallback_chain: {fb_providers}"
+    assert AIProviderEnum.OLLAMA_LOCAL in fb_providers, (
+        f"OLLAMA_LOCAL not in fallback_chain: {fb_providers}"
     )
     assert AIProviderEnum.NEMOTRON in fb_providers
     assert AIProviderEnum.CLAUDE in fb_providers

apps/api/tests/test_failover_alerter.py

@@ -68,7 +68,7 @@ async def test_alert_failover_dedup(mock_redis, mock_telegram_send):
         "to_provider": "gemini",
         "reason": "111 unhealthy",
         "model": "qwen3:8b",
-        "fallback_chain_str": "gemini → ollama_188",
+        "fallback_chain_str": "gemini → ollama_local",
     }
 
     # 第 1 次：dedup pass，發送

apps/api/tests/test_failover_e2e_dispatch.py

@@ -1,16 +1,15 @@
-# apps/api/tests/test_failover_e2e_dispatch.py | 2026-04-26 @ Asia/Taipei
-# 2026-04-26 Wave5 B4 by Claude Engineer-A4 — E2E executor dispatch 測試
-# 驗證 failover 切到 OLLAMA_188 後，HTTP 請求真的打到 OLLAMA_FALLBACK_URL
+# apps/api/tests/test_failover_e2e_dispatch.py | 2026-05-06 @ Asia/Taipei
+# 2026-05-06 Codex — 188 不再作為 Ollama Provider；驗證 ollama_local dispatch。
 """
 E2E：executor dispatch 層驗證
 ===============================
 測試覆蓋（補全 B4 — 整合測試只驗決策層，未驗執行層）：
 
-1. registry 確實有 ollama_188 provider（B1 修復後基本健全性）
-2. Ollama188Provider.is_enabled 在有 OLLAMA_FALLBACK_URL 時為 True
-3. Ollama188Provider.is_enabled 在 OLLAMA_FALLBACK_URL 空字串時為 False
-4. Ollama188Provider.analyze() 真的把 HTTP 打到 OLLAMA_FALLBACK_URL（攔截 httpx）
-5. executor.execute(provider_order=["ollama_188"]) 真的路由到 188 URL
+1. registry 確實有 ollama_local provider，且沒有 ollama_188 provider
+2. OllamaLocalProvider.is_enabled 在有 OLLAMA_FALLBACK_URL 時為 True
+3. OllamaLocalProvider.is_enabled 在 OLLAMA_FALLBACK_URL 空字串時為 False
+4. OllamaLocalProvider.analyze() 真的把 HTTP 打到 OLLAMA_FALLBACK_URL（攔截 httpx）
+5. executor.execute(provider_order=["ollama_local"]) 真的路由到 local URL
 6. Gemini quota pipeline 並行 5 次不超發（B3 atomic 驗證）
 7. Gemini quota TTL 第一次呼叫即設定
 """
@@ -28,31 +27,30 @@ import pytest
 # =============================================================================
 
 
-def test_registry_has_ollama_188_provider():
-    """B1 基本健全性：_init_registry() 後 registry 必須有 ollama_188"""
+def test_registry_has_ollama_local_provider_without_ollama_188():
+    """_init_registry() 後 registry 必須有 ollama_local，且不得有 ollama_188"""
     from src.services.ai_router import _init_registry
 
     registry = _init_registry()
     # registry.get() 只返回 is_enabled=True 的 provider
     # 用 _providers dict 直接檢查（不管 is_enabled）
-    assert "ollama_188" in registry._providers, (
-        "ollama_188 not found in registry._providers — B1 fix 未生效"
-    )
+    assert "ollama_local" in registry._providers
+    assert "ollama_188" not in registry._providers
 
 
-def test_ollama_188_provider_name():
-    """Ollama188Provider.name == 'ollama_188'"""
-    from src.services.ai_providers.ollama import Ollama188Provider
+def test_ollama_local_provider_name():
+    """OllamaLocalProvider.name == 'ollama_local'"""
+    from src.services.ai_providers.ollama import OllamaLocalProvider
 
-    p = Ollama188Provider()
-    assert p.name == "ollama_188"
+    p = OllamaLocalProvider()
+    assert p.name == "ollama_local"
 
 
-def test_ollama_188_provider_privacy_level():
-    """Ollama188Provider.privacy_level == 'local'（本地推理，可接機密資料）"""
-    from src.services.ai_providers.ollama import Ollama188Provider
+def test_ollama_local_provider_privacy_level():
+    """OllamaLocalProvider.privacy_level == 'local'（本地推理，可接機密資料）"""
+    from src.services.ai_providers.ollama import OllamaLocalProvider
 
-    p = Ollama188Provider()
+    p = OllamaLocalProvider()
     assert p.privacy_level == "local"
 
 
@@ -61,45 +59,44 @@ def test_ollama_188_provider_privacy_level():
 # =============================================================================
 
 
-def test_ollama_188_is_enabled_with_fallback_url(monkeypatch):
-    """OLLAMA_FALLBACK_URL 有值 + ENABLE_OLLAMA_188 未設 → is_enabled == True"""
-    from src.services.ai_providers.ollama import Ollama188Provider
-    from src.core.config import get_settings
+def test_ollama_local_is_enabled_with_fallback_url(monkeypatch):
+    """OLLAMA_FALLBACK_URL 有值 + ENABLE_OLLAMA_LOCAL 未設 → is_enabled == True"""
+    from src.services.ai_providers.ollama import OllamaLocalProvider
 
-    monkeypatch.setenv("ENABLE_OLLAMA_188", "true")
+    monkeypatch.setenv("ENABLE_OLLAMA_LOCAL", "true")
     # patch settings 的 OLLAMA_FALLBACK_URL
     mock_settings = MagicMock()
-    mock_settings.OLLAMA_FALLBACK_URL = "http://192.168.0.188:11434"
+    mock_settings.OLLAMA_FALLBACK_URL = "http://192.168.0.111:11434"
     mock_settings.OPENCLAW_TIMEOUT = "60"
 
-    p = Ollama188Provider()
+    p = OllamaLocalProvider()
     # 直接 patch module-level settings 物件
     with patch("src.services.ai_providers.ollama.settings", mock_settings):
         assert p.is_enabled is True
 
 
-def test_ollama_188_is_disabled_without_fallback_url(monkeypatch):
-    """OLLAMA_FALLBACK_URL 空字串 → is_enabled == False（188 節點未設定）"""
-    from src.services.ai_providers.ollama import Ollama188Provider
+def test_ollama_local_is_disabled_without_fallback_url(monkeypatch):
+    """OLLAMA_FALLBACK_URL 空字串 → is_enabled == False（local 節點未設定）"""
+    from src.services.ai_providers.ollama import OllamaLocalProvider
 
-    monkeypatch.setenv("ENABLE_OLLAMA_188", "true")
+    monkeypatch.setenv("ENABLE_OLLAMA_LOCAL", "true")
     mock_settings = MagicMock()
     mock_settings.OLLAMA_FALLBACK_URL = ""
 
-    p = Ollama188Provider()
+    p = OllamaLocalProvider()
     with patch("src.services.ai_providers.ollama.settings", mock_settings):
         assert p.is_enabled is False
 
 
-def test_ollama_188_is_disabled_by_env_flag(monkeypatch):
-    """ENABLE_OLLAMA_188=false → is_enabled == False（即使有 URL）"""
-    from src.services.ai_providers.ollama import Ollama188Provider
+def test_ollama_local_is_disabled_by_env_flag(monkeypatch):
+    """ENABLE_OLLAMA_LOCAL=false → is_enabled == False（即使有 URL）"""
+    from src.services.ai_providers.ollama import OllamaLocalProvider
 
-    monkeypatch.setenv("ENABLE_OLLAMA_188", "false")
+    monkeypatch.setenv("ENABLE_OLLAMA_LOCAL", "false")
     mock_settings = MagicMock()
-    mock_settings.OLLAMA_FALLBACK_URL = "http://192.168.0.188:11434"
+    mock_settings.OLLAMA_FALLBACK_URL = "http://192.168.0.111:11434"
 
-    p = Ollama188Provider()
+    p = OllamaLocalProvider()
     with patch("src.services.ai_providers.ollama.settings", mock_settings):
         assert p.is_enabled is False
 
@@ -110,14 +107,14 @@ def test_ollama_188_is_disabled_by_env_flag(monkeypatch):
 
 
 @pytest.mark.asyncio
-async def test_ollama_188_analyze_dispatches_to_fallback_url():
+async def test_ollama_local_analyze_dispatches_to_fallback_url():
     """
-    B4 核心：Ollama188Provider.analyze() 必須把 HTTP 打到 OLLAMA_FALLBACK_URL。
-    攔截 httpx.AsyncClient.post，記錄實際呼叫 URL，斷言包含 188 IP。
+    B4 核心：OllamaLocalProvider.analyze() 必須把 HTTP 打到 OLLAMA_FALLBACK_URL。
+    攔截 httpx.AsyncClient.post，記錄實際呼叫 URL，斷言包含本地 fallback IP。
     """
-    from src.services.ai_providers.ollama import Ollama188Provider
+    from src.services.ai_providers.ollama import OllamaLocalProvider
 
-    FALLBACK_URL = "http://192.168.0.188:11434"
+    FALLBACK_URL = "http://192.168.0.111:11434"
     captured_urls: list[str] = []
 
     mock_response = MagicMock()
@@ -149,7 +146,7 @@ async def test_ollama_188_analyze_dispatches_to_fallback_url():
         "top_p": 0.9,
     })
 
-    provider = Ollama188Provider()
+    provider = OllamaLocalProvider()
 
     with patch("src.services.ai_providers.ollama.settings", mock_settings):
         with patch("src.services.ai_providers.ollama.get_model_registry", return_value=mock_registry):
@@ -159,45 +156,45 @@ async def test_ollama_188_analyze_dispatches_to_fallback_url():
                 result = await provider.analyze("test prompt", context={})
 
     assert len(captured_urls) > 0, "analyze() 未發出任何 HTTP 請求"
-    assert any("192.168.0.188" in url for url in captured_urls), (
-        f"HTTP 請求未打到 188，實際 URL: {captured_urls}"
+    assert any("192.168.0.111" in url for url in captured_urls), (
+        f"HTTP 請求未打到 local fallback，實際 URL: {captured_urls}"
     )
-    assert result.provider == "ollama_188"
+    assert result.provider == "ollama_local"
 
 
 @pytest.mark.asyncio
-async def test_ollama_188_analyze_returns_error_when_no_fallback_url():
+async def test_ollama_local_analyze_returns_error_when_no_fallback_url():
     """OLLAMA_FALLBACK_URL 未設定 → analyze() 應返回 success=False，不發 HTTP"""
-    from src.services.ai_providers.ollama import Ollama188Provider
+    from src.services.ai_providers.ollama import OllamaLocalProvider
 
     mock_settings = MagicMock()
     mock_settings.OLLAMA_FALLBACK_URL = ""
 
-    provider = Ollama188Provider()
+    provider = OllamaLocalProvider()
     with patch("src.services.ai_providers.ollama.settings", mock_settings):
         result = await provider.analyze("test prompt")
 
     assert result.success is False
-    assert result.provider == "ollama_188"
+    assert result.provider == "ollama_local"
     assert "OLLAMA_FALLBACK_URL" in (result.error or "")
 
 
 @pytest.mark.asyncio
-async def test_executor_dispatches_ollama_188_to_fallback_url():
+async def test_executor_dispatches_ollama_local_to_fallback_url():
     """
-    B4 執行層：AIRouterExecutor.execute(provider_order=["ollama_188"])
-    應路由到 Ollama188Provider，且 HTTP 打到 OLLAMA_FALLBACK_URL。
+    B4 執行層：AIRouterExecutor.execute(provider_order=["ollama_local"])
+    應路由到 OllamaLocalProvider，且 HTTP 打到 OLLAMA_FALLBACK_URL。
     """
     from src.services.ai_router import AIProviderRegistry, AIRouterExecutor, reset_ai_router
-    from src.services.ai_providers.ollama import Ollama188Provider
+    from src.services.ai_providers.ollama import OllamaLocalProvider
     from src.services.ai_providers.interfaces import AIResult
 
     reset_ai_router()
 
-    FALLBACK_URL = "http://192.168.0.188:11434"
+    FALLBACK_URL = "http://192.168.0.111:11434"
     captured_urls: list[str] = []
 
-    # 建立真實 registry，只登錄 ollama_188
+    # 建立真實 registry，只登錄 ollama_local
     registry = AIProviderRegistry()
 
     # mock analyze 讓它回傳成功，但驗 URL 路徑
@@ -206,15 +203,15 @@ async def test_executor_dispatches_ollama_188_to_fallback_url():
         return AIResult(
             raw_response='{"action_title":"ok","confidence":0.9}',
             success=True,
-            provider="ollama_188",
+            provider="ollama_local",
             tokens=10,
         )
 
     mock_settings_global = MagicMock()
     mock_settings_global.OLLAMA_FALLBACK_URL = FALLBACK_URL
 
-    # 建立 Ollama188Provider，mock 其 analyze + is_enabled
-    provider = Ollama188Provider()
+    # 建立 OllamaLocalProvider，mock 其 analyze + is_enabled
+    provider = OllamaLocalProvider()
     provider.analyze = fake_analyze  # type: ignore[method-assign]
 
     # 強制 is_enabled = True（繞過 settings patch 的複雜度）
@@ -233,14 +230,14 @@ async def test_executor_dispatches_ollama_188_to_fallback_url():
             mock_settings.MOCK_MODE = False
             result = await executor.execute(
                 prompt="test alert",
-                provider_order=["ollama_188"],
+                provider_order=["ollama_local"],
                 context={},
             )
 
     assert result.success is True, f"execute 失敗: {result.error}"
-    assert result.provider == "ollama_188", f"provider 不是 ollama_188: {result.provider}"
-    assert any("192.168.0.188" in u for u in captured_urls), (
-        f"HTTP 未打到 188，captured: {captured_urls}"
+    assert result.provider == "ollama_local", f"provider 不是 ollama_local: {result.provider}"
+    assert any("192.168.0.111" in u for u in captured_urls), (
+        f"HTTP 未打到 local fallback，captured: {captured_urls}"
     )
 
 

apps/api/tests/test_model_regression.py

@@ -16,7 +16,7 @@ import httpx
 import pytest
 
 # Ollama 伺服器配置
-OLLAMA_URL = os.getenv("OLLAMA_URL", "http://192.168.0.188:11434")
+OLLAMA_URL = os.getenv("OLLAMA_URL", "http://192.168.0.111:11434")
 DEFAULT_MODEL = os.getenv("OLLAMA_MODEL", "qwen2.5:7b-instruct")
 TIMEOUT = 300  # 秒 (CPU 推理模式需 ~222-666 秒，見 2026-03-26 評估)
 
@@ -111,7 +111,7 @@ async def check_ollama_available() -> bool:
 
 @pytest.mark.integration
 class TestModelRegression:
-    """模型回歸測試 — 需要 Ollama 服務 (192.168.0.188:11434)"""
+    """模型回歸測試 — 需要 Ollama 服務（預設 111，可用 OLLAMA_URL 覆寫）"""
 
     @pytest.fixture(autouse=True)
     async def check_ollama(self):

apps/api/tests/test_model_version_probe.py

@@ -90,8 +90,8 @@ class TestProbeOllamaVersion:
         assert isinstance(info.captured_at, datetime)
 
     @pytest.mark.asyncio
-    async def test_success_188_provider(self):
-        """188 URL → provider='ollama_188'"""
+    async def test_success_local_provider(self):
+        """111 / local proxy URL → provider='ollama_local'"""
         model_entry = {
             "name": "deepseek-r1:14b",
             "modified_at": "2026-04-02T00:00:00Z",
@@ -106,10 +106,10 @@ class TestProbeOllamaVersion:
 
         with patch("httpx.AsyncClient", return_value=mock_client):
             info = await probe_ollama_version(
-                "http://192.168.0.188:11434", "deepseek-r1:14b"
+                "http://192.168.0.111:11434", "deepseek-r1:14b"
             )
 
-        assert info.provider == "ollama_188"
+        assert info.provider == "ollama_local"
 
     @pytest.mark.asyncio
     async def test_model_not_found_raises(self):
@@ -279,7 +279,7 @@ class TestProbeOpenclawNemoVersion:
 
         mock_settings = MagicMock()
         mock_settings.OPENCLAW_DEFAULT_MODEL = "deepseek-r1:14b"
-        mock_settings.OLLAMA_FALLBACK_URL = "http://192.168.0.188:11434"
+        mock_settings.OLLAMA_FALLBACK_URL = "http://192.168.0.111:11434"
 
         with patch("src.services.model_version_probe.settings", mock_settings), \
              patch("httpx.AsyncClient", return_value=mock_client):
@@ -301,7 +301,7 @@ class TestProbeOpenclawNemoVersion:
 
         mock_settings = MagicMock()
         mock_settings.OPENCLAW_DEFAULT_MODEL = "deepseek-r1:14b"
-        mock_settings.OLLAMA_FALLBACK_URL = "http://192.168.0.188:11434"
+        mock_settings.OLLAMA_FALLBACK_URL = "http://192.168.0.111:11434"
 
         with patch("src.services.model_version_probe.settings", mock_settings), \
              patch("httpx.AsyncClient", return_value=mock_client):
@@ -333,7 +333,7 @@ class TestProbeAllProviders:
         """5 個 provider 全部成功 → 回傳 5 筆 ProviderVersionInfo"""
         fake_results = [
             ProviderVersionInfo(provider="ollama", model="qwen2.5:7b-instruct", version="v1"),
-            ProviderVersionInfo(provider="ollama_188", model="qwen2.5:7b-instruct", version="v1"),
+            ProviderVersionInfo(provider="ollama_local", model="qwen2.5:7b-instruct", version="v1"),
             ProviderVersionInfo(provider="gemini", model="gemini-1.5-flash", version="gemini-1.5-flash"),
             ProviderVersionInfo(provider="claude", model="claude-haiku-4-5-20251001", version="claude-haiku-4-5-20251001"),
             ProviderVersionInfo(provider="openclaw_nemo", model="deepseek-r1:14b", version="v1"),
@@ -347,7 +347,7 @@ class TestProbeAllProviders:
 
             mock_settings = MagicMock()
             mock_settings.OLLAMA_URL = "http://34.143.170.20:11434"       # GCP-A（ADR-110）
-            mock_settings.OLLAMA_FALLBACK_URL = "http://192.168.0.188:11434"
+            mock_settings.OLLAMA_FALLBACK_URL = "http://192.168.0.111:11434"
             mock_settings.OLLAMA_HEALTH_CHECK_MODEL = "qwen2.5:7b-instruct"
 
             with patch("src.services.model_version_probe.settings", mock_settings):
@@ -364,8 +364,8 @@ class TestProbeAllProviders:
             raise RuntimeError("simulated failure")
 
         async def _fail_ollama(url, model):
-            if "188" in url:
-                raise RuntimeError("188 offline")
+            if "111" in url:
+                raise RuntimeError("local offline")
             return good
 
         with patch("src.services.model_version_probe.probe_ollama_version", side_effect=_fail_ollama), \
@@ -379,13 +379,13 @@ class TestProbeAllProviders:
 
             mock_settings = MagicMock()
             mock_settings.OLLAMA_URL = "http://34.143.170.20:11434"       # GCP-A（ADR-110）
-            mock_settings.OLLAMA_FALLBACK_URL = "http://192.168.0.188:11434"
+            mock_settings.OLLAMA_FALLBACK_URL = "http://192.168.0.111:11434"
             mock_settings.OLLAMA_HEALTH_CHECK_MODEL = "qwen2.5:7b-instruct"
 
             with patch("src.services.model_version_probe.settings", mock_settings):
                 results = await probe_all_providers()
 
-        # ollama(ok) + ollama_188(fail) + gemini(fail) + claude(ok) + openclaw_nemo(ok) → 3
+        # ollama(ok) + ollama_local(fail) + gemini(fail) + claude(ok) + openclaw_nemo(ok) → 3
         assert len(results) == 3
         providers = {r.provider for r in results}
         assert "ollama" in providers

apps/api/tests/test_model_version_tracker.py

@@ -48,7 +48,7 @@ def _make_info(provider: str, version: str = "v1", digest: str | None = "sha256:
 def _make_five() -> list[ProviderVersionInfo]:
     return [
         _make_info("ollama"),
-        _make_info("ollama_188"),
+        _make_info("ollama_local"),
         _make_info("gemini", digest=None),
         _make_info("claude", digest=None),
         _make_info("openclaw_nemo"),

apps/api/tests/test_ollama_failover_manager.py

@@ -310,7 +310,7 @@ class TestSelectProvider:
         )
 
         with patch.object(manager, "_write_failover_audit", return_value=None):
-            result = await manager.select_provider()
+            await manager.select_provider()
 
         # 並行 check 三台主機（GCP-A / GCP-B / Local）
         assert mock_monitor.check.call_count == 3
@@ -625,7 +625,6 @@ class TestWriteFailoverAudit:
     @pytest.mark.asyncio
     async def test_audit_uses_structlog_not_db(self):
         """_write_failover_audit 應呼叫 structlog，不呼叫 DB"""
-        import structlog
         manager = _make_manager()
         from src.services.ollama_failover_manager import OllamaEndpoint, OllamaRoutingResult
 
@@ -657,22 +656,22 @@ class TestWriteFailoverAudit:
 
 
 # =============================================================================
-# B2: AIProviderEnum.OLLAMA_188 存在
-# 2026-04-25 critic-fix Part2 by Claude Engineer-C2
+# B2: AIProviderEnum.OLLAMA_LOCAL 存在
+# 2026-05-06 Codex — 188 不再作為 Ollama Provider
 # =============================================================================
 
 
-class TestAIProviderEnumOllama188:
-    """B2 修復驗證：AIProviderEnum.OLLAMA_188 存在且 PROVIDER_LATENCY_BUDGET 有對應值"""
+class TestAIProviderEnumOllamaLocal:
+    """B2 修復驗證：AIProviderEnum.OLLAMA_LOCAL 存在且 PROVIDER_LATENCY_BUDGET 有對應值"""
 
-    def test_ollama_188_enum_exists(self):
+    def test_ollama_local_enum_exists(self):
         from src.services.ai_router import AIProviderEnum
-        assert AIProviderEnum.OLLAMA_188.value == "ollama_188"
+        assert AIProviderEnum.OLLAMA_LOCAL.value == "ollama_local"
 
-    def test_ollama_188_in_latency_budget(self):
+    def test_ollama_local_in_latency_budget(self):
         from src.services.ai_router import AIProviderEnum, PROVIDER_LATENCY_BUDGET
-        assert AIProviderEnum.OLLAMA_188 in PROVIDER_LATENCY_BUDGET
-        assert PROVIDER_LATENCY_BUDGET[AIProviderEnum.OLLAMA_188] == 120000
+        assert AIProviderEnum.OLLAMA_LOCAL in PROVIDER_LATENCY_BUDGET
+        assert PROVIDER_LATENCY_BUDGET[AIProviderEnum.OLLAMA_LOCAL] == 90000
 
 
 # =============================================================================

apps/api/tests/test_ollama_health_monitor.py

@@ -42,7 +42,7 @@ from src.services.ollama_health_monitor import (
 # =============================================================================
 
 HOST = "http://34.143.170.20:11434"      # GCP-A Primary（ADR-110 2026-05-03）
-HOST_188 = "http://192.168.0.188:11434"  # 歷史遺留參考常數（已移出主路由）
+HOST_LOCAL = "http://192.168.0.111:11434"  # Local fallback（已移出 188 主路由）
 
 
 @pytest.fixture(autouse=True)

apps/api/tests/test_openclaw_alert_cloud_fallback_gate.py

@@ -0,0 +1,231 @@
+from __future__ import annotations
+
+from dataclasses import dataclass
+from types import SimpleNamespace
+from typing import Any
+from unittest.mock import AsyncMock
+
+import pytest
+
+from src.services import ai_control as ai_control_module
+from src.services import ai_router as ai_router_module
+from src.services import openclaw as openclaw_module
+from src.services.ai_router import AIProviderEnum
+from src.services.intent_classifier import IntentType
+from src.services.openclaw import OpenClawService
+
+
+@dataclass
+class _FakeEndpoint:
+    provider_name: str
+    url: str = "http://example.test"
+
+
+class _FakeRoute:
+    def all_endpoints_in_order(self) -> list[_FakeEndpoint]:
+        return [
+            _FakeEndpoint("ollama_gcp_a"),
+            _FakeEndpoint("ollama_gcp_b"),
+            _FakeEndpoint("ollama_local"),
+            _FakeEndpoint("gemini", ""),
+        ]
+
+
+class _FakeFailoverManager:
+    def __init__(self) -> None:
+        self.task_types: list[str] = []
+
+    async def select_provider(self, task_type: str = "general") -> _FakeRoute:
+        self.task_types.append(task_type)
+        return _FakeRoute()
+
+
+class _UnorderedFailoverManager:
+    async def select_provider(self, task_type: str = "general") -> SimpleNamespace:
+        return SimpleNamespace(
+            all_endpoints_in_order=lambda: [
+                _FakeEndpoint("ollama_local"),
+                _FakeEndpoint("gemini"),
+                _FakeEndpoint("ollama_gcp_b"),
+                _FakeEndpoint("ollama_gcp_a"),
+            ],
+        )
+
+
+class _FakeRouter:
+    async def route(self, prompt: str, context: dict[str, Any]) -> SimpleNamespace:
+        return SimpleNamespace(
+            selected_provider=AIProviderEnum.GEMINI,
+            fallback_chain=[
+                (AIProviderEnum.CLAUDE, "claude"),
+                (AIProviderEnum.OLLAMA, "qwen2.5:7b-instruct"),
+            ],
+            intent=IntentType.DIAGNOSE,
+            routing_reason="high complexity would normally prefer cloud",
+        )
+
+
+class _FakeExecutor:
+    def __init__(self) -> None:
+        self.provider_order: list[str] | None = None
+
+    async def execute(
+        self,
+        *,
+        prompt: str,
+        provider_order: list[str],
+        context: dict[str, Any],
+        cache_ttl: int,
+        require_local: bool,
+    ) -> SimpleNamespace:
+        self.provider_order = provider_order
+        return SimpleNamespace(
+            raw_response='{"root_cause":"ok","suggested_action":"NO_ACTION"}',
+            provider=provider_order[0],
+            success=True,
+            tokens=42,
+            cost_usd=0.0,
+            latency_ms=10.0,
+        )
+
+
+@pytest.mark.asyncio
+async def test_alert_context_uses_ollama_lane_then_gemini_backup(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    fake_executor = _FakeExecutor()
+    fake_failover = _FakeFailoverManager()
+
+    monkeypatch.setattr(openclaw_module.settings, "USE_AI_ROUTER", True)
+    monkeypatch.setattr(openclaw_module.settings, "MOCK_MODE", False)
+    monkeypatch.setattr(openclaw_module.settings, "ALERT_AI_ALLOW_CLOUD_FALLBACK", True)
+    monkeypatch.setattr(openclaw_module.settings, "ALERT_AI_ENFORCE_OLLAMA_FIRST", True)
+    monkeypatch.setattr(ai_control_module, "get_ai_router_enabled", AsyncMock(return_value=None))
+    monkeypatch.setattr(ai_control_module, "get_primary_provider", AsyncMock(return_value=None))
+    monkeypatch.setattr(ai_control_module, "is_provider_disabled", AsyncMock(return_value=False))
+    monkeypatch.setattr(ai_router_module, "get_ai_router", lambda: _FakeRouter())
+    monkeypatch.setattr(ai_router_module, "get_ai_executor", lambda: fake_executor)
+    monkeypatch.setattr(openclaw_module, "get_ollama_failover_manager", lambda: fake_failover)
+
+    service = object.__new__(OpenClawService)
+    result = await service._call_with_fallback(
+        "diagnose alert",
+        alert_context={
+            "incident_id": "INC-1",
+            "alertname": "HostHighCpuLoad",
+            "target_resource": "node-exporter-110",
+        },
+    )
+
+    assert result == (
+        '{"root_cause":"ok","suggested_action":"NO_ACTION"}',
+        "ollama_gcp_a",
+        True,
+        42,
+        0.0,
+    )
+    assert fake_executor.provider_order == ["ollama_gcp_a", "ollama_gcp_b", "ollama_local", "gemini"]
+    assert fake_failover.task_types == ["diagnose"]
+
+
+@pytest.mark.asyncio
+async def test_alert_context_can_disable_cloud_backup_for_cost_stop(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    fake_executor = _FakeExecutor()
+    fake_failover = _FakeFailoverManager()
+
+    monkeypatch.setattr(openclaw_module.settings, "USE_AI_ROUTER", True)
+    monkeypatch.setattr(openclaw_module.settings, "MOCK_MODE", False)
+    monkeypatch.setattr(openclaw_module.settings, "ALERT_AI_ALLOW_CLOUD_FALLBACK", False)
+    monkeypatch.setattr(openclaw_module.settings, "ALERT_AI_ENFORCE_OLLAMA_FIRST", True)
+    monkeypatch.setattr(ai_control_module, "get_ai_router_enabled", AsyncMock(return_value=None))
+    monkeypatch.setattr(ai_control_module, "get_primary_provider", AsyncMock(return_value=None))
+    monkeypatch.setattr(ai_control_module, "is_provider_disabled", AsyncMock(return_value=False))
+    monkeypatch.setattr(ai_router_module, "get_ai_router", lambda: _FakeRouter())
+    monkeypatch.setattr(ai_router_module, "get_ai_executor", lambda: fake_executor)
+    monkeypatch.setattr(openclaw_module, "get_ollama_failover_manager", lambda: fake_failover)
+
+    service = object.__new__(OpenClawService)
+    await service._call_with_fallback(
+        "diagnose alert",
+        alert_context={"incident_id": "INC-1", "alertname": "HostHighCpuLoad"},
+    )
+
+    assert fake_executor.provider_order == ["ollama_gcp_a", "ollama_gcp_b", "ollama_local"]
+
+
+@pytest.mark.asyncio
+async def test_non_alert_context_keeps_router_cloud_order(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    fake_executor = _FakeExecutor()
+
+    monkeypatch.setattr(openclaw_module.settings, "USE_AI_ROUTER", True)
+    monkeypatch.setattr(openclaw_module.settings, "MOCK_MODE", False)
+    monkeypatch.setattr(openclaw_module.settings, "ALERT_AI_ALLOW_CLOUD_FALLBACK", False)
+    monkeypatch.setattr(openclaw_module.settings, "ALERT_AI_ENFORCE_OLLAMA_FIRST", True)
+    monkeypatch.setattr(ai_control_module, "get_ai_router_enabled", AsyncMock(return_value=None))
+    monkeypatch.setattr(ai_control_module, "get_primary_provider", AsyncMock(return_value=None))
+    monkeypatch.setattr(ai_control_module, "is_provider_disabled", AsyncMock(return_value=False))
+    monkeypatch.setattr(ai_router_module, "get_ai_router", lambda: _FakeRouter())
+    monkeypatch.setattr(ai_router_module, "get_ai_executor", lambda: fake_executor)
+
+    service = object.__new__(OpenClawService)
+    await service._call_with_fallback("general question", alert_context={"intent_hint": "query"})
+
+    assert fake_executor.provider_order == ["gemini", "claude", "ollama"]
+
+
+@pytest.mark.asyncio
+async def test_alert_context_uses_gcp_a_gcp_b_then_111_order(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    fake_failover = _FakeFailoverManager()
+
+    monkeypatch.setattr(openclaw_module.settings, "ALERT_AI_ALLOW_CLOUD_FALLBACK", False)
+    monkeypatch.setattr(openclaw_module.settings, "ALERT_AI_ENFORCE_OLLAMA_FIRST", True)
+    monkeypatch.setattr(openclaw_module, "get_ollama_failover_manager", lambda: fake_failover)
+
+    service = object.__new__(OpenClawService)
+    provider_order = await service._resolve_alert_provider_order(
+        task_type="diagnose",
+        alert_context={"incident_id": "INC-1", "alertname": "HostHighCpuLoad"},
+    )
+
+    assert provider_order == ["ollama_gcp_a", "ollama_gcp_b", "ollama_local"]
+
+
+@pytest.mark.asyncio
+async def test_alert_context_sorts_ollama_lane_and_drops_cloud_providers(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(openclaw_module.settings, "ALERT_AI_ALLOW_CLOUD_FALLBACK", False)
+    monkeypatch.setattr(openclaw_module.settings, "ALERT_AI_ENFORCE_OLLAMA_FIRST", True)
+    monkeypatch.setattr(openclaw_module, "get_ollama_failover_manager", lambda: _UnorderedFailoverManager())
+
+    service = object.__new__(OpenClawService)
+    provider_order = await service._resolve_alert_provider_order(
+        task_type="diagnose",
+        alert_context={"incident_id": "INC-1", "alertname": "HostHighCpuLoad"},
+    )
+
+    assert provider_order == ["ollama_gcp_a", "ollama_gcp_b", "ollama_local"]
+
+
+@pytest.mark.asyncio
+async def test_alert_context_sorts_ollama_lane_before_gemini_backup(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(openclaw_module.settings, "ALERT_AI_ALLOW_CLOUD_FALLBACK", True)
+    monkeypatch.setattr(openclaw_module.settings, "ALERT_AI_ENFORCE_OLLAMA_FIRST", True)
+    monkeypatch.setattr(openclaw_module, "get_ollama_failover_manager", lambda: _UnorderedFailoverManager())
+
+    service = object.__new__(OpenClawService)
+    provider_order = await service._resolve_alert_provider_order(
+        task_type="diagnose",
+        alert_context={"incident_id": "INC-1", "alertname": "HostHighCpuLoad"},
+        cloud_provider_order=["claude", "gemini", "ollama"],
+    )
+
+    assert provider_order == ["ollama_gcp_a", "ollama_gcp_b", "ollama_local", "gemini"]

apps/api/tests/test_prompt_validation.py

@@ -18,7 +18,7 @@ import pytest
 from src.core.prompts import OPENCLAW_TEST_PROMPT
 
 # Ollama 配置
-OLLAMA_URL = os.getenv("OLLAMA_URL", "http://192.168.0.188:11434")
+OLLAMA_URL = os.getenv("OLLAMA_URL", "http://192.168.0.111:11434")
 DEFAULT_MODEL = os.getenv("OLLAMA_MODEL", "qwen2.5:7b-instruct")
 TIMEOUT = 300  # 秒 (CPU 推理模式需 ~222-666 秒，見 2026-03-26 評估)
 

apps/web/Dockerfile

@@ -44,6 +44,8 @@ ARG NEXT_PUBLIC_SENTRY_DSN=
 ENV NEXT_PUBLIC_API_URL=${NEXT_PUBLIC_API_URL}
 ENV NEXT_PUBLIC_SENTRY_DSN=${NEXT_PUBLIC_SENTRY_DSN}
 ENV NEXT_TELEMETRY_DISABLED=1
+# 2026-05-05 ogt + Codex: keep self-hosted 110 runner builds from saturating CPU.
+ENV NEXT_PRIVATE_BUILD_WORKER_COUNT=1
 
 # 2026-04-06 ogt: --mount=type=cache 持久化 .next/cache，跨 build 增量編譯
 # 只有變更的頁面重新編譯，未變更頁面直接用 cache → 節省 3-4 min
@@ -51,7 +53,7 @@ ENV NEXT_TELEMETRY_DISABLED=1
 # /root/.cache/turbo 存放 turbo 的 task 輸出快取，避免每次重跑未變動的 packages
 RUN --mount=type=cache,target=/app/apps/web/.next/cache \
     --mount=type=cache,target=/root/.cache/turbo \
-    pnpm turbo build --filter=@awoooi/web
+    pnpm turbo build --filter=@awoooi/web --concurrency=1
 
 FROM base AS runner
 WORKDIR /app

apps/web/messages/en.json

@@ -67,7 +67,8 @@
     "operations": "Operations",
     "securityCompliance": "Security & Compliance",
     "classicAICenter": "Classic AI Center",
-    "governance": "AI Governance"
+    "governance": "AI Governance",
+    "awooop": "AwoooP"
   },
   "locale": {
     "switch": "Switch Language",
@@ -1480,4 +1481,4 @@
       "retry": "Retry"
     }
   }
-}
+}

apps/web/messages/zh-TW.json

@@ -67,7 +67,8 @@
     "operations": "營運",
     "securityCompliance": "安全合規",
     "classicAICenter": "經典 AI 中心",
-    "governance": "AI 治理"
+    "governance": "AI 治理",
+    "awooop": "AwoooP"
   },
   "locale": {
     "switch": "切換語系",
@@ -1481,4 +1482,4 @@
       "retry": "重試"
     }
   }
-}
+}

apps/web/src/app/[locale]/awooop/approvals/page.tsx

@@ -114,7 +114,7 @@ function ApprovalRow({ approval }: { approval: Approval }) {
     <tr
       className={cn(
         "border-b border-border hover:bg-accent/30 transition-colors",
-        isCritical && "bg-red-900/10 hover:bg-red-900/20"
+        isCritical && "bg-[#fff0ef] hover:bg-[#ffe4e1]"
       )}
     >
       <td className="px-4 py-3">
@@ -232,7 +232,7 @@ export default function ApprovalsPage() {
 
       {/* Error State */}
       {error && (
-        <div className="flex items-start gap-3 p-4 bg-red-900/20 border border-red-800/40 rounded-lg">
+        <div className="flex items-start gap-3 border border-[#e2a29b] bg-[#fff0ef] p-4">
           <AlertCircle className="w-5 h-5 text-red-400 flex-shrink-0 mt-0.5" aria-hidden="true" />
           <div>
             <p className="text-sm font-medium text-red-300">無法載入審批資料</p>
@@ -243,7 +243,7 @@ export default function ApprovalsPage() {
 
       {/* Empty State — 所有審批已處理 */}
       {!loading && approvals.length === 0 && !error && (
-        <div className="flex flex-col items-center justify-center py-16 bg-card border border-border rounded-xl">
+        <div className="flex flex-col items-center justify-center border border-[#e0ddd4] bg-white py-16">
           <ShieldCheck className="w-12 h-12 text-green-400 mb-3" aria-hidden="true" />
           <p className="text-sm font-medium text-foreground mb-1">審批佇列為空</p>
           <p className="text-xs text-muted-foreground">目前沒有待審批的 Run</p>
@@ -252,7 +252,7 @@ export default function ApprovalsPage() {
 
       {/* Table */}
       {(loading || approvals.length > 0) && (
-        <div className="bg-card border border-border rounded-xl overflow-hidden shadow-sm">
+        <div className="overflow-hidden border border-[#e0ddd4] bg-white shadow-[0_1px_4px_rgba(0,0,0,0.05)]">
           <div className="overflow-x-auto">
             <table className="w-full" role="table" aria-label="審批佇列">
               <thead>

apps/web/src/app/[locale]/awooop/contracts/page.tsx

@@ -19,20 +19,25 @@ import { cn } from "@/lib/utils";
 // Types
 // =============================================================================
 
-type ContractStatus = "draft" | "published" | "active";
+type ContractStatus = "draft" | "published" | "active" | "revoked";
 
 interface Contract {
-  id: string;
+  contract_id: string;
   contract_family: string;
   project_id: string;
-  status: ContractStatus;
+  lifecycle_status: ContractStatus;
   body_hash: string;
   created_at: string;
 }
 
 interface Tenant {
   project_id: string;
-  name: string;
+  display_name: string;
+}
+
+interface ContractsResponse {
+  contracts?: Contract[];
+  items?: Contract[];
 }
 
 // =============================================================================
@@ -47,21 +52,27 @@ const STATUS_CONFIG: Record<
 > = {
   draft: {
     label: "草稿",
-    bg: "bg-gray-800",
-    text: "text-gray-300",
-    border: "border-gray-600",
+    bg: "bg-[#f4f1e8]",
+    text: "text-[#5f5b52]",
+    border: "border-[#d8d3c7]",
   },
   published: {
     label: "已發佈",
-    bg: "bg-blue-900/40",
-    text: "text-blue-300",
-    border: "border-blue-600/40",
+    bg: "bg-[#eef5ff]",
+    text: "text-[#1f5b9b]",
+    border: "border-[#9bb6d9]",
   },
   active: {
     label: "生效中",
-    bg: "bg-green-900/40",
-    text: "text-green-300",
-    border: "border-green-600/40",
+    bg: "bg-[#f0faf2]",
+    text: "text-[#17602a]",
+    border: "border-[#9bc7a4]",
+  },
+  revoked: {
+    label: "已撤銷",
+    bg: "bg-[#fff0ef]",
+    text: "text-[#9f2f25]",
+    border: "border-[#e2a29b]",
   },
 };
 
@@ -109,7 +120,7 @@ function ContractRow({ contract }: { contract: Contract }) {
         </span>
       </td>
       <td className="px-4 py-3">
-        <StatusBadge status={contract.status} />
+        <StatusBadge status={contract.lifecycle_status} />
       </td>
       <td className="px-4 py-3">
         <span className="font-mono text-xs text-muted-foreground bg-muted px-2 py-0.5 rounded">
@@ -140,7 +151,10 @@ export default function ContractsPage() {
   useEffect(() => {
     fetch(`${API_BASE}/api/v1/platform/tenants`)
       .then((r) => r.json())
-      .then((data) => setTenants(Array.isArray(data.items) ? data.items : []))
+      .then((data) => {
+        const rows = Array.isArray(data.tenants) ? data.tenants : data.items;
+        setTenants(Array.isArray(rows) ? rows : []);
+      })
       .catch(() => {});
   }, []);
 
@@ -154,8 +168,9 @@ export default function ContractsPage() {
         `${API_BASE}/api/v1/platform/contracts?${params.toString()}`
       );
       if (!res.ok) throw new Error(`HTTP ${res.status}`);
-      const data = await res.json();
-      setContracts(Array.isArray(data.items) ? data.items : []);
+      const data: ContractsResponse = await res.json();
+      const rows = Array.isArray(data.contracts) ? data.contracts : data.items;
+      setContracts(Array.isArray(rows) ? rows : []);
     } catch (err) {
       setError(err instanceof Error ? err.message : "載入失敗");
     } finally {
@@ -192,7 +207,7 @@ export default function ContractsPage() {
       </div>
 
       {/* Filters */}
-      <div className="flex items-center gap-3 p-4 bg-card border border-border rounded-xl">
+      <div className="flex items-center gap-3 border border-[#e0ddd4] bg-white p-4">
         <Filter className="w-4 h-4 text-muted-foreground flex-shrink-0" aria-hidden="true" />
         <span className="text-sm text-muted-foreground">篩選：</span>
         <div className="relative">
@@ -205,7 +220,7 @@ export default function ContractsPage() {
             <option value="">所有租戶</option>
             {tenants.map((t) => (
               <option key={t.project_id} value={t.project_id}>
-                {t.name || t.project_id}
+                {t.display_name || t.project_id}
               </option>
             ))}
           </select>
@@ -225,7 +240,7 @@ export default function ContractsPage() {
       )}
 
       {/* Table */}
-      <div className="bg-card border border-border rounded-xl overflow-hidden shadow-sm">
+      <div className="overflow-hidden border border-[#e0ddd4] bg-white shadow-[0_1px_4px_rgba(0,0,0,0.05)]">
         <div className="overflow-x-auto">
           <table className="w-full" role="table" aria-label="合約清單">
             <thead>
@@ -269,7 +284,7 @@ export default function ContractsPage() {
                 </tr>
               ) : (
                 contracts.map((contract) => (
-                  <ContractRow key={contract.id} contract={contract} />
+                  <ContractRow key={contract.contract_id + contract.body_hash} contract={contract} />
                 ))
               )}
             </tbody>

apps/web/src/app/[locale]/awooop/layout.tsx

@@ -6,8 +6,9 @@
 
 "use client";
 
+import { AppLayout } from "@/components/layout";
 import { Link, usePathname } from "@/i18n/routing";
-import { Building2, FileText, Activity, ShieldCheck } from "lucide-react";
+import { Activity, BrainCircuit, Building2, ClipboardList, FileText, ShieldCheck } from "lucide-react";
 import { cn } from "@/lib/utils";
 
 // =============================================================================
@@ -15,6 +16,11 @@ import { cn } from "@/lib/utils";
 // =============================================================================
 
 const navItems = [
+  {
+    label: "工作鏈路",
+    href: "/awooop/work-items" as const,
+    icon: ClipboardList,
+  },
   {
     label: "租戶管理",
     href: "/awooop/tenants" as const,
@@ -43,64 +49,72 @@ const navItems = [
 
 export default function AwoooPLayout({
   children,
+  params,
 }: {
   children: React.ReactNode;
+  params: { locale: string };
 }) {
   const pathname = usePathname();
 
   return (
-    <div className="min-h-full flex flex-col">
-      {/* Console Header */}
-      <div className="bg-card border-b border-border px-6 py-4">
-        <div className="flex items-center justify-between mb-4">
-          <div>
-            <h1 className="text-xl font-bold text-foreground tracking-tight">
-              AwoooP Operator Console
-            </h1>
-            <p className="text-xs text-muted-foreground mt-0.5">
-              Agent 平台管理員後台 — 租戶 · 合約 · Run · 審批
-            </p>
-          </div>
-          <div className="flex items-center gap-2">
-            <span className="inline-flex items-center gap-1.5 px-2.5 py-1 rounded-full text-xs font-medium bg-brand-accent/10 text-brand-accent border border-brand-accent/20">
-              <span className="w-1.5 h-1.5 rounded-full bg-brand-accent animate-pulse" />
+    <AppLayout locale={params.locale} showBackground={false}>
+      <div className="min-h-[calc(100vh-116px)] bg-[#f7f5ee] border border-[#e0ddd4]">
+        <div className="border-b border-[#e0ddd4] bg-[#faf9f3] px-5 py-4">
+          <div className="flex flex-wrap items-center justify-between gap-3">
+            <div className="flex items-center gap-3">
+              <span className="flex h-9 w-9 items-center justify-center border border-[#d8d3c7] bg-white text-[#141413]">
+                <BrainCircuit className="h-4 w-4" aria-hidden="true" />
+              </span>
+              <div>
+                <h1 className="text-lg font-semibold tracking-normal text-[#141413]">
+                  AwoooP Operator Console
+                </h1>
+                <div className="mt-1 flex items-center gap-2 text-xs text-[#77736a]">
+                  <span className="font-mono">Control Plane</span>
+                  <span className="h-1 w-1 rounded-full bg-[#d97757]" />
+                  <span className="font-mono">Shadow First</span>
+                </div>
+              </div>
+            </div>
+            <span className="inline-flex items-center gap-2 border border-[#d8d3c7] bg-white px-3 py-1.5 text-xs font-semibold text-[#141413]">
+              <span className="h-1.5 w-1.5 rounded-full bg-[#22c55e]" />
               OPERATOR
             </span>
           </div>
+
+          <nav
+            className="mt-4 flex flex-wrap gap-1"
+            role="navigation"
+            aria-label="AwoooP 主要導航"
+          >
+            {navItems.map((item) => {
+              const Icon = item.icon;
+              const isActive =
+                pathname === item.href ||
+                pathname?.startsWith(item.href + "/");
+
+              return (
+                <Link
+                  key={item.href}
+                  href={item.href}
+                  aria-current={isActive ? "page" : undefined}
+                  className={cn(
+                    "inline-flex items-center gap-2 border px-3 py-2 text-sm font-medium transition-colors",
+                    isActive
+                      ? "border-[#d97757] bg-white text-[#141413]"
+                      : "border-transparent text-[#77736a] hover:border-[#d8d3c7] hover:bg-white hover:text-[#141413]"
+                  )}
+                >
+                  <Icon className="h-4 w-4" aria-hidden="true" />
+                  {item.label}
+                </Link>
+              );
+            })}
+          </nav>
         </div>
 
-        {/* Tab Navigation */}
-        <nav className="flex gap-1" role="navigation" aria-label="AwoooP 主要導航">
-          {navItems.map((item) => {
-            const Icon = item.icon;
-            const isActive =
-              pathname === item.href ||
-              pathname?.startsWith(item.href + "/");
-
-            return (
-              <Link
-                key={item.href}
-                href={item.href}
-                aria-current={isActive ? "page" : undefined}
-                className={cn(
-                  "flex items-center gap-2 px-4 py-2 rounded-lg text-sm font-medium transition-all duration-150",
-                  isActive
-                    ? "bg-brand-accent/15 text-brand-accent border border-brand-accent/30"
-                    : "text-muted-foreground hover:text-foreground hover:bg-accent"
-                )}
-              >
-                <Icon className="w-4 h-4" aria-hidden="true" />
-                {item.label}
-              </Link>
-            );
-          })}
-        </nav>
+        <main className="px-5 py-5">{children}</main>
       </div>
-
-      {/* Page Content */}
-      <main className="flex-1 px-6 py-6">
-        {children}
-      </main>
-    </div>
+    </AppLayout>
   );
 }

apps/web/src/app/[locale]/awooop/page.tsx

@@ -2,8 +2,12 @@
 // WOOO AIOps - AwoooP Console 入口重導向
 // =============================================================================
 
-import { redirect } from "@/i18n/routing";
+import { redirect } from "next/navigation";
 
-export default function AwoooPPage() {
-  redirect("/awooop/tenants");
+export default function AwoooPPage({
+  params,
+}: {
+  params: { locale: string };
+}) {
+  redirect(`/${params.locale}/awooop/work-items`);
 }

apps/web/src/app/[locale]/awooop/runs/page.tsx

@@ -23,16 +23,14 @@ import { cn } from "@/lib/utils";
 // =============================================================================
 
 type RunState =
-  | "CREATED"
-  | "QUEUED"
-  | "POLICY_RESOLVED"
-  | "RUNNING"
-  | "WAITING_TOOL"
-  | "WAITING_APPROVAL"
-  | "RESUMED"
-  | "COMPLETED"
-  | "FAILED"
-  | "CANCELLED";
+  | "pending"
+  | "running"
+  | "waiting_tool"
+  | "waiting_approval"
+  | "completed"
+  | "failed"
+  | "cancelled"
+  | "timeout";
 
 interface Run {
   run_id: string;
@@ -40,18 +38,19 @@ interface Run {
   agent_id: string;
   state: RunState;
   is_shadow: boolean;
-  token_usage_input: number | null;
-  token_usage_output: number | null;
+  cost_usd: number | string;
+  step_count: number;
   created_at: string;
 }
 
 interface Tenant {
   project_id: string;
-  name: string;
+  display_name: string;
 }
 
 interface RunsResponse {
-  items: Run[];
+  runs?: Run[];
+  items?: Run[];
   total: number;
   page: number;
   per_page: number;
@@ -69,66 +68,54 @@ const STATE_CONFIG: Record<
   RunState,
   { label: string; bg: string; text: string; border: string; pulse?: boolean }
 > = {
-  CREATED: {
-    label: "已建立",
-    bg: "bg-gray-800",
-    text: "text-gray-300",
-    border: "border-gray-600",
+  pending: {
+    label: "待執行",
+    bg: "bg-[#f4f1e8]",
+    text: "text-[#5f5b52]",
+    border: "border-[#d8d3c7]",
   },
-  QUEUED: {
-    label: "排隊中",
-    bg: "bg-gray-800",
-    text: "text-gray-400",
-    border: "border-gray-600",
-  },
-  POLICY_RESOLVED: {
-    label: "策略已解析",
-    bg: "bg-blue-900/40",
-    text: "text-blue-300",
-    border: "border-blue-600/40",
-  },
-  RUNNING: {
+  running: {
     label: "執行中",
-    bg: "bg-green-900/40",
-    text: "text-green-300",
-    border: "border-green-600/40",
+    bg: "bg-[#f0faf2]",
+    text: "text-[#17602a]",
+    border: "border-[#9bc7a4]",
     pulse: true,
   },
-  WAITING_TOOL: {
+  waiting_tool: {
     label: "等待工具",
-    bg: "bg-yellow-900/40",
-    text: "text-yellow-300",
-    border: "border-yellow-600/40",
+    bg: "bg-[#fff7e8]",
+    text: "text-[#8a5a08]",
+    border: "border-[#d9b36f]",
   },
-  WAITING_APPROVAL: {
+  waiting_approval: {
     label: "等待審批",
-    bg: "bg-yellow-900/40",
-    text: "text-yellow-300",
-    border: "border-yellow-600/40",
+    bg: "bg-[#fff7e8]",
+    text: "text-[#8a5a08]",
+    border: "border-[#d9b36f]",
   },
-  RESUMED: {
-    label: "已恢復",
-    bg: "bg-purple-900/40",
-    text: "text-purple-300",
-    border: "border-purple-600/40",
-  },
-  COMPLETED: {
+  completed: {
     label: "已完成",
-    bg: "bg-green-900/40",
-    text: "text-green-400",
-    border: "border-green-600/40",
+    bg: "bg-[#f0faf2]",
+    text: "text-[#17602a]",
+    border: "border-[#9bc7a4]",
   },
-  FAILED: {
+  failed: {
     label: "失敗",
-    bg: "bg-red-900/40",
-    text: "text-red-300",
-    border: "border-red-600/40",
+    bg: "bg-[#fff0ef]",
+    text: "text-[#9f2f25]",
+    border: "border-[#e2a29b]",
   },
-  CANCELLED: {
+  cancelled: {
     label: "已取消",
-    bg: "bg-red-900/30",
-    text: "text-red-400",
-    border: "border-red-700/40",
+    bg: "bg-[#fff0ef]",
+    text: "text-[#9f2f25]",
+    border: "border-[#e2a29b]",
+  },
+  timeout: {
+    label: "已超時",
+    bg: "bg-[#fff0ef]",
+    text: "text-[#9f2f25]",
+    border: "border-[#e2a29b]",
   },
 };
 
@@ -137,7 +124,7 @@ const STATE_CONFIG: Record<
 // =============================================================================
 
 function RunStateBadge({ state }: { state: RunState }) {
-  const config = STATE_CONFIG[state] ?? STATE_CONFIG.CREATED;
+  const config = STATE_CONFIG[state] ?? STATE_CONFIG.pending;
   return (
     <span
       className={cn(
@@ -158,7 +145,7 @@ function RunStateBadge({ state }: { state: RunState }) {
 function ShadowBadge({ isShadow }: { isShadow: boolean }) {
   if (!isShadow) return <span className="text-muted-foreground text-sm">--</span>;
   return (
-    <span className="inline-flex items-center px-2 py-0.5 rounded text-xs font-medium bg-gray-800 text-gray-400 border border-gray-600">
+    <span className="inline-flex items-center border border-[#d8d3c7] bg-white px-2 py-0.5 text-xs font-medium text-[#5f5b52]">
       Shadow
     </span>
   );
@@ -174,8 +161,7 @@ function RunRow({ run }: { run: Run }) {
       })
     : "--";
 
-  const totalTokens =
-    (run.token_usage_input ?? 0) + (run.token_usage_output ?? 0);
+  const cost = Number(run.cost_usd ?? 0);
 
   return (
     <tr className="border-b border-border hover:bg-accent/30 transition-colors">
@@ -202,12 +188,12 @@ function RunRow({ run }: { run: Run }) {
       </td>
       <td className="px-4 py-3">
         <span className="flex items-center gap-1 text-sm font-mono text-muted-foreground">
-          {totalTokens > 0 ? (
+          {run.step_count > 0 || cost > 0 ? (
             <>
               <Cpu className="w-3.5 h-3.5" aria-hidden="true" />
-              {totalTokens.toLocaleString()}
+              {run.step_count.toLocaleString()} steps
               <span className="text-xs text-muted-foreground/60">
-                ({run.token_usage_input ?? 0}↑ {run.token_usage_output ?? 0}↓)
+                (${cost.toFixed(4)})
               </span>
             </>
           ) : (
@@ -244,7 +230,10 @@ export default function RunsPage() {
   useEffect(() => {
     fetch(`${API_BASE}/api/v1/platform/tenants`)
       .then((r) => r.json())
-      .then((data) => setTenants(Array.isArray(data.items) ? data.items : []))
+      .then((data) => {
+        const rows = Array.isArray(data.tenants) ? data.tenants : data.items;
+        setTenants(Array.isArray(rows) ? rows : []);
+      })
       .catch(() => {});
   }, []);
 
@@ -253,7 +242,7 @@ export default function RunsPage() {
       setError(null);
       const params = new URLSearchParams();
       if (projectFilter) params.set("project_id", projectFilter);
-      if (statusFilter) params.set("status", statusFilter);
+      if (statusFilter) params.set("state", statusFilter);
       params.set("page", String(page));
       params.set("per_page", String(PER_PAGE));
 
@@ -262,7 +251,8 @@ export default function RunsPage() {
       );
       if (!res.ok) throw new Error(`HTTP ${res.status}`);
       const data: RunsResponse = await res.json();
-      setRuns(Array.isArray(data.items) ? data.items : []);
+      const rows = Array.isArray(data.runs) ? data.runs : data.items;
+      setRuns(Array.isArray(rows) ? rows : []);
       setTotal(data.total ?? 0);
       setLastRefresh(new Date());
     } catch (err) {
@@ -320,7 +310,7 @@ export default function RunsPage() {
       </div>
 
       {/* Filters */}
-      <div className="flex items-center gap-3 p-4 bg-card border border-border rounded-xl flex-wrap">
+      <div className="flex flex-wrap items-center gap-3 border border-[#e0ddd4] bg-white p-4">
         <Filter className="w-4 h-4 text-muted-foreground flex-shrink-0" aria-hidden="true" />
         <span className="text-sm text-muted-foreground">篩選：</span>
 
@@ -335,7 +325,7 @@ export default function RunsPage() {
             <option value="">所有租戶</option>
             {tenants.map((t) => (
               <option key={t.project_id} value={t.project_id}>
-                {t.name || t.project_id}
+                {t.display_name || t.project_id}
               </option>
             ))}
           </select>
@@ -373,7 +363,7 @@ export default function RunsPage() {
       )}
 
       {/* Table */}
-      <div className="bg-card border border-border rounded-xl overflow-hidden shadow-sm">
+      <div className="overflow-hidden border border-[#e0ddd4] bg-white shadow-[0_1px_4px_rgba(0,0,0,0.05)]">
         <div className="overflow-x-auto">
           <table className="w-full" role="table" aria-label="Run 清單">
             <thead>
@@ -394,7 +384,7 @@ export default function RunsPage() {
                   Shadow
                 </th>
                 <th className="text-left px-4 py-3 text-xs font-medium text-muted-foreground uppercase tracking-wider">
-                  Token 用量
+                  成本 / Steps
                 </th>
                 <th className="text-left px-4 py-3 text-xs font-medium text-muted-foreground uppercase tracking-wider">
                   建立時間

apps/web/src/app/[locale]/awooop/tenants/page.tsx

@@ -21,6 +21,7 @@ import { cn } from "@/lib/utils";
 // =============================================================================
 
 type MigrationMode =
+  | "legacy_awoooi_default"
   | "shadow"
   | "canary"
   | "read_only"
@@ -29,14 +30,15 @@ type MigrationMode =
 
 interface Tenant {
   project_id: string;
-  name: string;
+  display_name: string;
   migration_mode: MigrationMode;
-  budget_limit_usd: number | null;
-  is_suspended: boolean;
+  budget_limit_usd: number | string | null;
+  is_active: boolean;
 }
 
 interface ApiResponse {
-  items: Tenant[];
+  tenants?: Tenant[];
+  items?: Tenant[];
   total: number;
 }
 
@@ -50,35 +52,41 @@ const MIGRATION_MODE_CONFIG: Record<
   MigrationMode,
   { label: string; bg: string; text: string; border: string }
 > = {
+  legacy_awoooi_default: {
+    label: "Legacy",
+    bg: "bg-white",
+    text: "text-[#5f5b52]",
+    border: "border-[#d8d3c7]",
+  },
   shadow: {
     label: "Shadow",
-    bg: "bg-gray-800",
-    text: "text-gray-300",
-    border: "border-gray-600",
+    bg: "bg-[#f4f1e8]",
+    text: "text-[#5f5b52]",
+    border: "border-[#d8d3c7]",
   },
   canary: {
     label: "Canary",
-    bg: "bg-yellow-900/40",
-    text: "text-yellow-300",
-    border: "border-yellow-600/40",
+    bg: "bg-[#fff7e8]",
+    text: "text-[#8a5a08]",
+    border: "border-[#d9b36f]",
   },
   read_only: {
     label: "Read Only",
-    bg: "bg-blue-900/40",
-    text: "text-blue-300",
-    border: "border-blue-600/40",
+    bg: "bg-[#eef5ff]",
+    text: "text-[#1f5b9b]",
+    border: "border-[#9bb6d9]",
   },
   suggest: {
     label: "Suggest",
-    bg: "bg-purple-900/40",
-    text: "text-purple-300",
-    border: "border-purple-600/40",
+    bg: "bg-[#f6f0ff]",
+    text: "text-[#6541a5]",
+    border: "border-[#baa7de]",
   },
   auto_remediate: {
     label: "Auto Remediate",
-    bg: "bg-green-900/40",
-    text: "text-green-300",
-    border: "border-green-600/40",
+    bg: "bg-[#f0faf2]",
+    text: "text-[#17602a]",
+    border: "border-[#9bc7a4]",
   },
 };
 
@@ -104,12 +112,12 @@ function MigrationModeBadge({ mode }: { mode: MigrationMode }) {
 
 function SuspendedBadge({ suspended }: { suspended: boolean }) {
   return suspended ? (
-    <span className="inline-flex items-center gap-1 px-2.5 py-0.5 rounded-md text-xs font-medium bg-red-900/40 text-red-300 border border-red-600/40">
+    <span className="inline-flex items-center gap-1 border border-[#e2a29b] bg-[#fff0ef] px-2.5 py-0.5 text-xs font-medium text-[#9f2f25]">
       <Ban className="w-3 h-3" aria-hidden="true" />
       停用
     </span>
   ) : (
-    <span className="inline-flex items-center gap-1 px-2.5 py-0.5 rounded-md text-xs font-medium bg-green-900/40 text-green-300 border border-green-600/40">
+    <span className="inline-flex items-center gap-1 border border-[#9bc7a4] bg-[#f0faf2] px-2.5 py-0.5 text-xs font-medium text-[#17602a]">
       <CheckCircle2 className="w-3 h-3" aria-hidden="true" />
       正常
     </span>
@@ -117,6 +125,9 @@ function SuspendedBadge({ suspended }: { suspended: boolean }) {
 }
 
 function TenantRow({ tenant }: { tenant: Tenant }) {
+  const budget =
+    tenant.budget_limit_usd == null ? null : Number(tenant.budget_limit_usd);
+
   return (
     <tr className="border-b border-border hover:bg-accent/30 transition-colors">
       <td className="px-4 py-3">
@@ -125,7 +136,7 @@ function TenantRow({ tenant }: { tenant: Tenant }) {
         </span>
       </td>
       <td className="px-4 py-3">
-        <span className="text-sm text-foreground font-medium">{tenant.name || "--"}</span>
+        <span className="text-sm text-foreground font-medium">{tenant.display_name || "--"}</span>
       </td>
       <td className="px-4 py-3">
         <MigrationModeBadge mode={tenant.migration_mode} />
@@ -135,7 +146,7 @@ function TenantRow({ tenant }: { tenant: Tenant }) {
           {tenant.budget_limit_usd != null ? (
             <>
               <DollarSign className="w-3.5 h-3.5" aria-hidden="true" />
-              {tenant.budget_limit_usd.toLocaleString("en-US", {
+              {budget?.toLocaleString("en-US", {
                 minimumFractionDigits: 2,
               })}
             </>
@@ -145,7 +156,7 @@ function TenantRow({ tenant }: { tenant: Tenant }) {
         </span>
       </td>
       <td className="px-4 py-3">
-        <SuspendedBadge suspended={tenant.is_suspended} />
+        <SuspendedBadge suspended={!tenant.is_active} />
       </td>
     </tr>
   );
@@ -166,7 +177,8 @@ export default function TenantsPage() {
       const res = await fetch(`${API_BASE}/api/v1/platform/tenants`);
       if (!res.ok) throw new Error(`HTTP ${res.status}`);
       const data: ApiResponse = await res.json();
-      setTenants(Array.isArray(data.items) ? data.items : []);
+      const rows = Array.isArray(data.tenants) ? data.tenants : data.items;
+      setTenants(Array.isArray(rows) ? rows : []);
     } catch (err) {
       setError(err instanceof Error ? err.message : "載入失敗");
     } finally {
@@ -214,7 +226,7 @@ export default function TenantsPage() {
       )}
 
       {/* Table */}
-      <div className="bg-card border border-border rounded-xl overflow-hidden shadow-sm">
+      <div className="overflow-hidden border border-[#e0ddd4] bg-white shadow-[0_1px_4px_rgba(0,0,0,0.05)]">
         <div className="overflow-x-auto">
           <table className="w-full" role="table" aria-label="租戶清單">
             <thead>

apps/web/src/app/[locale]/awooop/work-items/page.tsx

@@ -0,0 +1,254 @@
+// =============================================================================
+// WOOO AIOps - AwoooP 工作鏈路
+// =============================================================================
+// 將 AwoooP 實施項目對齊到 Operator Console 可觀測面。
+
+"use client";
+
+import {
+  Activity,
+  ArrowRight,
+  ClipboardList,
+  Database,
+  Gauge,
+  GitBranch,
+  Network,
+  ShieldCheck,
+} from "lucide-react";
+import { Link } from "@/i18n/routing";
+import { cn } from "@/lib/utils";
+
+type WorkStatus = "live" | "in_progress" | "blocked" | "watching";
+
+type WorkItem = {
+  phase: string;
+  title: string;
+  status: WorkStatus;
+  surface: string;
+  source: string;
+  gate: string;
+  href: "/awooop/tenants" | "/awooop/contracts" | "/awooop/runs" | "/awooop/approvals";
+};
+
+const statusConfig: Record<WorkStatus, { label: string; className: string }> = {
+  live: {
+    label: "已接線",
+    className: "border-[#9bc7a4] bg-[#f0faf2] text-[#17602a]",
+  },
+  in_progress: {
+    label: "推進中",
+    className: "border-[#d9b36f] bg-[#fff7e8] text-[#8a5a08]",
+  },
+  blocked: {
+    label: "阻塞",
+    className: "border-[#e2a29b] bg-[#fff0ef] text-[#9f2f25]",
+  },
+  watching: {
+    label: "觀察期",
+    className: "border-[#9bb6d9] bg-[#eef5ff] text-[#1f5b9b]",
+  },
+};
+
+const workItems: WorkItem[] = [
+  {
+    phase: "P0",
+    title: "AI 路由以 GCP-A/GCP-B/111 Ollama 優先",
+    status: "live",
+    surface: "Run 監控",
+    source: "ai_routing_decision / ollama_failover_decision",
+    gate: "Gemini 僅能作為 fallback",
+    href: "/awooop/runs",
+  },
+  {
+    phase: "P0",
+    title: "飛輪 KPI 改讀 auto_repair_executions",
+    status: "in_progress",
+    surface: "工作鏈路 / 系統報告",
+    source: "auto_repair_executions",
+    gate: "修復率不得再讀 incidents.outcome",
+    href: "/awooop/runs",
+  },
+  {
+    phase: "P0",
+    title: "審批與 Run State 對齊",
+    status: "live",
+    surface: "審批佇列",
+    source: "awooop_run_state",
+    gate: "waiting_approval 才能 decide",
+    href: "/awooop/approvals",
+  },
+  {
+    phase: "P1",
+    title: "Contract Lifecycle",
+    status: "watching",
+    surface: "合約儀表板",
+    source: "awooop_contract_revisions",
+    gate: "draft → published → active",
+    href: "/awooop/contracts",
+  },
+  {
+    phase: "P1",
+    title: "Tenant Migration State",
+    status: "watching",
+    surface: "租戶管理",
+    source: "awooop_projects",
+    gate: "shadow gate 需量化",
+    href: "/awooop/tenants",
+  },
+  {
+    phase: "P1",
+    title: "MCP Gateway 與 Context Firewall",
+    status: "watching",
+    surface: "Run 監控",
+    source: "mcp_gateway audit / redaction",
+    gate: "tool call 必須帶 project_id",
+    href: "/awooop/runs",
+  },
+  {
+    phase: "P2",
+    title: "Communication Hub",
+    status: "watching",
+    surface: "Run 監控 / 審批佇列",
+    source: "conversation_event / outbound_message",
+    gate: "Telegram 先 mirror 再切流",
+    href: "/awooop/runs",
+  },
+  {
+    phase: "P2",
+    title: "Operator Console 正式接入主站",
+    status: "in_progress",
+    surface: "AwoooP",
+    source: "apps/web/src/app/[locale]/awooop",
+    gate: "/zh-TW/awooop 不得再 redirect 異常",
+    href: "/awooop/tenants",
+  },
+];
+
+function StatusBadge({ status }: { status: WorkStatus }) {
+  const config = statusConfig[status];
+  return (
+    <span
+      className={cn(
+        "inline-flex items-center border px-2 py-0.5 text-xs font-semibold",
+        config.className
+      )}
+    >
+      {config.label}
+    </span>
+  );
+}
+
+const summary = [
+  { label: "Live", value: workItems.filter((i) => i.status === "live").length, icon: Activity },
+  { label: "In Progress", value: workItems.filter((i) => i.status === "in_progress").length, icon: GitBranch },
+  { label: "Watching", value: workItems.filter((i) => i.status === "watching").length, icon: Gauge },
+  { label: "Blocked", value: workItems.filter((i) => i.status === "blocked").length, icon: ShieldCheck },
+];
+
+export default function AwoooPWorkItemsPage() {
+  return (
+    <div className="space-y-5">
+      <div className="flex flex-wrap items-center justify-between gap-3">
+        <div className="flex items-center gap-3">
+          <ClipboardList className="h-5 w-5 text-[#d97757]" aria-hidden="true" />
+          <div>
+            <h2 className="text-lg font-semibold tracking-normal text-[#141413]">
+              工作鏈路
+            </h2>
+            <p className="text-xs text-[#77736a]">
+              {workItems.length} 個控制點
+            </p>
+          </div>
+        </div>
+      </div>
+
+      <div className="grid gap-px border border-[#e0ddd4] bg-[#e0ddd4] md:grid-cols-4">
+        {summary.map((item) => {
+          const Icon = item.icon;
+          return (
+            <div key={item.label} className="bg-white px-4 py-3">
+              <div className="flex items-center justify-between">
+                <span className="text-xs font-medium text-[#77736a]">{item.label}</span>
+                <Icon className="h-4 w-4 text-[#87867f]" aria-hidden="true" />
+              </div>
+              <div className="mt-2 font-mono text-2xl font-semibold text-[#141413]">
+                {item.value}
+              </div>
+            </div>
+          );
+        })}
+      </div>
+
+      <div className="overflow-hidden border border-[#e0ddd4] bg-white">
+        <div className="overflow-x-auto">
+          <table className="w-full" role="table" aria-label="AwoooP 工作鏈路">
+            <thead>
+              <tr className="border-b border-[#e0ddd4] bg-[#faf9f3]">
+                <th className="px-4 py-3 text-left text-xs font-semibold text-[#77736a]">
+                  Phase
+                </th>
+                <th className="px-4 py-3 text-left text-xs font-semibold text-[#77736a]">
+                  工作項目
+                </th>
+                <th className="px-4 py-3 text-left text-xs font-semibold text-[#77736a]">
+                  狀態
+                </th>
+                <th className="px-4 py-3 text-left text-xs font-semibold text-[#77736a]">
+                  操作面
+                </th>
+                <th className="px-4 py-3 text-left text-xs font-semibold text-[#77736a]">
+                  Source
+                </th>
+                <th className="px-4 py-3 text-left text-xs font-semibold text-[#77736a]">
+                  Gate
+                </th>
+                <th className="px-4 py-3 text-left text-xs font-semibold text-[#77736a]">
+                  Link
+                </th>
+              </tr>
+            </thead>
+            <tbody>
+              {workItems.map((item) => (
+                <tr key={`${item.phase}-${item.title}`} className="border-b border-[#eee9dd] last:border-b-0">
+                  <td className="px-4 py-3 font-mono text-xs font-semibold text-[#141413]">
+                    {item.phase}
+                  </td>
+                  <td className="px-4 py-3 text-sm font-medium text-[#141413]">
+                    {item.title}
+                  </td>
+                  <td className="px-4 py-3">
+                    <StatusBadge status={item.status} />
+                  </td>
+                  <td className="px-4 py-3 text-sm text-[#5f5b52]">
+                    {item.surface}
+                  </td>
+                  <td className="px-4 py-3">
+                    <span className="inline-flex items-center gap-1.5 font-mono text-xs text-[#5f5b52]">
+                      <Database className="h-3.5 w-3.5" aria-hidden="true" />
+                      {item.source}
+                    </span>
+                  </td>
+                  <td className="px-4 py-3">
+                    <span className="inline-flex items-center gap-1.5 text-sm text-[#5f5b52]">
+                      <Network className="h-3.5 w-3.5" aria-hidden="true" />
+                      {item.gate}
+                    </span>
+                  </td>
+                  <td className="px-4 py-3">
+                    <Link
+                      href={item.href}
+                      className="inline-flex items-center gap-1.5 border border-[#d8d3c7] px-2.5 py-1 text-xs font-semibold text-[#141413] hover:border-[#d97757]"
+                    >
+                      開啟
+                      <ArrowRight className="h-3.5 w-3.5" aria-hidden="true" />
+                    </Link>
+                  </td>
+                </tr>
+              ))}
+            </tbody>
+          </table>
+        </div>
+      </div>
+    </div>
+  );
+}

apps/web/src/components/layout/sidebar.tsx

@@ -87,6 +87,7 @@ const NAV_SECTIONS: NavSection[] = [
       { id: 'security-compliance', href: '/security-compliance',    labelKey: 'securityCompliance',Icon: Shield },
       { id: 'knowledge',           href: '/knowledge',              labelKey: 'knowledge',         Icon: BookOpen },
       { id: 'governance',          href: '/governance',             labelKey: 'governance',        Icon: ShieldCheck },
+      { id: 'awooop',              href: '/awooop',                 labelKey: 'awooop',           Icon: BrainCircuit },
     ],
   },
   {

docs/LOGBOOK.md

@@ -6,6 +6,71 @@
 
 ---
 
+## 2026-05-05 | 重開機後排程與 startup baseline 修復
+
+**背景**：四台主機非預期重開機後，統帥要求確認所有服務、網站、工具、資料庫與排程都能正常恢復，不能只看容器 `healthy`。
+
+**本次排程/啟動鏈修補**：
+- 120/121 K3s 回到 Ready；CD workflow 目標從 121 改為 120，避免 121 worker kubeconfig `127.0.0.1:6443` 造成 Secrets patch 失敗；120 已驗證 limited sudo kubectl 可用。
+- K8s CronJob 修正：`k3s-status-report`、`weekly-report`、`km-vectorize` 改用存在的 service account、live API image、cluster service DNS；手動 job 驗證 drift/k3s/weekly 可完成，歷史 failed jobs 已清掉。
+- KM embedding schema 從 768/錯誤 typmod 修為 `vector(1024)`；原 embedding 已備份到 `knowledge_entries_embedding_backup_20260505`，正在以 `bge-m3:latest` 重建。
+- 188 momo backup script 修正 quote/validation/Telegram optional/error cleanup；成功產出 `/home/ollama/momo_backups/momo_analytics_20260505_212032.sql.gz`。
+- 188 `backup-from-110.sh` 因 SSH config 權限錯誤導致 `HostBackupFailed`；修正 `.ssh/config` 權限與 110 identity 設定後，以低優先權手動備份成功，Prometheus `backup_110_last_success_timestamp` 已更新。
+- 188 momo-scheduler 修正 dashboard URL：容器內改打 `http://momo-pro-system`，不再打 `127.0.0.1:5000`。
+- 188 Google Drive token 從 legacy pickle 轉為 JSON，scheduler 容器內 `GoogleDriveService().authenticate()` 通過。
+- 188 daily sales import 修正 Excel sheet 選擇，優先讀 `即時業績明細`；手動匯入成功 `19934` 筆，日期 `2026-04-01 ~ 2026-05-03`。
+- 188 import 尾端驗證修正：改比對本次匯入日期範圍，不再用全表筆數硬比；`daily_sales_snapshot` 與 `realtime_sales_monthly` 在該日期範圍皆 `19934` 筆且驗證通過。
+- 110 startup 修復：移除 `/etc/sysctl.conf` 中誤寫的非法敏感純文字行；`systemd-sysctl` 恢復成功。
+- 110 停用兩個過期 startup units：`momo-startup-complete.service`（指向不存在路徑/錯 host）與 `wooo-staggered-startup.service`（舊 GitLab 延遲啟動且會增加重開機負載）。
+- 110 `awoooi-startup-110.service` timeout 從 5 分鐘延長到 15 分鐘，重跑後 `ActiveState=active`、`SubState=exited`、`Result=success`，`systemctl --failed` 為 0。
+- 110 certbot timer 失敗追查：`grist.wooo.work` / `registry.wooo.work` public route 目前被導向 `aiops.wooo.work`，HTTP-01 無法從 110 成功；已將兩個 stale renewal config 移至 `/etc/letsencrypt/renewal-disabled-codex-*`，並 reset certbot failed state。憑證 archive 未刪除；後續需修 public route 或改 DNS-01。
+- `scripts/reboot-recovery/full-stack-cold-start-check.sh` 新增 `P2-SCHEDULES`，覆蓋 188/110/120/121 cron、textfile mtime、188 backup freshness、110 failed units、K8s CronJob/Job/Pod 狀態、121 DR drill cron。
+- `docs/runbooks/FULL-STACK-COLD-START-SOP.md` 新增排程驗證章節與 done criteria，要求排程真正可執行才算 reboot recovery 完成。
+
+**最終驗證**：
+- KM reembed 完成：`1774/1774` success、`0` failed；DB 目前 `knowledge_entries` total `1785`、embedded `1776`、vector dims `1024..1024`，舊 embedding backup `1691` rows。
+- 手動 `km-vectorize` CronJob `km-vectorize-codex-220715` 完成，回 `embed-all: 200 {"total":0,"success":0,"failed":0}`。
+- `bash scripts/reboot-recovery/full-stack-cold-start-check.sh --send-alert-test` → `PASS=50 WARN=0 BLOCKED=0`，包含 Alertmanager webhook E2E、public routes、cron/CronJob/textfile/systemd schedule checks。
+- Prometheus firing alerts 已從 `HostBackupFailed + FlywheelExecutionRateMissing` 收斂為僅剩 `FlywheelExecutionRateMissing`；HostBackupFailed 解除。
+- 188/110 負載回到低檔；K3s node CPU 約 3-6%，KM reembed 未造成主機過載。
+
+**下一步**：
+- 將本次 runtime hotfix 對應的 repo changes 走正式 deploy，避免下一版 image 覆蓋 hotfix。
+- 修 `grist.wooo.work` / `registry.wooo.work` public route 或改 DNS-01 renewal；目前舊 renewal config 已停用以避免 certbot timer 每次失敗。
+
+## 2026-05-05 | 110 Sentry resource limits persistence gap closed
+
+**背景**：110 guardrail 告警已清，但主機 load 仍有長尾；統帥擔心 Claude Code 只做 live `docker update`，重建後配置又失效。
+
+**現場結論**：
+- 188 已回穩：load 約 `2.26 / 2.84 / 3.21`，momo/litellm/SignOz 核心容器都有 live CPU/memory guardrail；仍有 `HostBackupFailed`，但與 CPU/load 無關。
+- 110 仍是 Sentry 長尾，不是 runner 或 momo 類事故：ClickHouse 約 2.2-3.0 cores，Kafka 約 0.6 core，taskworker/taskbroker/taskscheduler/redis/uptime-checker 合計形成背景 load。
+- ClickHouse 目前不是查詢卡死：`system.processes` 無長查詢，`system.mutations` 無 pending，`system.merges` 只看到短 transaction merge；最大資料表是 `eap_items_1_local` 約 `6.68 GiB`。
+- Kafka consumer lag 查詢未見 backlog 膨脹；目前不應再靠降低 ClickHouse/Kafka memory 或泛用 restart。
+- 真正缺口：110 live limit 已存在，但 `/opt/sentry/docker-compose.yml` 只持久化了 `process-spans`；ClickHouse/Kafka/taskworker/taskbroker/taskscheduler/redis 一旦 compose recreate 可能回到 unlimited。
+
+**本次 live 修補**：
+- 110 `/opt/sentry/docker-compose.yml` 已備份為 `docker-compose.yml.bak-20260505-155707-codex-resource-limits`。
+- 持久化 Sentry 核心 guardrail：ClickHouse `2 CPU / 8 GiB / 16 GiB swap`、Kafka `2 CPU / 3 GiB / 6 GiB swap`、taskworker `2 CPU / 2 GiB / 4 GiB swap`、taskbroker `1 CPU / 512 MiB / 1 GiB swap`、taskscheduler `0.5 CPU / 512 MiB / 1 GiB swap`、redis `0.5 CPU / 512 MiB / 1 GiB swap`、uptime-checker `0.5 CPU / 512 MiB / 1 GiB swap`。
+- 只對 uptime-checker 補 live `docker update`，未重啟 Sentry/ClickHouse/Kafka；容器仍 `Up 5 days`。
+- 110 `/opt/sentry/clickhouse/config.xml` 已備份為 `config.xml.bak-20260505-160120-codex-merge-pool4`；ClickHouse 背景 merge 從 pool `8` 降到 `4`，三門檻從 `6/4/6` 降到 `3/2/3`，`max_bytes_to_merge_at_max_space_in_pool` 從 `512MiB` 降到 `256MiB`。
+- `SYSTEM RELOAD CONFIG` 不會熱套用這些 ClickHouse 25.3 設定，因此只重啟 `sentry-self-hosted-clickhouse-1`；重啟前 active foreground processes `1`（查詢本身）、pending mutations `0`。
+
+**驗證**：
+- `/opt/sentry/docker-compose.yml` `docker compose config` passed（僅 upstream `version` obsolete warning）。
+- `docker inspect` 顯示 ClickHouse/Kafka/taskworker/taskbroker/taskscheduler/redis/uptime-checker live limit 全部與 compose baseline 一致。
+- 110 load 從約 `12.50 / 13.10 / 13.35` 降到 `7.41 / 10.60 / 12.35`；`HostLoadAverageSustainedHigh` 未 firing，`DockerContainerCpuSustainedHigh` 僅 pending 於 Sentry ClickHouse。
+- ClickHouse 重啟後 16 秒 healthy；runtime setting 已確認 `background_pool_size=4`、三門檻 `3/2/3`、merge 上限 `268435456` bytes；active merges `0`、pending mutations `0`、ClickHouse CPU 約從 `2.1-2.7 cores` 降到 `0.67 core`。
+- 因 4 條 merge thread 仍可讓 ClickHouse 短暫回到 2.7 cores，將 live + compose CPU quota 從 `4` 收到 `2`，記憶體維持 `8 GiB`；後續 topk 顯示 ClickHouse 約 `2.0 cores`，由 CPU quota 保護 host。
+- 後續 host `ps` 顯示剩餘 `HostHighCpuLoad` 主因之一是 CD Web image build：`node /app/.../next build` 約 `1.4 cores`，疊加 Gitea/ClickHouse/Kafka；已在 `apps/web/Dockerfile` 加 `NEXT_PRIVATE_BUILD_WORKER_COUNT=1`，並將 `pnpm turbo build --filter=@awoooi/web` 改為 `--concurrency=1`，避免 Web build 再把 110 推到長時間高 CPU。
+- 舊 `HostHighCpuLoad` 從 `CPU >80% for 5m` 調成 `CPU >90% for 10m` 的早期 warning；真正長時間過載/自動診斷交給 `HostLoadAverageSustainedHigh` 的 `load5/core >1.5 for 15m`。
+- Prometheus firing alerts 只剩 `FlywheelExecutionRateMissing` 與 188 `HostBackupFailed`；Docker/runner guardrail alerts clean。
+
+**下一步**：
+- 110 若 ClickHouse sustained CPU 仍 pending 超過 drain window，下一步查 EAP/profiling/replay/uptime 是否需要保留；不要先降 ClickHouse memory 或重啟。
+- 將其他 unlimited 低流量容器分批納入 baseline，不一次全量加，避免把 Sentry/Harbor/monitoring 次要服務壓出新事故。
+- 188 優先修 `HostBackupFailed` 與 momo scheduler Google Drive/白頁檢查雜訊，CPU/load 不是當前阻塞。
+
 ## 2026-05-05 | 110/188 CPU/Mem 配額全景盤點 + Docker baseline 監控落地
 
 **背景**：統帥擔心 Claude Code 對 110/188 服務 CPU/memory limit 亂配置，造成服務卡死或慢性過載；本輪接續盤點 live Docker inspect / docker stats / compose 宣告。
@@ -3033,3 +3098,130 @@ C1（evolver 加 YAML_RULE guard）+ C2（seeder SQL `AND status != 'deprecated'
 ```bash
 psql $DATABASE_URL -f apps/api/migrations/cleanup_duplicate_deprecated_playbooks.sql
 ```
+
+---
+
+## 2026-05-05（台北）— 四主機重開機後全站冷啟動救援
+
+**觸發**：110 / 120 / 121 / 188 同時重開機後，多數服務異常；統帥要求先恢復所有網站、主機、核心服務，並建立完整冷啟動 SOP。
+
+### 已恢復
+
+| 範圍 | 結果 |
+|------|------|
+| 188 host PostgreSQL | WAL checkpoint 損壞；已備份後 `pg_resetwal`，`k3s_datastore` `REINDEX` + `VACUUM ANALYZE` 完成 |
+| K3s datastore | 刪除並備份可重建的腐壞 HPA / VPA / VPA checkpoint / `mon1` node rows；120 / 121 重新 Ready |
+| AWOOI prod | `awoooi-api` / `awoooi-web` / `awoooi-worker` Running；VIP `192.168.0.125` 內網驗證 API 200 / Web 307 |
+| mo.wooo.work | `momo-db` WAL redo 損壞；備份後 `pg_resetwal`，`momo-pro-system` / scheduler / bot / DB 全部 healthy；公網 `/` 200、`/health` 200 |
+| 110 host overload | actions runner units 維持最後放行；Sentry ClickHouse/Kafka 已從 dirty-reboot 損壞中恢復，Sentry stack healthy |
+| 188 SignOz | SignOz ClickHouse volume 出現 filesystem corruption；已 clean-clone 可讀資料並保留原始 corrupt volume，SignOz HTTP 恢復 |
+| 冷啟動 SOP | 新增 `docs/runbooks/FULL-STACK-COLD-START-SOP.md` 與 `scripts/reboot-recovery/full-stack-cold-start-check.sh` |
+
+### 驗證
+
+```bash
+bash scripts/reboot-recovery/full-stack-cold-start-check.sh --send-alert-test
+# PASS=31 WARN=0 BLOCKED=0
+# Result: GREEN. Full stack is ready for controlled runner/CD release.
+```
+
+### Dirty reboot 資料保全
+
+- 110 Sentry ClickHouse：原始壞 volume 保留為 `/var/lib/docker/volumes/sentry-clickhouse/_data.corrupt-20260505-203346`；以 clean-clone 恢復可讀資料並加 `force_restore_data`。
+- 110 Sentry Kafka：malformed checkpoint 已備份至 `/var/backups/sentry-kafka-checkpoints-20260505-203942`，只重建 checkpoint，不刪 topic/log data。
+- 188 SignOz ClickHouse：原始壞 volume 保留為 `/var/lib/docker/volumes/signoz-clickhouse/_data.corrupt-20260505-203735`；以 clean-clone 恢復可讀資料。
+- 188 `momo-db`：WAL reset 前備份 `/var/backups/postgresql/momo-db-before-pg-resetwal-20260505-200834.tgz`。
+
+### 已知隔離 / 後續
+
+- 110 actions runner units 仍按策略最後放行：guardrail 已套用，`CPUQuota=200%`、`MemoryMax=2G`、`WatchdogUSec=0`；需在 load/core 穩定後逐步開啟。
+- `Bad message` / `Structure needs cleaning` 是 host filesystem 層訊號；線上 clean-clone 已恢復服務，但完整歷史資料追溯需安排離線 `fsck` 或備份驗證。
+- `drift-scanner-29633040-qrf8w` 為單次 CronJob Error，不阻斷主服務；後續可清理或調查。
+
+---
+
+## 2026-05-05（台北）— GCP Ollama 告警路徑止血與內網化決策
+
+**觸發**：告警卡仍顯示 `Router: Gemini`，且 GCP-A / GCP-B Ollama 先前在告警 JSON prompt 上連續 504，導致 Gemini 備援產生費用。
+
+### 已執行
+
+| 範圍 | 結果 |
+|------|------|
+| 告警模型 | 將告警專用 Ollama 模型固定為 `gemma3:4b`，避免 `qwen3:14b` / `qwen2.5-coder:32b` 冷啟動拖入 Gemini |
+| Production image | `awoooi-api` / `awoooi-worker` 已手動切到 `192.168.0.110:5000/awoooi/api:787acd3bda918f53b977f37133e0b5c73558033e` |
+| Production env | 已明確設定 `ALERT_AI_ENFORCE_OLLAMA_FIRST=true`、`ALERT_AI_ALLOW_CLOUD_FALLBACK=true`、`ALERT_OLLAMA_MODEL=gemma3:4b` |
+| GCP Ollama 保溫 | GCP-A / GCP-B 已卸載 14B / 32B 重模型，並以 `keep_alive=8h` 保溫 `gemma3:4b` |
+| Meta W-6 降噪 | Trust Drift 未達 20% 時不再升級為 Meta System；現場 Redis 已加 6h dedup 防止重複通知 |
+
+### 現場驗證
+
+```bash
+kubectl -n awoooi-prod get deploy awoooi-api awoooi-worker -o jsonpath='{range .items[*]}{.metadata.name}{"\t"}{range .spec.template.spec.containers[*]}{.name}={.image}{" "}{end}{"\n"}{end}'
+# awoooi-api    api=192.168.0.110:5000/awoooi/api:787acd3bda918f53b977f37133e0b5c73558033e
+# awoooi-worker worker=192.168.0.110:5000/awoooi/api:787acd3bda918f53b977f37133e0b5c73558033e
+
+kubectl -n awoooi-prod exec deploy/awoooi-api -- printenv | grep -E 'ALERT_OLLAMA_MODEL|ALERT_AI_|OLLAMA_.*URL'
+# ALERT_OLLAMA_MODEL=gemma3:4b
+# ALERT_AI_ALLOW_CLOUD_FALLBACK=true
+# ALERT_AI_ENFORCE_OLLAMA_FIRST=true
+# OLLAMA_URL=http://192.168.0.110:11435
+# OLLAMA_SECONDARY_URL=http://192.168.0.110:11436
+# OLLAMA_FALLBACK_URL=http://192.168.0.111:11434
+```
+
+### 架構決策
+
+- 目前 `192.168.0.110:11435/11436` 是經由 110 nginx 轉發到 GCP 公網 IP，屬於過渡方案，不應作為長期 primary Ollama lane。
+- 建議建立 WireGuard site-to-site private mesh，讓 K3s / 110 / 111 / GCP-A / GCP-B 以私網 IP 互連，Ollama 僅綁定 mesh interface，並由 AwoooP Inference Gateway 統一路由、熔斷、佇列與模型保溫。
+- 注意：目前 GCP-A / GCP-B `/api/ps` 顯示 `size_vram: 0`，內網化可解決連線與安全問題，但無法讓 CPU-only GCP 等同 111 的 VRAM/GPU 效能；大模型應留在 111 或改用 GPU 型 GCP 節點。
+
+---
+
+## 2026-05-06（台北）— AwoooP Operator Console 與飛輪 KPI 對齊
+
+**觸發**：00:30 系統報告顯示「全系統正常」，但飛輪狀態為 `修復 0/15 (0%)`，使用者指出 AI 自動化幾乎沒有做；同步要求 AwoooP 工作項目必須與前端頁面、邏輯、操作面對齊。
+
+### 已修正
+
+| 範圍 | 結果 |
+|------|------|
+| 心跳報告 | `HeartbeatReportService._get_flywheel_stats()` 改讀 `auto_repair_executions`，不再用已失準的 `incidents.outcome` 推估修復率 |
+| 飛輪 Prometheus KPI | `FlywheelStatsService._playbook_stats()` 優先以 `auto_repair_executions` 計算 24h execution success rate，Redis playbook counter 僅作 fallback |
+| AI Success | `MetricsDBRepository` 改用 `UPPER(status::text)` 對齊實際 `APPROVED / EXECUTION_SUCCESS / EXECUTION_FAILED` 狀態值 |
+| Auto-repair metric | `AutoRepairService.execute_auto_repair()` 成功/失敗都呼叫 `record_auto_repair()`，修正 Prometheus 指標零 caller 問題 |
+| K8s Pod 報告 | Completed/Succeeded CronJob pod 不再顯示為紅色失敗；Telegram 報告會顯示 phase |
+| AwoooP 前端 | `/zh-TW/awooop` redirect 修正，Console 接入主 `AppLayout` 與 sidebar；新增 `工作鏈路` 頁映射 P0/P1/P2 工作項目、source of truth、gate 與操作面 |
+| AwoooP API | `GET /api/v1/platform/approvals?run_id=` 支援 M8 詳情頁查單筆 waiting approval |
+
+### 驗證
+
+```bash
+DATABASE_URL='postgresql+asyncpg://test:test@localhost:5432/test' \
+  apps/api/.venv/bin/python -m py_compile \
+  apps/api/src/repositories/metrics_repository.py \
+  apps/api/src/services/heartbeat_report_service.py \
+  apps/api/src/services/auto_repair_service.py \
+  apps/api/src/services/flywheel_stats_service.py \
+  apps/api/src/api/v1/platform/operator_runs.py \
+  apps/api/src/services/platform_operator_service.py
+
+DATABASE_URL='postgresql+asyncpg://test:test@localhost:5432/test' \
+  apps/api/.venv/bin/python -m ruff check --select E9,F401,F821 \
+  apps/api/src/repositories/metrics_repository.py \
+  apps/api/src/services/heartbeat_report_service.py \
+  apps/api/src/services/auto_repair_service.py \
+  apps/api/src/services/flywheel_stats_service.py \
+  apps/api/src/api/v1/platform/operator_runs.py \
+  apps/api/src/services/platform_operator_service.py
+# All checks passed!
+
+pnpm --filter @awoooi/web typecheck
+# tsc --noEmit passed
+```
+
+### 後續
+
+- 仍需處理 `approval_records.matched_playbook_id = NULL` 問題，否則執行結果無法完整回寫 Playbook trust。
+- 仍需攔截 AI action hallucination（alertname 被當 deployment/host、namespace 亂填）進入 approval 前的路徑。
+- AwoooP Console 下一步應接入真實 run step journal / trace view，而不是只列 run state。

docs/runbooks/FULL-STACK-COLD-START-SOP.md

@@ -0,0 +1,497 @@
+# AWOOOI Full-Stack Cold Start SOP
+
+> Version: v1.0
+> Last updated: 2026-05-05 Asia/Taipei
+> Scope: 110 / 120 / 121 / 188 full-stack reboot recovery. 112 Kali is recorded as P3 optional and is not part of this recovery path.
+
+---
+
+## 0. When To Use This
+
+Use this SOP when any of these happen:
+
+- 110/120/121/188 reboot unexpectedly.
+- All services are abnormal after a power/network event.
+- K3s is stuck `activating`.
+- Host load remains high during startup and service health is mixed.
+- Monitoring, alerting, CD, AI auto-repair, and Docker Compose services disagree about the real state.
+
+The rule is simple: **recover the dependency chain, not the loudest symptom.**
+
+---
+
+## 1. Golden Startup Order
+
+```text
+0. Freeze automation and preserve evidence
+1. Physical/network layer
+2. 188 data layer
+3. 110 registry/observability layer
+4. 120/121 K3s layer
+5. AWOOOI workload layer
+6. Public routes and alert chain
+7. High-load batch/consumer/crawler services
+8. Runner/CD
+9. AI auto-remediation
+10. 112 Kali scanner, if needed
+```
+
+Never start runner/CD before 188 PostgreSQL, 110 Harbor, K3s nodes, and AWOOOI API are healthy.
+
+---
+
+## 2. Automation Freeze
+
+Cold start creates noisy metrics and partial failures. During P0/P1, keep automation in observe-only mode.
+
+| Item | Cold-start policy | Reason |
+|------|-------------------|--------|
+| Gitea/GitHub runners | Last | Build jobs can saturate 110 CPU/RAM. |
+| momo-scheduler / crawlers | Last | Chrome and batch work can saturate 188. |
+| Sentry/Snuba consumers | Controlled | Kafka backlog and ClickHouse merge can create temporary high load. |
+| Alertmanager outbound notification | Gate | Avoid alert storms before API webhook and Telegram are verified. |
+| AI auto-repair | Observe-only | Metrics, Redis, KM, and playbooks may be incomplete. |
+| Stateful DB restart | Human approval | PostgreSQL, Redis, ClickHouse, Harbor DB, Sentry DB are not generic restart targets. |
+
+---
+
+## 3. P0 Evidence And Network
+
+Run from any machine on the same LAN:
+
+```bash
+for h in 110 120 121 188; do
+  ping -c 2 -W 2 192.168.0.$h >/dev/null && echo "PING_OK 192.168.0.$h" || echo "PING_FAIL 192.168.0.$h"
+done
+
+arp -an | grep -E '192\.168\.0\.(110|120|121|188)'
+for h in 110 120 121 188; do
+  nc -G 3 -z 192.168.0.$h 22 && echo "SSH_OK 192.168.0.$h" || echo "SSH_FAIL 192.168.0.$h"
+done
+```
+
+Then capture reboot evidence:
+
+```bash
+ssh ollama@192.168.0.188 'hostname; date; uptime; who -b; last -x reboot shutdown | head -20'
+ssh wooo@192.168.0.110 'hostname; date; uptime; who -b; last -x reboot shutdown | head -20'
+ssh wooo@192.168.0.120 'hostname; date; uptime; who -b; last -x reboot shutdown | head -20'
+ssh wooo@192.168.0.121 'hostname; date; uptime; who -b; last -x reboot shutdown | head -20'
+```
+
+If any host has ARP `incomplete` or SSH port down, stop here and fix physical/network first.
+
+---
+
+## 4. P0 188 Data Layer
+
+188 is the first real service dependency because K3s datastore and AWOOOI DB depend on PostgreSQL.
+
+### 4.1 Startup order
+
+1. `containerd`
+2. `docker`
+3. `postgresql@14-main`
+4. `k3s_datastore.kine` maintenance
+5. `redis-server` on `6380`
+6. `ollama` or current AI proxy dependencies
+7. `nginx`
+8. Docker networks
+9. MinIO / OpenClaw / SignOz
+10. momo / litellm / batch services after load is stable
+
+### 4.2 Read-only check
+
+```bash
+ssh ollama@192.168.0.188 '
+hostname; date; uptime; free -h
+systemctl is-active containerd docker postgresql@14-main redis-server ollama nginx || true
+pg_isready -h localhost -p 5432 || true
+redis-cli -p 6380 ping 2>/dev/null || redis-cli ping 2>/dev/null || true
+docker ps --format "{{.Names}}\t{{.Status}}\t{{.Ports}}" | head -120
+'
+```
+
+### 4.3 PostgreSQL WAL checkpoint damage
+
+Signature:
+
+```text
+PANIC: could not locate a valid checkpoint record
+invalid primary checkpoint record
+unexpected pageaddr ... in log segment ...
+```
+
+This blocks:
+
+- `188:5432`
+- K3s startup on 120/121
+- AWOOOI API DB access
+- Alertmanager webhook if API cannot start
+
+Human-approved recovery command on 188:
+
+```bash
+sudo systemctl stop postgresql@14-main
+sudo install -d -m 700 -o postgres -g postgres /var/backups/postgresql
+sudo tar -C /var/lib/postgresql/14 -czf /var/backups/postgresql/14-main-before-pg-resetwal-$(date +%Y%m%d-%H%M%S).tgz main
+sudo -u postgres /usr/lib/postgresql/14/bin/pg_resetwal -f /var/lib/postgresql/14/main
+sudo systemctl start postgresql@14-main
+pg_isready -h localhost -p 5432
+sudo -u postgres psql -d k3s_datastore -c "VACUUM ANALYZE kine;"
+```
+
+Do not run `DROP`, reinitialize the cluster, delete `/var/lib/postgresql`, or restore an old backup unless the commander explicitly approves it.
+
+---
+
+## 5. P0/P1 110 Registry And Observability
+
+110 must recover Harbor/Gitea/Monitoring early, but runners last.
+
+### 5.1 Startup order
+
+1. `docker`
+2. Remove `Exited (128)` / `Exited (137)` orphan containers
+3. Harbor `harbor-log`
+4. Harbor full stack
+5. Gitea
+6. Prometheus / Alertmanager / Grafana / exporters
+7. Langfuse
+8. SignOz
+9. Sentry DB layer
+10. Sentry web/worker/consumer layer
+11. Gitea host runner and actions runners
+
+### 5.2 Checks
+
+```bash
+ssh wooo@192.168.0.110 '
+hostname; date; uptime; free -h
+systemctl is-active docker || true
+curl -s -o /dev/null -w "harbor=%{http_code}\n" --max-time 5 http://127.0.0.1:5000/v2/ || true
+curl -s -o /dev/null -w "gitea=%{http_code}\n" --max-time 5 http://127.0.0.1:3001/ || true
+curl -s --max-time 5 http://127.0.0.1:9090/-/ready || true
+curl -s --max-time 5 http://127.0.0.1:9093/-/healthy || true
+curl -s -o /dev/null -w "sentry=%{http_code}\n" --max-time 10 http://127.0.0.1:9000/ || true
+docker ps --format "{{.Names}}\t{{.Status}}" | head -120
+'
+```
+
+Harbor healthy means `/v2/` returns `200` or `401`. Do not treat `401` as failure.
+
+### 5.3 Runner gate
+
+Runner may start only after all are true:
+
+- `188 PostgreSQL` ready
+- `110 Harbor` ready
+- `110 Gitea` ready
+- `120/121 K3s` nodes ready
+- AWOOOI API health passes
+- 110 load/core is below `1.0` for at least 15 minutes
+- runner systemd guardrails are active: `CPUQuota=200%`, `MemoryMax=2G`, `WatchdogUSec=0`
+
+Check:
+
+```bash
+ssh wooo@192.168.0.110 '
+for u in $(systemctl list-units "actions.runner.*" --all --no-legend --plain | awk "{print \$1}"); do
+  echo "=== $u ==="
+  systemctl show "$u" -p ActiveState -p SubState -p CPUQuotaPerSecUSec -p MemoryMax -p WatchdogUSec -p NRestarts
+done
+'
+```
+
+If `WatchdogUSec` is not `0`, apply the guardrail script manually with sudo:
+
+```bash
+sudo /home/wooo/scripts/apply-runner-systemd-guardrails.sh --apply
+```
+
+---
+
+## 6. P1 120/121 K3s
+
+K3s must wait for 188 PostgreSQL and 110 Harbor.
+
+### 6.1 Startup order
+
+1. 120 `k3s.service`
+2. 121 `k3s-agent.service` or its live role
+3. CNI / kube-proxy
+4. Nodes Ready
+5. Core pods
+6. `awoooi-prod` pods
+7. keepalived VIP `192.168.0.125`
+8. NodePorts `32334` and `32335`
+
+### 6.2 Checks
+
+```bash
+ssh wooo@192.168.0.120 '
+hostname; uptime
+pg_isready -h 192.168.0.188 -p 5432 || true
+systemctl is-active k3s k3s-agent keepalived 2>/dev/null || true
+kubectl get nodes -o wide 2>/dev/null || true
+kubectl get pods -A 2>/dev/null | grep -v -E "Running|Completed" || true
+kubectl get pods -n awoooi-prod -o wide 2>/dev/null || true
+ip addr show | grep 192.168.0.125 || true
+'
+
+ssh wooo@192.168.0.121 '
+hostname; uptime
+systemctl is-active k3s k3s-agent keepalived 2>/dev/null || true
+ip addr show | grep 192.168.0.125 || true
+'
+```
+
+If K3s is `activating` while 188 PostgreSQL is down, fix PostgreSQL first. Restarting K3s repeatedly will not solve it.
+
+---
+
+## 7. P2 AWOOOI Workloads
+
+Run after K3s nodes are Ready:
+
+```bash
+ssh wooo@192.168.0.120 '
+kubectl get deploy -n awoooi-prod
+kubectl get pods -n awoooi-prod -o wide
+kubectl get svc -n awoooi-prod
+kubectl get events -n awoooi-prod --sort-by=.lastTimestamp | tail -40
+'
+
+curl -s --max-time 8 http://192.168.0.125:32334/api/v1/health
+curl -s -o /dev/null -w "web=%{http_code}\n" --max-time 8 http://192.168.0.125:32335/
+```
+
+If pods are `ImagePullBackOff`, go back to 110 Harbor.
+
+If API health fails because DB/Redis is down, go back to 188.
+
+---
+
+## 8. P2 Alert Chain
+
+Current main path:
+
+```text
+Prometheus/Alertmanager on 110
+  -> http://192.168.0.125:32334/api/v1/webhooks/alertmanager
+  -> AWOOOI API
+  -> TelegramGateway
+  -> Telegram
+```
+
+Alertmanager health alone is not enough. Run E2E:
+
+```bash
+curl -s -X POST http://192.168.0.125:32334/api/v1/webhooks/alertmanager \
+  -H 'Content-Type: application/json' \
+  -d '{"receiver":"cold-start-test","status":"firing","alerts":[{"status":"firing","labels":{"alertname":"ColdStartE2ETest","severity":"info"},"annotations":{"summary":"Cold start E2E test, ignore"},"startsAt":"2026-05-05T11:00:00Z","endsAt":"0001-01-01T00:00:00Z","generatorURL":""}],"groupLabels":{},"commonLabels":{},"commonAnnotations":{},"externalURL":"","version":"4","groupKey":"cold-start-test"}'
+```
+
+Expected: API returns success and Telegram receives the test alert.
+
+---
+
+## 9. P2 Schedules And Delayed Work
+
+Do not mark the reboot complete until scheduled work is proven runnable. A container can be healthy while its cron path is broken.
+
+| Host / Layer | Required check | Success baseline |
+|--------------|----------------|------------------|
+| 188 cron | `systemctl is-active cron` and `crontab -l` | cron active; backup, restart exporter, stats exporter entries present |
+| 188 backup-from-110 | `backup_110_last_success_timestamp` in textfile/Prometheus | last success age `< 25h` |
+| 188 momo-scheduler | `docker logs momo-scheduler` | `全部排程任務已註冊`; Google Drive auth works; dashboard URLs use container-reachable hostnames |
+| 188 momo import | manual `run_auto_import_task()` after parser changes | selected sheet is `即時業績明細`; imported date range has matching rows in `daily_sales_snapshot` and `realtime_sales_monthly` |
+| 110 cron | `systemctl is-active cron` | cron active; Docker/systemd textfile exporters fresh |
+| 110 startup units | `systemctl --failed` | zero failed units; stale `momo-startup-complete` and `wooo-staggered-startup` disabled |
+| 120 K8s CronJobs | `kubectl get cronjobs -n awoooi-prod` | unsuspended; no failed Jobs remain after current validation |
+| 121 DR drill | `crontab -l` | DR drill cron present unless explicitly paused |
+
+Useful checks:
+
+```bash
+ssh ollama@192.168.0.188 'systemctl is-active cron; crontab -l; ls -l /home/ollama/node_exporter_textfiles/*.prom'
+ssh wooo@192.168.0.110 'systemctl --failed --no-pager; systemctl is-active cron; crontab -l'
+ssh wooo@192.168.0.120 'sudo kubectl get cronjobs,jobs -n awoooi-prod'
+ssh wooo@192.168.0.121 'systemctl is-active cron; crontab -l'
+```
+
+If a schedule succeeds but emits a false verification alert, fix the verification rule before releasing AI auto-remediation. False positives train operators to ignore real alarms.
+
+---
+
+## 10. P2/P3 Stateful Service Guardrails
+
+| Tier | Examples | Automation |
+|------|----------|------------|
+| BLOCK | PostgreSQL data dir, ClickHouse data dir, Harbor DB, Sentry DB | No automatic destructive action. Human approval only. |
+| CRITICAL_HITL | Redis, Kafka, MinIO, SignOz ClickHouse, Sentry ClickHouse | Human-in-the-loop restart/repair. |
+| STANDARD_HITL | API/Web/worker, OpenClaw, litellm | Restart only with evidence and blast-radius check. |
+| AUTO | Stateless exporters, blackbox, nginx exporter | Auto restart allowed after verification. |
+
+Never use generic `docker restart $(docker ps -q)` during cold start.
+
+### 10.1 Dirty-Reboot Storage Corruption
+
+Treat these log signatures as storage corruption, not ordinary service flakiness:
+
+- `Bad message`
+- `Structure needs cleaning`
+- `Unknown codec`
+- `PANIC: could not locate a valid checkpoint record`
+- Kafka `Malformed line` in checkpoint files
+- ClickHouse `broken and needs manual correction`
+
+Cold-start automation may stop a restart storm and collect evidence, but it must not delete the original data directory. If a filesystem returns `Bad message` or `Structure needs cleaning`, the real root cause is below the container layer. Online recovery can restore service from readable data, but complete historical recovery requires an offline filesystem check or backup restore.
+
+### 10.2 ClickHouse Clean-Clone Recovery Pattern
+
+Use this pattern for Sentry ClickHouse or SignOz ClickHouse when individual corrupted parts cannot be moved because the host filesystem rejects reads.
+
+```text
+1. Stop the compose stack or at least stop dependent consumers.
+2. Disable restart loops for the failing container.
+3. Save logs and build an exclude list from unreadable store paths.
+4. Preserve the original volume as _data.corrupt-YYYYMMDD-HHMMSS.
+5. Create a clean _data clone with readable files only.
+6. Add flags/force_restore_data.
+7. Start ClickHouse first, then web/API, then consumers.
+8. Verify HTTP, merge backlog, and restart count before releasing high-load services.
+```
+
+Do not replace this with `rm -rf store/...` unless the unreadable path is already backed up or the commander explicitly accepts data loss. The preferred incident artifact is:
+
+```text
+/var/lib/docker/volumes/<volume>/_data.corrupt-YYYYMMDD-HHMMSS
+/var/backups/<service>-<component>-YYYYMMDD-HHMMSS
+```
+
+### 10.3 Kafka Checkpoint Recovery Pattern
+
+If Kafka refuses to start with malformed checkpoint files after a dirty reboot, preserve and move only checkpoint files:
+
+```text
+log-start-offset-checkpoint
+recovery-point-offset-checkpoint
+replication-offset-checkpoint
+```
+
+Then start Kafka and confirm health before starting Snuba/Sentry consumers. Do not delete topic directories or Kafka logs during cold-start recovery.
+
+---
+
+## 11. P3 High-Load Services
+
+Only release these after P0/P1/P2 gates are green:
+
+| Host | Service | Release condition |
+|------|---------|-------------------|
+| 188 | momo-scheduler / crawler | load/core < 1.0 for 15 minutes and DB healthy |
+| 188 | SignOz ClickHouse | healthy and merge backlog trending down |
+| 188 | litellm | `/health/liveliness` good and provider route verified |
+| 110 | Sentry Snuba consumers | ClickHouse healthy and Kafka backlog decreasing |
+| 110 | Sentry uptime-checker | Sentry web/DB healthy |
+| 110 | runners | all previous gates green and load/core < 1.0 for 15 minutes |
+
+---
+
+## 12. Baseline And AI Auto-Remediation Gate
+
+### 12.1 Stable Runtime Baseline
+
+These are release gates after the first cold-start recovery pass:
+
+| Area | Baseline |
+|------|----------|
+| 188 host | PostgreSQL accepting, Redis PONG, momo `/health` 200, SignOz HTTP reachable, load/core < 1.0 sustained before crawlers |
+| 110 host | Harbor `/v2/` 200/401, Gitea 200/302, Prometheus ready, Alertmanager healthy, Sentry HTTP 200/302/400, no ClickHouse/Kafka restart loop |
+| K3s | 120/121 nodes Ready, VIP `192.168.0.125` present, AWOOOI API 2xx/3xx, Web 2xx/3xx |
+| Public routes | `https://awoooi.wooo.work/api/v1/health` 2xx/3xx, `https://mo.wooo.work/health` 2xx/3xx |
+| Guardrails | Docker/systemd textfile exporters fresh, runner `CPUQuota=200%`, `MemoryMax=2G`, `WatchdogUSec=0` |
+| Schedules | cron active on 110/188/120/121; K8s CronJobs unsuspended; no current failed Jobs; 188 backup success `< 25h` |
+| Backlog | ClickHouse merges and Kafka/Snuba lag trending down, not increasing for two consecutive checks |
+
+If service health is green but load average remains high, check live CPU and IO before changing memory limits. High load after Sentry/Snuba or ClickHouse startup can be backlog drain; high CPU from runners/builds/crawlers is a release-order problem.
+
+### 12.2 AI Auto-Remediation Gate
+
+AI auto-repair can move from observe-only to limited execution only after:
+
+- Prometheus rules are loaded.
+- docker/systemd textfile exporter files are fresh.
+- blackbox probes have stable results.
+- cron/CronJob schedule checks are green.
+- AWOOOI API `/api/v1/health` passes.
+- Alertmanager E2E webhook passes.
+- Redis/KM/playbook health is available.
+- No active restart storm.
+- Host load/core remains below `1.0` for 15 minutes.
+
+Until then:
+
+- diagnose only
+- notify only
+- require human approval for remediation
+- no DB/ClickHouse/Harbor/Sentry destructive action
+- no generic restart action against stateful services
+
+---
+
+## 13. One-Command Readiness Script
+
+Run:
+
+```bash
+bash scripts/reboot-recovery/full-stack-cold-start-check.sh
+```
+
+The script is read-only. It reports gates:
+
+- `P0-NETWORK`
+- `P0-188-DATA`
+- `P0-110-REGISTRY`
+- `P1-K3S`
+- `P2-WORKLOAD`
+- `P2-ALERTCHAIN`
+- `P2-PUBLIC-ROUTES`
+- `P2-SCHEDULES`
+- runner guardrail state inside `P0-110-REGISTRY-OBSERVABILITY`
+
+If it prints `BLOCKED`, fix the first blocked gate before moving forward.
+
+---
+
+## 14. Done Criteria
+
+All must be true:
+
+- Four hosts reachable by SSH.
+- 188 PostgreSQL and Redis healthy.
+- 110 Harbor, Gitea, Prometheus, Alertmanager healthy.
+- 120/121 K3s nodes Ready.
+- VIP `192.168.0.125` present.
+- AWOOOI API and Web reachable through NodePort/VIP.
+- Alertmanager E2E webhook succeeds.
+- cron/CronJob schedules are active, unsuspended, and verified.
+- Sentry and SignOz are either healthy or explicitly in controlled backlog recovery.
+- High-load batch services are capped or delayed.
+- Runners are guarded and released last.
+- AI auto-remediation is not in full execution mode until all gates are green.
+
+---
+
+## 15. Known Drift To Fix After Recovery
+
+These must be cleaned after the incident, not during P0:
+
+- `SERVICE-ENDPOINTS.md` still has old Prometheus/Alertmanager locations.
+- Audit older docs for direct node webhook targets; current main path should be VIP `192.168.0.125:32334`.
+- OpenClaw `8088` vs `8089` must be live-confirmed and normalized.
+- 188 compose paths drift between `/home/ollama/*` and Ansible `/opt/*`.
+- 110 runner docs still mention Docker runner in places; live startup prefers host `gitea-act-runner-host.service`.
+- `scripts/setup-runner-watchdog.sh` conflicts with the 2026-05-05 runner watchdog disablement guardrail.
+- `grist.wooo.work` / `registry.wooo.work` public HTTP/HTTPS currently route to `aiops.wooo.work`; their old 110 certbot renewal configs are disabled until public routing is corrected or DNS-01 renewal is configured.

docs/runbooks/HOST-RESOURCE-BASELINE-110-188.md

@@ -9,11 +9,13 @@
 
 | Service | Live Limit | Live Usage Snapshot | Verdict |
 |---|---:|---:|---|
-| Sentry ClickHouse | 4 CPU / 8 GiB | ~235-291% CPU / 3.3-3.4 GiB | CPU capped but still hottest. Do not lower memory; keep merge settings explicit. |
+| Sentry ClickHouse | 2 CPU / 8 GiB, merge pool 4 | capped near 2 cores after pool 8 -> 4 restart | Do not lower memory. CPU quota intentionally slows background merge so Sentry cannot dominate 110. If backlog grows, inspect `MergeMutate` and Sentry high-volume features before raising it. |
 | Sentry Kafka | 2 CPU / 3 GiB | ~40-55% CPU / 2.5 GiB (84%) | Memory is close to pressure. Do not reduce memory. |
 | Sentry taskworker | 2 CPU / 2 GiB, concurrency 2 | ~120-181% CPU after restart | Concurrency reduced from 4 to 2 after Kafka lag cleared. Watch Sentry task latency before further changes. |
 | Sentry taskbroker | 1 CPU / 512 MiB | ~70-98% CPU / 160 MiB | CPU is tight; increasing may improve backlog but can raise host load. |
 | Sentry taskscheduler | 0.5 CPU / 512 MiB | ~13% CPU / 387 MiB (76%) | Memory is tight; alert at 85% before it stalls. |
+| Sentry redis | 0.5 CPU / 512 MiB | ~15-30% CPU / 19 MiB | Live and compose cap are aligned. |
+| Sentry uptime-checker | 0.5 CPU / 512 MiB | ~26-30% CPU / 43-187 MiB | Capped after it showed sustained background CPU. |
 | Gitea | 3 CPU / 3 GiB | ~4% CPU / 2.18 GiB (73%) | Good cap; memory headroom is not huge. |
 | GitHub/Gitea runners | unlimited systemd services | one runner had WatchdogSec=5min and 8,490 restarts; `act` CI containers caused load spikes | Must be monitored outside Docker. Remove bad watchdog drop-in and apply per-runner CPU/Memory quotas. |
 | node-exporter | 1 CPU / 256 MiB | ~0-5% CPU / 8 MiB | Good after disabling expensive `arp`, `netclass`, and `netdev` collectors. |
@@ -28,8 +30,11 @@
 | SignOz ClickHouse | 4 CPU / 24 GiB | ~93-133% CPU / 1.1 GiB | Healthy enough; keep current cap. |
 | SignOz Zookeeper | 1 CPU / 2 GiB | ~8-18% CPU / 1.09 GiB | OK. |
 | cadvisor | 1.5 CPU / 1 GiB | ~0% CPU / 28 MiB | Good. |
-| litellm | unlimited | ~0.6-0.9% CPU / 780 MiB | Add modest cap after observing traffic; do not re-add DATABASE_URL. |
-| momo-pro-system / momo-db | unlimited | DB had short CPU bursts, then ~0.6% with no active long query | Needs service-specific limits after scheduler/schema pressure is controlled. |
+| litellm | 1 CPU / 1 GiB | ~0.5-0.9% CPU / 780 MiB | Good cap; keep stateless mode and do not re-add `DATABASE_URL`. |
+| momo-pro-system | 2 CPU / 2 GiB | ~1-2% CPU / 740 MiB | Good cap; startup cache prewarm must stay single-flight. |
+| momo-scheduler | 2 CPU / 2 GiB | ~0.3% CPU / 105-163 MiB after crawler burst | CPU cap is working. Next fix is crawler concurrency and failed background jobs, not lower CPU. |
+| momo-telegram-bot | 0.5 CPU / 512 MiB | ~0.7% CPU / 66 MiB | Good cap. |
+| momo-db | 2 CPU / 4 GiB | DB had short CPU bursts, then ~0.6-29% with no active long query | Good cap; current bursts are query/workload, not limit pressure. |
 | Monitoring tools / websites / exporters | mostly unlimited | low | Add caps gradually with textfile alerts watching pressure. |
 
 ## Baseline Policy
@@ -69,12 +74,13 @@ Use these thresholds for alerting and AI triage:
 
 1. Deploy `scripts/ops/docker-stats-textfile-exporter.py` to 110 and 188 textfile collector cron.
 2. Reload Prometheus rules with the new Docker CPU/memory/restart baseline alerts.
-3. Observe 110 for one drain window after node-exporter collector trim and taskworker concurrency 2. Kafka lag is now near zero; if ClickHouse remains high, tune merge/query behavior, not Kafka consumers.
-4. Tune `momo-scheduler` crawler concurrency on 188; keep 2 CPU / 2 GiB until success rate and latency prove it is too low.
-5. Fix 188 Elephant Alpha/OpenClaw allowed-action drift before enabling resource auto-repair beyond diagnosis.
-6. Add modest caps to currently unlimited low-risk services in small batches.
-7. Deploy `scripts/ops/stop-stale-gitea-actions-jobs.sh` to 110 as `/home/wooo/scripts/stop-stale-gitea-actions-jobs.sh`; keep Prometheus auto action in dry-run mode.
-8. Fix 110 runner services with sudo-capable host maintenance:
+3. Persist live limits in the owning compose files before considering the host repaired; live `docker update` alone is not durable.
+4. Observe 110 for one drain window after node-exporter collector trim and taskworker concurrency 2. Kafka lag is now near zero; if ClickHouse remains high, tune merge/query behavior or reduce high-volume Sentry features, not Kafka memory.
+5. Tune `momo-scheduler` crawler concurrency on 188; keep 2 CPU / 2 GiB until success rate and latency prove it is too low.
+6. Fix 188 Elephant Alpha/OpenClaw allowed-action drift before enabling resource auto-repair beyond diagnosis.
+7. Add modest caps to currently unlimited low-risk services in small batches. Do not alert every unlimited auxiliary container at once; promote candidates only after 24h usage data.
+8. Deploy `scripts/ops/stop-stale-gitea-actions-jobs.sh` to 110 as `/home/wooo/scripts/stop-stale-gitea-actions-jobs.sh`; keep Prometheus auto action in dry-run mode.
+9. Fix 110 runner services with sudo-capable host maintenance:
 
 ```bash
 sudo /home/wooo/scripts/apply-runner-systemd-guardrails.sh --apply
@@ -88,3 +94,4 @@ sudo /home/wooo/scripts/apply-runner-systemd-guardrails.sh --apply
 - Treating "no alert" as healthy when cAdvisor or textfile exporters are missing.
 - Letting monitoring collectors spend seconds per scrape; this turns observability into load.
 - Leaving self-hosted runners unlimited on the same host as Sentry/ClickHouse/Gitea.
+- Applying live `docker update` without persisting the same guardrail in compose/systemd/IaC.

docs/runbooks/RUNBOOK-OLLAMA-FAILOVER.md

@@ -13,15 +13,15 @@ Dashboard 路徑：`Ollama 容災監控`（uid: `ollama-failover-p23`）
 
 ### Panel 1 — Ollama 可用性 (Stat)
 
-**看什麼**：`up{job=~"ollama_111|ollama_188"}` × 100，顯示每台 Ollama 主機的 scrape 存活狀態。
+**看什麼**：`up{job=~"ollama_gcp_a|ollama_gcp_b|ollama_local|ollama_111"}` × 100，顯示每個 Ollama provider endpoint 的 scrape 存活狀態。
 
 | 顏色 | 意義 |
 |------|------|
 | 綠色 100% | Prometheus 探測正常，主機在線 |
-| 黃色 50% | 一台離線，另一台在線（容災中） |
-| 紅色 0% | 兩台全離線，高風險 |
+| 黃色 | 部分 endpoint 離線，系統應進入容災 |
+| 紅色 0% | Ollama provider pool 全離線，高風險 |
 
-**注意**：此面板反映 Prometheus scrape 狀態，需要 scrape job 命名為 `ollama_111` / `ollama_188`。
+**注意**：此面板反映 Prometheus scrape 狀態，需要 scrape job 命名對齊 `ollama_gcp_a` / `ollama_gcp_b` / `ollama_local`。
 設定檔位於 `ops/monitoring/generated/prometheus-scrape-generated.yaml`。
 
 ---
@@ -47,9 +47,10 @@ Dashboard 路徑：`Ollama 容災監控`（uid: `ollama-failover-p23`）
 
 | 分布 | 意義 |
 |------|------|
-| ollama 佔 >90% | 正常，111 健康 |
-| gemini 佔多數 | 111 SLOW/DEGRADED/OFFLINE，容災中 |
-| ollama_188 出現 | Gemini 配額耗盡備援，或 111 和 Gemini 同時失敗 |
+| ollama / ollama_gcp_a 佔 >90% | 正常，GCP-A 健康 |
+| ollama_gcp_b 佔多數 | GCP-A SLOW/DEGRADED/OFFLINE，容災到 GCP-B |
+| ollama_local 出現 | GCP-A/B 均不可用，容災到 111 local |
+| gemini 佔多數 | Ollama provider pool 全部不可用，使用付費備援 |
 | 全部 nemotron/claude | 極端情況，所有主力 provider 失敗 |
 
 ---
@@ -71,10 +72,10 @@ Dashboard 路徑：`Ollama 容災監控`（uid: `ollama-failover-p23`）
 
 ### `OllamaInstanceDown` — Ollama 主機離線
 
-**觸發條件**：`up{job=~"ollama_111|ollama_188"} == 0` 持續 2 分鐘。
+**觸發條件**：`up{job=~"ollama_gcp_a|ollama_gcp_b|ollama_local|ollama_111"} == 0` 持續 2 分鐘。
 
 **影響評估**：
-- 系統應已自動切至 Gemini（查 Panel 3 確認）
+- 系統應已依序切至 GCP-B / 111 local / Gemini（查 Panel 3 確認）
 - 查 Panel 4 是否有 Failover 計數上升
 
 **排查步驟**：
@@ -82,11 +83,9 @@ Dashboard 路徑：`Ollama 容災監控`（uid: `ollama-failover-p23`）
 ```bash
 # 步驟 1：確認主機存活
 ping -c 3 192.168.0.111
-ping -c 3 192.168.0.188
 
 # 步驟 2：SSH 進主機確認 ollama 服務狀態
 ssh wooo@192.168.0.111 'systemctl status ollama'
-ssh wooo@192.168.0.188 'systemctl status ollama'
 
 # 步驟 3：查 ollama 最近的 journal log
 ssh wooo@192.168.0.111 'journalctl -u ollama -n 50 --no-pager'
@@ -210,8 +209,9 @@ ssh wooo@192.168.0.111 'systemctl status ollama && nvidia-smi'
 
 | Metric | 類型 | 狀態 | 說明 |
 |--------|------|------|------|
-| `up{job="ollama_111"}` | Gauge | ✅ 現有 | Prometheus scrape 存活 |
-| `up{job="ollama_188"}` | Gauge | ✅ 現有 | Prometheus scrape 存活 |
+| `up{job="ollama_gcp_a"}` | Gauge | ✅ 現有 | Prometheus scrape 存活 |
+| `up{job="ollama_gcp_b"}` | Gauge | ✅ 現有 | Prometheus scrape 存活 |
+| `up{job="ollama_local"}` | Gauge | ✅ 現有 | Prometheus scrape 存活 |
 | `ollama_failover_triggered_total` | Counter | ✅ P2.3 補入 | failover 切換次數，labels: from_provider, to_provider |
 | `ollama_recovery_triggered_total` | Counter | ✅ P2.3 補入 | recovery 切回次數，labels: from_provider |
 | `ollama_health_status{host}` | Gauge | ✅ P2.3 補入 | 健康狀態 1=healthy, 0=not_healthy |

k8s/awoooi-prod/04-configmap.yaml.patch-188-fallback

@@ -1,34 +0,0 @@
-# ============================================================================
-# PATCH: 188 CPU-only Ollama 備援端點
-# 日期: 2026-04-25 (台北時區)
-# 負責人: ogt + Claude Sonnet 4.6
-# ADR 參考: plan_complete_v3.md P0.5
-# 診斷實測數據:
-#   主機: 192.168.0.188, Intel Xeon Silver 4214 @ 2.2GHz, 12 核, CPU-only
-#   RAM: 62GB (used 14GB), Disk: 982GB (used 221GB)
-#   GPU: 無
-#   現有模型: qwen2.5:7b-instruct (4.5GB), llama3.2:3b (1.9GB),
-#            deepseek-r1:14b (8.5GB), nomic-embed-text (261MB)
-#   推理延遲實測: qwen2.5:7b-instruct → total=111s, eval_rate=0.09 token/s
-#                llama3.2:3b → total=155s (cold start, 比 7b 更慢)
-#   目標 ~30s 無法達到 (CPU 推理硬上限 ~0.09 token/s)
-#   決策: qwen2.5:7b-instruct 已存在，設為備援 (111s 延遲，使用者需知情)
-#   連通性: 110 → 188:11434 ✅ 已驗證
-# ⚠️  注意: 188 推理極慢(~111s)，應只在 111 GPU Ollama 完全失效時啟用
-#   建議: 程式碼層應設 OLLAMA_FALLBACK_188_TIMEOUT_SEC = 150
-# ============================================================================
-#
-# 將以下兩行加入 /Users/ogt/awoooi/k8s/awoooi-prod/04-configmap.yaml
-# 建議位置: OLLAMA_URL 行 (第 20 行) 之後
-#
-# --- 新增內容 ---
-  # 2026-04-25 ogt + Claude Sonnet 4.6: 188 CPU-only Ollama 備援 (plan_complete_v3 P0.5)
-  # ⚠️  188 推理延遲實測 ~111s (0.09 token/s, CPU-only Xeon 4214)，僅作 111 完全失效時的降級備援
-  # 模型已存在: qwen2.5:7b-instruct (4.5GB), 無需重拉
-  OLLAMA_FALLBACK_188: "http://192.168.0.188:11434"
-  OLLAMA_188_MODEL: "qwen2.5:7b-instruct"
-# --- 新增內容結束 ---
-#
-# 使用方式 (需用戶 review 後手動 apply):
-#   kubectl -n awoooi-prod apply -f k8s/awoooi-prod/04-configmap.yaml
-#   kubectl -n awoooi-prod rollout restart deployment/awoooi-api

k8s/awoooi-prod/13-cronjob-k3s-report.yaml

@@ -42,8 +42,11 @@ spec:
           restartPolicy: OnFailure
           containers:
             - name: k3s-report
-              image: 192.168.0.110:5000/awoooi-api:latest
-              imagePullPolicy: Always
+              # 2026-05-05 Codex: keep the API image placeholder so CD
+              # injects the same immutable tag used by API/worker. The old
+              # awoooi-api:latest repo returns 400 from Harbor after reboot.
+              image: 192.168.0.110:5000/library/api:IMAGE_TAG_PLACEHOLDER
+              imagePullPolicy: IfNotPresent
               command:
                 - python
                 - -m
@@ -63,5 +66,7 @@ spec:
                 limits:
                   cpu: "200m"
                   memory: "128Mi"
-          # 使用 API 的 ServiceAccount (需要 RBAC)
-          serviceAccountName: awoooi-api
+          # 2026-05-05 Codex: this report only calls Prometheus and Telegram.
+          # The old awoooi-api ServiceAccount does not exist, which prevented
+          # Job pods from being created after reboot.
+          serviceAccountName: default

k8s/awoooi-prod/14-cronjob-weekly-report.yaml

@@ -42,8 +42,11 @@ spec:
           restartPolicy: OnFailure
           containers:
             - name: weekly-report
-              image: 192.168.0.110:5000/awoooi-api:latest
-              imagePullPolicy: Always
+              # 2026-05-05 Codex: keep the API image placeholder so CD
+              # injects the same immutable tag used by API/worker. The old
+              # awoooi-api:latest repo returns 400 from Harbor after reboot.
+              image: 192.168.0.110:5000/library/api:IMAGE_TAG_PLACEHOLDER
+              imagePullPolicy: IfNotPresent
               command:
                 - python
                 - -m
@@ -63,5 +66,7 @@ spec:
                 limits:
                   cpu: "500m"
                   memory: "256Mi"
-          # 使用 API 的 ServiceAccount (需要 RBAC)
-          serviceAccountName: awoooi-api
+          # 2026-05-05 Codex: this report only calls app services, Prometheus,
+          # Git, and Telegram. The old awoooi-api ServiceAccount does not
+          # exist, which prevented Job pods from being created after reboot.
+          serviceAccountName: default

k8s/awoooi-prod/15-cronjob-km-vectorize.yaml

@@ -27,7 +27,10 @@ spec:
   jobTemplate:
     spec:
       backoffLimit: 2
-      activeDeadlineSeconds: 300
+      # 2026-05-05 Codex: allow post-reboot/post-migration catch-up batches.
+      # The script now fails if the API reports failed rows, so this longer
+      # deadline does not hide partial vectorization.
+      activeDeadlineSeconds: 1800
       template:
         metadata:
           labels:
@@ -37,8 +40,11 @@ spec:
           restartPolicy: OnFailure
           containers:
             - name: km-vectorize
-              image: 192.168.0.110:5000/awoooi-api:latest
-              imagePullPolicy: Always
+              # 2026-05-05 Codex: keep the API image placeholder so CD
+              # injects the same immutable tag used by API/worker. The old
+              # awoooi-api:latest repo returns 400 from Harbor after reboot.
+              image: 192.168.0.110:5000/library/api:IMAGE_TAG_PLACEHOLDER
+              imagePullPolicy: IfNotPresent
               command:
                 - python
                 - /app/scripts/cron_km_vectorize.py
@@ -46,7 +52,9 @@ spec:
                 - name: TZ
                   value: "Asia/Taipei"
                 - name: INTERNAL_API_URL
-                  value: "http://awoooi-api.awoooi-prod.svc.cluster.local:8000"
+                  # 2026-05-05 Codex: use the actual Service name; the old
+                  # awoooi-api DNS name does not exist in awoooi-prod.
+                  value: "http://awoooi-api-svc.awoooi-prod.svc.cluster.local:8000"
               resources:
                 requests:
                   cpu: "50m"
@@ -54,4 +62,7 @@ spec:
                 limits:
                   cpu: "200m"
                   memory: "128Mi"
-          serviceAccountName: awoooi-api
+          # 2026-05-05 Codex: this job only calls the internal API. The old
+          # awoooi-api ServiceAccount does not exist, which prevented Job pods
+          # from being created after reboot.
+          serviceAccountName: default

k8s/drift-cronjob.yaml

@@ -63,10 +63,11 @@ spec:
                           print(f"status={r.status_code} body={r.text[:200]}")
                   asyncio.run(run())
               env:
-                # 2026-04-09 Claude Sonnet 4.6: ClusterIP 和 DNS 在 Job Pod 均不可達
-                # 改用 NodePort 直連 K3s worker node（同 K8s_API_SERVER_URL 解法）
+                # 2026-05-05 Codex: call the in-cluster Service instead of a
+                # fixed worker NodePort. After reboot, 121 can be unavailable
+                # while the Service and VIP are already healthy.
                 - name: INTERNAL_API_URL
-                  value: "http://192.168.0.121:32334"
+                  value: "http://awoooi-api-svc.awoooi-prod.svc.cluster.local:8000"
                 - name: DRIFT_SCAN_NAMESPACES
                   value: "awoooi-prod"
               resources:

k8s/monitoring/alert-chain-monitor.yaml

@@ -88,7 +88,7 @@ spec:
         # -----------------------------------------------------------------
         - alert: NoAlertsReceived2Hours
           expr: |
-            time() - max by (source)(awoooi_alert_chain_last_success_timestamp) > 7200
+            time() - max by (source)(awoooi_alert_chain_last_success_timestamp{source="alertmanager"}) > 7200
           for: 5m
           labels:
             severity: warning

ops/monitoring/alerts-unified.yml

@@ -15,6 +15,39 @@
 
 groups:
 
+  # =========================================================================
+  # Full-stack recovery scorecard recording rules
+  # =========================================================================
+  - name: full_stack_recovery_scorecard_rules
+    interval: 60s
+    rules:
+      - record: awoooi_recovery_core_ready
+        expr: |
+          sum without(result) (
+            awoooi_cold_start_last_result{host="110",scope="110_120_121_188",result="green"} == bool 1
+          )
+          * on(host,scope) (
+            awoooi_cold_start_warn_gates{host="110",scope="110_120_121_188"} == bool 0
+          )
+          * on(host,scope) (
+            awoooi_cold_start_blocked_gates{host="110",scope="110_120_121_188"} == bool 0
+          )
+          * on(host,scope) (
+            (time() - awoooi_cold_start_last_green_timestamp{host="110",scope="110_120_121_188"}) < bool 3600
+          )
+
+      - record: awoooi_recovery_dr_offsite_ready
+        expr: |
+          max by(host) (
+            awoooi_backup_offsite_configured{host="110"} == bool 1
+          )
+          * on(host) max by(host) (
+            awoooi_backup_offsite_fresh{host="110"} == bool 1
+          )
+          * on(host) min by(host) (
+            awoooi_backup_credential_escrow_fresh{host="110"} == bool 1
+          )
+
   # =========================================================================
   # 主機層告警 (host_alerts)
   # =========================================================================
@@ -33,20 +66,22 @@ groups:
           description: "Node Exporter 無回應超過 1 分鐘"
 
       - alert: HostHighCpuLoad
-        expr: 100 - (avg by(host) (rate(node_cpu_seconds_total{mode="idle"}[5m])) * 100) > 80
-        for: 5m
+        # 2026-05-05 ogt + Codex: keep this as early warning only.
+        # Sustained overload/root-cause automation is handled by HostLoadAverageSustainedHigh.
+        expr: 100 - (avg by(host) (rate(node_cpu_seconds_total{mode="idle"}[5m])) * 100) > 90
+        for: 10m
         labels:
           severity: warning
           layer: systemd-188
           team: ops
-          auto_repair: "true"
+          auto_repair: "false"
           # MCP Phase 2a (ADR-071, 2026-04-11 Claude Sonnet 4.6): SSH MCP 路由標籤
           mcp_provider: "ssh_host"
           host_type: "bare_metal"
           alert_category: "host_resource"
         annotations:
           summary: "主機 {{ $labels.host }} CPU 高負載"
-          description: "CPU 使用率超過 80%"
+          description: "CPU 使用率超過 90% 持續 10 分鐘；若 load5/core 未超過 1.5，先視為容量觀察與診斷，不直接修復。"
 
       - alert: HostLoadAverageSustainedHigh
         # 2026-05-05 ogt + Codex: 110/188 長時間過載基線。
@@ -165,7 +200,7 @@ groups:
           description: "過去 24 小時有備份失敗"
 
       - alert: VeleroBackupNotRun
-        expr: time() - velero_backup_last_successful_timestamp > 86400
+        expr: max by(host, namespace) (awoooi_velero_latest_completed_backup_fresh{host="110",namespace="velero"}) == 0
         for: 10m
         labels:
           severity: critical
@@ -175,7 +210,7 @@ groups:
           auto_repair: "false"
         annotations:
           summary: "Velero 超過 24 小時未成功備份"
-          description: "最後一次成功備份超過 24 小時"
+          description: "backup health exporter 顯示 latest Completed Velero backup 超過 25 小時或不存在。"
 
       # Sprint C-2 Host rsync 備份告警 (2026-04-11 Claude Sonnet 4.6)
       # backup-from-110.sh 成功時寫入 /var/run/backup-110.last_success
@@ -505,7 +540,7 @@ groups:
           description: "Sentry 錯誤可能無法正確處理"
 
       - alert: NoAlertsReceived2Hours
-        expr: time() - max by (source)(awoooi_alert_chain_last_success_timestamp) > 7200
+        expr: time() - max by (source)(awoooi_alert_chain_last_success_timestamp{source="alertmanager"}) > 7200
         for: 5m
         labels:
           severity: warning
@@ -665,7 +700,7 @@ groups:
 
       - alert: DockerContainerMissingResourceLimit
         # 2026-05-05 ogt + Codex: catch Compose services that silently run with unlimited CPU/memory.
-        expr: (docker_container_cpu_limit_cores{container_name=~"momo-.*|litellm|gitea|sentry-self-hosted-(clickhouse|kafka|taskworker|process-spans).*"} == 0) or (docker_container_memory_limit_bytes{container_name=~"momo-.*|litellm|gitea|sentry-self-hosted-(clickhouse|kafka|taskworker|process-spans).*"} == 0)
+        expr: (docker_container_cpu_limit_cores{container_name=~"momo-.*|litellm|gitea|sentry-self-hosted-(clickhouse|kafka|taskworker|taskbroker|taskscheduler|redis|uptime-checker|process-spans).*"} == 0) or (docker_container_memory_limit_bytes{container_name=~"momo-.*|litellm|gitea|sentry-self-hosted-(clickhouse|kafka|taskworker|taskbroker|taskscheduler|redis|uptime-checker|process-spans).*"} == 0)
         for: 30m
         labels:
           severity: warning
@@ -1011,10 +1046,10 @@ groups:
   # 備份還原告警 (awoooi_backup_restore) — 從主機補回 2026-04-12
   # =========================================================================
   - name: awoooi_backup_restore
-    interval: 1h
+    interval: 1m
     rules:
       - alert: BackupRestoreTestFailed
-        expr: awoooi_backup_restore_test_success == 0
+        expr: max by(host, namespace, cronjob) (awoooi_velero_restore_test_failed_jobs{host="110",namespace="velero",cronjob="backup-restore-test"}) > 0
         for: 5m
         labels:
           severity: critical
@@ -1023,11 +1058,37 @@ groups:
           auto_repair: "false"
         annotations:
           summary: "備份還原 dry-run 測試失敗"
-          description: "Velero restore dry-run 失敗，備份可能無法還原。立即人工驗證備份狀態。"
-          runbook: "執行 velero backup describe awoooi-daily 及 velero restore create --from-backup awoooi-daily --dry-run"
+          description: "velero namespace 中保留了失敗的 backup-restore-test Job，備份可能無法還原。立即人工驗證備份狀態。"
+          runbook: "先找最新 Completed Velero backup，再執行 restore dry-run；禁止在 production namespace 做真還原"
+
+      - alert: BackupRestoreTestMissing
+        expr: absent(awoooi_velero_restore_test_cron_present{host="110",namespace="velero",cronjob="backup-restore-test"})
+        for: 30m
+        labels:
+          severity: warning
+          alert_category: infrastructure
+          notification_type: TYPE-3
+          auto_repair: "false"
+        annotations:
+          summary: "備份還原 dry-run 監控指標缺失"
+          description: "Prometheus 沒有收到 awoooi_velero_restore_test_cron_present；110 backup health exporter 或 120 kubectl 查詢可能失效。"
+          runbook: "檢查 110 backup_health.prom、SSH 110→120、以及 velero namespace 的 backup-restore-test CronJob"
+
+      - alert: BackupRestoreTestCronMissing
+        expr: max by(host, namespace, cronjob) (awoooi_velero_restore_test_cron_present{host="110",namespace="velero",cronjob="backup-restore-test"}) == 0
+        for: 15m
+        labels:
+          severity: critical
+          alert_category: infrastructure
+          notification_type: TYPE-3
+          auto_repair: "false"
+        annotations:
+          summary: "備份還原 dry-run CronJob 缺失"
+          description: "velero namespace 找不到 backup-restore-test CronJob；備份可還原性沒有定期驗證。"
+          runbook: "kubectl apply k8s/awoooi-prod/17-configmap-backup-restore-scripts.yaml 與 16-cronjob-backup-restore-test.yaml"
 
       - alert: BackupRestoreTestStale
-        expr: (time() - awoooi_backup_restore_test_timestamp_seconds) > 691200
+        expr: max by(host, namespace, cronjob) (awoooi_velero_restore_test_last_success_fresh{host="110",namespace="velero",cronjob="backup-restore-test"}) == 0
         for: 10m
         labels:
           severity: warning
@@ -1036,9 +1097,375 @@ groups:
           auto_repair: "false"
         annotations:
           summary: "備份還原測試超過 8 天未執行"
-          description: "上次備份測試距今 {{ $value | humanizeDuration }}，週排程 CronJob 可能失效。"
+          description: "backup-restore-test CronJob 沒有 8 天內成功紀錄；週排程 CronJob 可能失效。"
           runbook: "檢查 velero namespace 中的 backup-restore-test CronJob 狀態"
 
+  # =========================================================================
+  # Host / service / config backup health
+  # =========================================================================
+  - name: full_stack_backup_health_alerts
+    interval: 1m
+    rules:
+      - alert: BackupHealthMonitorMissing110
+        expr: absent(awoooi_backup_health_monitor_up{host="110"})
+        for: 20m
+        labels:
+          severity: warning
+          layer: host-backup
+          component: backup-health-monitor
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "110 備份健康指標缺失"
+          description: "110 沒有輸出 backup_health.prom，無法確認資料庫、設定檔與服務備份是否新鮮。"
+          runbook: "用 Ansible `110-devops.yml --tags textfile_exporters` 或手動部署 scripts/ops/backup-health-textfile-exporter.py"
+
+      - alert: BackupHealthMonitorMissing188
+        expr: absent(awoooi_backup_health_monitor_up{host="188"})
+        for: 20m
+        labels:
+          severity: warning
+          layer: host-backup
+          component: backup-health-monitor
+          host: "188"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "188 備份健康指標缺失"
+          description: "188 沒有輸出 backup_health.prom，無法確認 110 rsync 與 momo PostgreSQL 備份是否新鮮。"
+          runbook: "用 Ansible `188-ai-web.yml --tags textfile_exporters` 或手動部署 scripts/ops/backup-health-textfile-exporter.py"
+
+      - alert: BackupHealthMonitorStale
+        expr: time() - awoooi_backup_health_last_run_timestamp{host=~"110|188"} > 1800
+        for: 10m
+        labels:
+          severity: warning
+          layer: host-backup
+          component: backup-health-monitor
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "{{ $labels.host }} 備份健康 exporter 超過 30 分鐘未更新"
+          description: "backup health textfile exporter stale，備份狀態不可觀測。"
+          runbook: "SSH 主機檢查 cron、/tmp/awoooi-backup-health-textfile-exporter.cron.log 與 node-exporter textfile collector"
+
+      - alert: BackupExpectedJobMissing
+        expr: awoooi_backup_job_configured{host=~"110|188"} == 0
+        for: 15m
+        labels:
+          severity: critical
+          layer: host-backup
+          component: backup-cron
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-3
+          auto_repair: "false"
+        annotations:
+          summary: "{{ $labels.host }} 備份排程缺失：{{ $labels.exported_job }}"
+          description: "預期備份 cron/config 不存在；下一次重開機後資料可能沒有可用還原點。"
+          runbook: "依 docs/runbooks/FULL-STACK-COLD-START-SOP.md 的備份章節補回 cron，先 dry-run 再執行"
+
+      - alert: BackupScheduleDuplicateActiveEntries
+        expr: awoooi_backup_cron_active_duplicate_count{host="110"} > 0
+        for: 15m
+        labels:
+          severity: warning
+          layer: host-backup
+          component: backup-cron
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "110 備份 crontab 有重複 active entries"
+          description: "110 crontab 目前有 {{ $value }} 個 exact duplicate active entry；可能造成 offsite sync、verifier 或 status job 重複執行。"
+          runbook: "SSH 110 執行 `crontab -l | awk 'NF && $0 !~ /^#/ {count[$0]++} END {for (line in count) if (count[line] > 1) print count[line], line}'`，只移除重複 active entry，不要刪除未理解的備份排程。"
+
+      - alert: BackupScheduleSingletonMismatch
+        expr: awoooi_backup_cron_singular_entry_ok{host="110"} == 0
+        for: 15m
+        labels:
+          severity: warning
+          layer: host-backup
+          component: backup-cron
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "110 備份排程單一入口異常：{{ $labels.entry }}"
+          description: "{{ $labels.entry }} 應該剛好只有一個 active cron entry；目前 count={{ $value }}，可能造成排程缺失或重複執行。"
+          runbook: "用 Ansible `110-devops.yml --tags backup_jobs` 收斂排程，並用 `scripts/reboot-recovery/reboot-recovery-readiness-audit.sh --live --no-color` 驗證。"
+
+      - alert: BackupScriptMissing
+        expr: awoooi_backup_script_present{host=~"110|188"} == 0
+        for: 15m
+        labels:
+          severity: critical
+          layer: host-backup
+          component: backup-script
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-3
+          auto_repair: "false"
+        annotations:
+          summary: "{{ $labels.host }} 備份腳本缺失：{{ $labels.script }}"
+          description: "備份排程可能存在，但實際腳本不存在或路徑漂移。"
+          runbook: "從 repo 部署對應 scripts/backup 或 scripts/ops 腳本，確認權限 0755"
+
+      - alert: BackupJobStale
+        expr: awoooi_backup_job_fresh{host=~"110|188"} == 0
+        for: 15m
+        labels:
+          severity: critical
+          layer: host-backup
+          component: backup-freshness
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-3
+          auto_repair: "false"
+        annotations:
+          summary: "{{ $labels.host }} 備份過舊：{{ $labels.exported_job }}"
+          description: "{{ $labels.exported_job }} 最新成功證據超過 {{ $labels.max_age_hours }} 小時或不存在；來源 {{ $labels.source }}，目標 {{ $labels.target }}。"
+          runbook: "先檢查備份 log 與磁碟空間，再手動執行對應備份；禁止直接刪除舊備份或 production 資料"
+
+      - alert: BackupAggregateRunFailed
+        expr: awoooi_backup_last_run_failed_count{host="110",exported_job="backup_all"} > 0
+        for: 10m
+        labels:
+          severity: warning
+          layer: host-backup
+          component: backup-all
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "110 全服務備份最近一次有 {{ $value }} 個失敗項目"
+          description: "backup-all.sh 最近一次 aggregate run 仍有失敗；即使個別 DB 備份已手動補跑，也要重跑 aggregate backup 清除紅燈。"
+          runbook: "SSH 110 檢查 /backup/logs/cron.log 與 /backup/logs/backup.log，修正後執行 /backup/scripts/backup-all.sh"
+
+      - alert: BackupConfigCapturePartial
+        expr: awoooi_backup_config_capture_ok{host="110",critical="true"} == 0
+        for: 10m
+        labels:
+          severity: warning
+          layer: host-backup
+          component: backup-config-capture
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "110 設定檔備份缺少關鍵目標：{{ $labels.target }}"
+          description: "configs restic snapshot 雖可能存在，但最新設定檔備份未成功捕捉 {{ $labels.target }}；source={{ $labels.source }}。這會影響下一次冷啟動或災難還原的設定還原完整性。"
+          runbook: "先修復對應主機或 K8s API 可達性，再執行 /backup/scripts/backup-configs.sh，確認 awoooi_backup_config_capture_ok 回到 1，最後補跑 Google Drive/rclone offsite sync。"
+
+      - alert: BackupConfigCaptureStatusStale
+        expr: absent(awoooi_backup_config_capture_status_timestamp{host="110"}) or (time() - awoooi_backup_config_capture_status_timestamp{host="110"} > 172800)
+        for: 30m
+        labels:
+          severity: warning
+          layer: host-backup
+          component: backup-config-capture
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "110 設定檔備份覆蓋率狀態缺失或過舊"
+          description: "backup-configs.sh 沒有新鮮的 capture status；無法判斷 110/120/121/188/K8s 設定檔是否真的被最新 snapshot 捕捉。"
+          runbook: "部署新版 /backup/scripts/backup-configs.sh 與 /home/wooo/scripts/backup-health-textfile-exporter.py，執行 /backup/scripts/backup-configs.sh 後刷新 textfile exporter。"
+
+      - alert: BackupIntegrityCheckMissingOrFailed
+        expr: absent(awoooi_backup_integrity_fresh{host="110",scope="restic_check"}) or awoooi_backup_integrity_fresh{host="110",scope="restic_check"} == 0
+        for: 30m
+        labels:
+          severity: critical
+          layer: host-backup
+          component: backup-integrity
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-3
+          auto_repair: "false"
+        annotations:
+          summary: "110 備份倉庫完整性檢查缺失或失敗"
+          description: "每週 restic check 沒有成功證據，或有 repo 檢查失敗；目前不能假設備份可讀。"
+          runbook: "SSH 110 執行 `/backup/scripts/check-backup-integrity.sh --mode check`，先看 /backup/logs/backup-integrity.log；禁止刪 repo 或 prune 直到確認原因"
+
+      - alert: BackupRestoreDrillMissingOrFailed
+        expr: absent(awoooi_backup_integrity_fresh{host="110",scope="restore_drill"}) or awoooi_backup_integrity_fresh{host="110",scope="restore_drill"} == 0
+        for: 30m
+        labels:
+          severity: warning
+          layer: host-backup
+          component: backup-restore-drill
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "110 備份抽樣還原演練缺失或失敗"
+          description: "每月 restore drill 沒有成功證據，備份雖可能新鮮，但尚未驗證可讀取還原。"
+          runbook: "SSH 110 執行 `/backup/scripts/check-backup-integrity.sh --mode restore-drill`；只允許還原到隔離暫存目錄，不得覆蓋 production"
+
+      - alert: BackupOffsiteCopyNotConfigured
+        expr: sum by(host) (awoooi_backup_offsite_configured{host="110"}) == 0
+        for: 1m
+        labels:
+          severity: warning
+          layer: host-backup
+          component: backup-offsite
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "110 尚未配置離機備份 provider"
+          description: "backup health exporter 未偵測到 Google Drive/rclone 或其他 offsite provider 配置；本地 restic 全綠仍不等於異地可恢復。"
+          runbook: "在 110 以 `/backup/scripts/configure-offsite-rclone.sh --interactive` 建立 Google Drive remote，產生 `/backup/offsite/*last_success` 證據；不得把 provider token 寫入 repo、Telegram 或 Prometheus label。"
+
+      - alert: BackupOffsiteCopyStale
+        expr: |
+          (
+            (sum by(host) (awoooi_backup_offsite_configured{host="110"}) > 0)
+            and
+            (sum by(host) (awoooi_backup_offsite_fresh{host="110"}) == 0)
+          )
+          and
+          (
+            (sum by(host) (awoooi_backup_offsite_full_sync_enabled{host="110"}) == 0)
+            or
+            ((time() - max by(host) (awoooi_backup_offsite_full_sync_enabled_timestamp{host="110"})) > 30 * 3600)
+          )
+        for: 2h
+        labels:
+          severity: warning
+          layer: host-backup
+          component: backup-offsite
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "110 離機備份超過 48 小時未成功"
+          description: "已偵測到 offsite provider 配置，但沒有新鮮成功標記；本地備份可能無法抵抗整台 110 遺失。"
+          runbook: "SSH 110 檢查 Google Drive/rclone 同步 log 與 `/backup/offsite/*last_success`；full sync 需在 enable marker 與低負載門檻成立後由 `/backup/scripts/sync-offsite-backups.sh --mode sync` 鏡像本地 latest-only repo。"
+
+      - alert: BackupRetentionPolicyNotLatestOnly
+        expr: |
+          absent(awoooi_backup_retention_latest_only{host="110"})
+          or
+          awoooi_backup_retention_latest_only{host="110"} != 1
+          or
+          absent(awoooi_backup_retention_offsite_delete_old_enabled{host="110",provider="rclone"})
+          or
+          awoooi_backup_retention_offsite_delete_old_enabled{host="110",provider="rclone"} != 1
+        for: 15m
+        labels:
+          severity: warning
+          layer: host-backup
+          component: backup-retention
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "110 備份保留策略不是 latest-only"
+          description: "operator 要求所有備份只保留最新一份；本地 restic 必須 keep-last=1，Google Drive/rclone 必須在成功 mirror 後刪除遠端舊檔。"
+          runbook: "檢查 `/backup/scripts/common.sh` 的 BACKUP_RETENTION_MODE=latest、KEEP_LAST=1 與 OFFSITE_SYNC_DELETE_OLD=1，刷新 backup-health textfile；必要時在備份成功後執行 `/backup/scripts/enforce-latest-only-retention.sh`。"
+
+      - alert: BackupSnapshotRetentionExceeded
+        expr: awoooi_backup_job_snapshot_count{host="110",type="restic"} > 1
+        for: 30m
+        labels:
+          severity: warning
+          layer: host-backup
+          component: backup-retention
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "110 備份 repo {{ $labels.exported_job }} 保留超過 1 份 snapshot"
+          description: "{{ $labels.exported_job }} 目前有 {{ $value }} 份 restic snapshot；latest-only 策略要求每個 repo 全域只保留最新 1 份。"
+          runbook: "SSH 110 執行 `/backup/scripts/enforce-latest-only-retention.sh`；若仍未收斂，確認 `common.sh` 使用 `restic forget --group-by \"\" --keep-last 1 --prune`，避免 restic 依 path/tag 分組保留多份。"
+
+      - alert: BackupOffsiteFullVerifyFailed
+        expr: |
+          awoooi_backup_offsite_fresh{host="110",provider="rclone"} == 1
+          unless on(host, provider)
+          (awoooi_backup_offsite_remote_verify_ok{host="110",provider="rclone"} == 1)
+        for: 30m
+        labels:
+          severity: warning
+          layer: host-backup
+          component: backup-offsite
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "110 Google Drive full sync 完成但遠端驗證未通過"
+          description: "full offsite marker 已 fresh，但 verify-offsite-full-sync.sh 沒有證明 13 個 Google Drive repo 都可列出且符合 latest-only。"
+          runbook: "SSH 110 執行 `/backup/scripts/verify-offsite-full-sync.sh --write-textfile --no-color`，檢查 `/backup/logs/offsite-full-sync-verify.log` 與 `/home/wooo/node_exporter_textfiles/offsite_full_sync_verify.prom`。"
+
+      - alert: BackupOffsiteRemoteSnapshotRetentionExceeded
+        expr: |
+          (awoooi_backup_offsite_remote_snapshot_count{host="110",provider="rclone"} > 1)
+          and on(host, provider)
+          (awoooi_backup_offsite_fresh{host="110",provider="rclone"} == 1)
+        for: 30m
+        labels:
+          severity: warning
+          layer: host-backup
+          component: backup-retention
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "Google Drive repo {{ $labels.repo }} 保留超過 1 份 snapshot"
+          description: "{{ $labels.repo }} 在 Google Drive/rclone 遠端目前有 {{ $value }} 份 snapshot；latest-only 策略要求遠端也只保留最新一份。"
+          runbook: "確認 110 `/backup/scripts/sync-offsite-backups.sh --mode sync` 使用 `rclone sync`、`OFFSITE_SYNC_DELETE_OLD=1`、`RCLONE_DRIVE_USE_TRASH=false`，再於低峰重新執行 full sync 與 verifier。"
+
+      - alert: BackupCredentialEscrowEvidenceMissing
+        expr: awoooi_backup_credential_escrow_fresh{host="110"} == 0
+        for: 1m
+        labels:
+          severity: warning
+          layer: host-backup
+          component: credential-escrow
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "備份憑證金庫證據缺失或過期：{{ $labels.item }}"
+          description: "{{ $labels.item }} 沒有 31 天內人工驗證證據；重建時可能找不到 restic/offsite/break-glass/DNS/OAuth 復原材料。"
+          runbook: "在密碼管理器或離線加密金庫完成雙人覆核後，只建立不含 secret 的 `/backup/escrow-evidence/{{ $labels.item }}.last_verified` 時間戳證據。"
+
   # =========================================================================
   # 基礎設施詳細告警 (awoooi_infrastructure_detailed) — 從主機補回 2026-04-12
   # =========================================================================
@@ -1321,3 +1748,284 @@ groups:
           summary: "Prometheus ({{ $labels.instance }}) 停擺"
           description: "Prometheus 自己停擺 → 所有其他告警失效"
           runbook: "SSH 110 docker compose -f /home/wooo/monitoring/docker-compose.yml up -d prometheus"
+
+  # =========================================================================
+  # Full-stack cold-start recovery gate
+  # =========================================================================
+  - name: cold_start_recovery_alerts
+    rules:
+      - alert: PrometheusRuleDriftGuardFailed
+        expr: |
+          absent(awoooi_prometheus_rule_drift_guard_last_run_timestamp{host="110"})
+          or
+          (time() - max by(host) (awoooi_prometheus_rule_drift_guard_last_run_timestamp{host="110"}) > 900)
+          or
+          (awoooi_prometheus_rule_drift_guard_missing_required_count{host="110"} > 0)
+          or
+          (awoooi_prometheus_rule_drift_guard_current_matches_canonical{host="110"} == 0)
+        for: 10m
+        labels:
+          severity: critical
+          layer: systemd-110
+          component: prometheus-rule-drift-guard
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-3
+          auto_repair: "false"
+        annotations:
+          summary: "Prometheus 規則漂移防護失效"
+          description: "110 Prometheus rule drift guard 沒有新鮮成功指標、required rules 缺失，或 active alerts.yml 不等於 canonical rules。"
+          runbook: "執行 `bash scripts/ops/deploy-alerts.sh` 重新部署 canonical rules 與 drift guard，等待 1-2 個 Prometheus evaluation cycle 後重跑 readiness audit。"
+
+      - alert: PrometheusRuleDriftAutoRepaired
+        expr: awoooi_prometheus_rule_drift_guard_repaired{host="110"} > 0
+        for: 1m
+        labels:
+          severity: warning
+          layer: systemd-110
+          component: prometheus-rule-drift-guard
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "Prometheus 規則漂移已被自動修復"
+          description: "110 drift guard 最近一次偵測到 active Prometheus rules 漂移，已回復 canonical rules 並 reload Prometheus。"
+          runbook: "檢查 `/home/wooo/logs/prometheus-rule-drift-guard.log` 與 `/home/wooo/monitoring/alerts.yml.guard.bak.*`，找出誰覆寫了 active rules。"
+
+      - alert: ColdStartMonitorMissing
+        expr: absent(awoooi_cold_start_monitor_up{host="110",scope="110_120_121_188"})
+        for: 15m
+        labels:
+          severity: warning
+          layer: systemd-110
+          component: cold-start-monitor
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "Cold-start monitor textfile metric missing"
+          description: "110 沒有輸出 awoooi_cold_start_monitor_up；重開機恢復 gate 目前不可觀測。"
+          runbook: "執行 scripts/reboot-recovery/install-cold-start-monitor-110.sh，確認 /home/wooo/node_exporter_textfiles/cold_start_recovery.prom"
+
+      - alert: ColdStartMonitorStale
+        expr: time() - awoooi_cold_start_last_run_timestamp{host="110",scope="110_120_121_188"} > 900
+        for: 10m
+        labels:
+          severity: warning
+          layer: systemd-110
+          component: cold-start-monitor
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "Cold-start monitor stale"
+          description: "cold-start monitor 超過 15 分鐘沒有更新，距離上次執行 {{ $value | humanizeDuration }}。"
+          runbook: "SSH 110 檢查 crontab、/tmp/awoooi-cold-start-monitor.cron.log、cold-start-last.log"
+
+      - alert: ColdStartRecoveryBlocked
+        expr: awoooi_cold_start_blocked_gates{host="110",scope="110_120_121_188"} > 0
+        for: 5m
+        labels:
+          severity: critical
+          layer: full-stack
+          component: cold-start-gate
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-3
+          auto_repair: "false"
+        annotations:
+          summary: "Full-stack cold-start recovery BLOCKED"
+          description: "cold-start gate 有 {{ $value }} 個 BLOCKED gate。AI 修復需保持 observe-only，先處理第一個 blocked gate。"
+          runbook: "查看 /home/wooo/reboot-recovery/cold-start-last.log；依 docs/runbooks/FULL-STACK-COLD-START-SOP.md 的 P0→P2 順序修復"
+
+      - alert: K3sNodeFilesystemErrorGateBlocked
+        expr: awoooi_cold_start_blocker_reason{host="110",scope="110_120_121_188",reason="k3s_node_filesystem_error",target="120"} > 0
+        for: 5m
+        labels:
+          severity: critical
+          layer: k3s
+          component: node-filesystem
+          host: "120"
+          target_host: "120"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-3
+          auto_repair: "false"
+        annotations:
+          summary: "120 K3s 節點 filesystem error 阻擋重開機放行"
+          description: "cold-start log 偵測到 120 Node event 仍有 EXT4/I/O/deleted inode 類錯誤；即使 Pod Running、網站 200，也不可宣告下一次重開機安全。"
+          runbook: "查看 110 `/home/wooo/reboot-recovery/cold-start-last.log`，執行 `scripts/reboot-recovery/120-fsck-maintenance-checklist.sh` 做維護前只讀檢查；維護窗內用 console/rescue 對 120 root LV 執行 fsck，禁止 online fsck。"
+
+      - alert: ColdStartHost120Unreachable
+        expr: awoooi_cold_start_blocker_reason{host="110",scope="110_120_121_188",reason="host_unreachable",target="120"} > 0
+        for: 3m
+        labels:
+          severity: critical
+          layer: host
+          component: host-reachability
+          host: "120"
+          target_host: "120"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-3
+          auto_repair: "false"
+        annotations:
+          summary: "120 主機不可達，Full-stack cold-start 已阻擋"
+          description: "110 cold-start monitor 無法 ping/SSH 192.168.0.120；目前只能由 121/VIP 撐住 K3s，不能宣告所有主機重開機恢復完成。"
+          runbook: "查看 120 console。若停在 initramfs/manual fsck，先對 root LV 做離線 fsck；若主機關機或網卡異常，先恢復電源/網路，再重跑 full-stack cold-start gate。禁止從自動修復直接重啟其他服務掩蓋主機離線。"
+
+      - alert: ColdStartRecoveryDegraded
+        expr: awoooi_cold_start_warn_gates{host="110",scope="110_120_121_188"} > 0
+        for: 15m
+        labels:
+          severity: warning
+          layer: full-stack
+          component: cold-start-gate
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "Full-stack cold-start recovery DEGRADED"
+          description: "cold-start gate 有 {{ $value }} 個 WARN gate；核心可用但不應放行 runner/CD/AI auto-repair full execution。"
+          runbook: "查看 /home/wooo/reboot-recovery/cold-start-last.log，修到 PASS/WARN/BLOCKED = green"
+
+      - alert: ColdStartLastGreenTooOld
+        expr: time() - awoooi_cold_start_last_green_timestamp{host="110",scope="110_120_121_188"} > 3600
+        for: 15m
+        labels:
+          severity: warning
+          layer: full-stack
+          component: cold-start-gate
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "Full-stack cold-start gate has not been GREEN recently"
+          description: "距離上次 GREEN 已超過 {{ $value | humanizeDuration }}；需要確認 110/120/121/188 與排程/網站 gate。"
+          runbook: "執行 SSH_BATCH_MODE=yes bash scripts/reboot-recovery/full-stack-cold-start-check.sh --send-alert-test"
+
+  # =========================================================================
+  # Host storage health / dirty reboot evidence
+  # =========================================================================
+  - name: host_storage_health_alerts
+    rules:
+      - alert: Host110StorageHealthMonitorMissing
+        expr: absent(awoooi_host_storage_monitor_up{host="110"})
+        for: 15m
+        labels:
+          severity: warning
+          layer: systemd-110
+          component: storage-health-monitor
+          host: "110"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "110 storage health textfile metric missing"
+          description: "110 沒有輸出 storage_health.prom；dirty reboot、root read-only 與 fsck 證據目前不可觀測。"
+          runbook: "用 Ansible `110-devops.yml --tags textfile_exporters` 或手動部署 scripts/ops/storage-health-textfile-exporter.py，確認 /home/wooo/node_exporter_textfiles/storage_health.prom"
+
+      - alert: Host188StorageHealthMonitorMissing
+        expr: absent(awoooi_host_storage_monitor_up{host="188"})
+        for: 15m
+        labels:
+          severity: warning
+          layer: systemd-188
+          component: storage-health-monitor
+          host: "188"
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "188 storage health textfile metric missing"
+          description: "188 沒有輸出 storage_health.prom；dirty reboot、root read-only 與 fsck 證據目前不可觀測。"
+          runbook: "用 Ansible `188-ai-web.yml --tags textfile_exporters` 或手動部署 scripts/ops/storage-health-textfile-exporter.py，確認 /home/ollama/node_exporter_textfiles/storage_health.prom"
+
+      - alert: HostStorageHealthMonitorStale
+        expr: time() - awoooi_host_storage_last_run_timestamp{host=~"110|188"} > 900
+        for: 10m
+        labels:
+          severity: warning
+          layer: host-storage
+          component: storage-health-monitor
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "主機 {{ $labels.host }} storage health textfile stale"
+          description: "storage health exporter 超過 15 分鐘沒有更新；重開機後檔案系統風險不可觀測。"
+          runbook: "SSH 主機檢查 cron、/tmp/awoooi-storage-health-textfile-exporter.cron.log 與 node-exporter textfile collector"
+
+      - alert: HostRootFilesystemReadOnly
+        expr: awoooi_host_root_filesystem_readonly{host=~"110|188",mountpoint="/"} > 0
+        for: 1m
+        labels:
+          severity: critical
+          layer: host-storage
+          component: root-filesystem
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-3
+          auto_repair: "false"
+        annotations:
+          summary: "主機 {{ $labels.host }} root filesystem 已變成 read-only"
+          description: "root filesystem 被掛載為唯讀，服務可能仍暫時存活但寫入會失敗；禁止自動修復，先保全證據並規劃維護窗。"
+          runbook: "依 docs/runbooks/FULL-STACK-COLD-START-SOP.md §16：保全 journal/df/mount 證據，確認備份，再安排 console/offline fsck"
+
+      - alert: HostCurrentBootStorageErrorsDetected
+        expr: awoooi_host_storage_error_count{host=~"110|188",source="journalctl-kernel",boot="current"} > 0
+        for: 5m
+        labels:
+          severity: critical
+          layer: host-storage
+          component: kernel-storage
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-3
+          auto_repair: "false"
+        annotations:
+          summary: "主機 {{ $labels.host }} current boot 有 storage/kernel 錯誤"
+          description: "目前開機週期已出現 filesystem、I/O 或 fsck 類錯誤；不可只重啟容器掩蓋問題。"
+          runbook: "先執行 read-only 診斷：journalctl -k -p warning..alert、mount、df、smartctl/raid 狀態；必要時進入維護窗處理"
+
+      - alert: HostPreviousBootStorageErrorsDetected
+        expr: awoooi_host_storage_error_count{host=~"110|188",source="journalctl-kernel",boot="previous"} > 0
+        for: 30m
+        labels:
+          severity: warning
+          layer: host-storage
+          component: dirty-reboot-evidence
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "主機 {{ $labels.host }} previous boot 保留 storage/fsck 錯誤證據"
+          description: "上一個開機週期留有 storage/fsck 錯誤，代表這次重開機事故需要完成 fsck、備份與容量後續檢查。"
+          runbook: "把證據寫入 docs/LOGBOOK.md，確認 full-stack cold-start gate 與 P3 gate；下一次維護窗補 offline fsck/SMART/RAID 檢查"
+
+      - alert: HostFsckLogErrorsDetected
+        expr: sum by(host) (awoooi_host_storage_error_count{host=~"110|188",boot="last-fsck-log"}) > 0
+        for: 30m
+        labels:
+          severity: warning
+          layer: host-storage
+          component: fsck-log
+          team: ops
+          alert_category: infrastructure
+          notification_type: TYPE-1
+          auto_repair: "false"
+        annotations:
+          summary: "主機 {{ $labels.host }} fsck log 保留錯誤證據"
+          description: "主機 fsck log 內仍有 inconsistency 或 I/O 類錯誤文字；這是事故後追蹤項，不應交給自動修復直接處理。"
+          runbook: "確認 /run/initramfs/fsck.log 與 /var/log/fsck/*，將結果納入重開機事故報告與下次維護窗檢查項"

ops/monitoring/alerts.yml

@@ -33,8 +33,10 @@ groups:
           description: "Node Exporter 無回應超過 1 分鐘"
 
       - alert: HostHighCpuLoad
-        expr: 100 - (avg by(host) (rate(node_cpu_seconds_total{mode="idle"}[5m])) * 100) > 80
-        for: 5m
+        # 2026-05-05 ogt + Codex: keep this as early warning only.
+        # Sustained overload/root-cause automation is handled by HostLoadAverageSustainedHigh.
+        expr: 100 - (avg by(host) (rate(node_cpu_seconds_total{mode="idle"}[5m])) * 100) > 90
+        for: 10m
         labels:
           severity: warning
           layer: systemd-188
@@ -46,7 +48,7 @@ groups:
           alert_category: "host_resource"
         annotations:
           summary: "主機 {{ $labels.host }} CPU 高負載"
-          description: "CPU 使用率超過 80%"
+          description: "CPU 使用率超過 90% 持續 10 分鐘；若 load5/core 未超過 1.5，先視為容量觀察與診斷，不直接修復。"
           # 2026-05-02 ogt + Claude Sonnet 4.6: 引導 LLM 走 SSH 診斷而非 kubectl
           auto_repair_action: "ssh {{ $labels.instance }} 'ps aux --sort=-%cpu | head -20' (host CPU 診斷；禁 kubectl restart awoooi-* — 主因常為第三方服務 Sentry/ClickHouse/Snuba)"
           runbook: "host CPU 高負載排查：先 SSH ps aux 看 top 進程；若為第三方服務（Sentry/ClickHouse 等）寫 ADR 升級資源或調 limit，禁止 kubectl restart 跨 domain"
@@ -671,7 +673,7 @@ groups:
 
       - alert: DockerContainerMissingResourceLimit
         # 2026-05-05 ogt + Codex: catch Compose services that silently run with unlimited CPU/memory.
-        expr: (docker_container_cpu_limit_cores{container_name=~"momo-.*|litellm|gitea|sentry-self-hosted-(clickhouse|kafka|taskworker|process-spans).*"} == 0) or (docker_container_memory_limit_bytes{container_name=~"momo-.*|litellm|gitea|sentry-self-hosted-(clickhouse|kafka|taskworker|process-spans).*"} == 0)
+        expr: (docker_container_cpu_limit_cores{container_name=~"momo-.*|litellm|gitea|sentry-self-hosted-(clickhouse|kafka|taskworker|taskbroker|taskscheduler|redis|uptime-checker|process-spans).*"} == 0) or (docker_container_memory_limit_bytes{container_name=~"momo-.*|litellm|gitea|sentry-self-hosted-(clickhouse|kafka|taskworker|taskbroker|taskscheduler|redis|uptime-checker|process-spans).*"} == 0)
         for: 30m
         labels:
           severity: warning

ops/monitoring/generated/blackbox-targets-generated.yaml

@@ -26,8 +26,18 @@
 - labels:
     criticality: P0
     owner: ai-team
-    service: ollama
-  url: http://192.168.0.188:11434/api/tags
+    service: ollama-gcp-a
+  url: http://192.168.0.110:11435/api/tags
+- labels:
+    criticality: P0
+    owner: ai-team
+    service: ollama-gcp-b
+  url: http://192.168.0.110:11436/api/tags
+- labels:
+    criticality: P0
+    owner: ai-team
+    service: ollama-local
+  url: http://192.168.0.110:11437/api/tags
 - labels:
     criticality: P0
     owner: ai-team

ops/monitoring/generated/prometheus-scrape-generated.yaml

@@ -92,7 +92,9 @@ scrape_configs:
       service: ollama
       type: docker
     targets:
-    - 192.168.0.188:11434
+    - 192.168.0.110:11435
+    - 192.168.0.110:11436
+    - 192.168.0.110:11437
 - job_name: openclaw
   static_configs:
   - labels:

ops/monitoring/grafana/dashboards/ollama_failover.json

@@ -82,11 +82,11 @@
         "textMode": "auto"
       },
       "title": "Ollama 可用性",
-      "description": "up{job=~\"ollama_111|ollama_188\"} × 100\n- 綠色 100% = 主機在線\n- 紅色 0% = 主機離線（容災應已觸發）\n\n資料來源: Prometheus scrape job ollama_111 / ollama_188",
+      "description": "up{job=~\"ollama_gcp_a|ollama_gcp_b|ollama_local|ollama_111\"} × 100\n- 綠色 100% = 主機在線\n- 紅色 0% = 主機離線（容災應已觸發）\n\n資料來源: Prometheus scrape job ollama_gcp_a / ollama_gcp_b / ollama_local",
       "targets": [
         {
           "datasource": { "type": "prometheus", "uid": "${datasource}" },
-          "expr": "up{job=~\"ollama_111|ollama_188\"} * 100",
+          "expr": "up{job=~\"ollama_gcp_a|ollama_gcp_b|ollama_local|ollama_111\"} * 100",
           "legendFormat": "{{ job }}",
           "refId": "A"
         }
@@ -188,7 +188,7 @@
         "tooltip": { "mode": "single", "sort": "none" }
       },
       "title": "AI Provider 路由分布",
-      "description": "sum by (provider) (rate(ai_router_selected_provider_total[5m]))\n- 正常狀態: ollama 佔大多數\n- failover 中: gemini / ollama_188 比例上升\n- 全走 gemini = 111 完全 offline\n\n資料來源: OLLAMA_FAILOVER_TRIGGERED_TOTAL + AI_ROUTER_PROVIDER_TOTAL (src/core/metrics.py)",
+      "description": "sum by (provider) (rate(ai_router_selected_provider_total[5m]))\n- 正常狀態: ollama / ollama_gcp_a 佔大多數\n- failover 中: ollama_gcp_b / ollama_local / gemini 比例上升\n- 全走 gemini = Ollama provider pool 完全 offline\n\n資料來源: OLLAMA_FAILOVER_TRIGGERED_TOTAL + AI_ROUTER_PROVIDER_TOTAL (src/core/metrics.py)",
       "targets": [
         {
           "datasource": { "type": "prometheus", "uid": "${datasource}" },

ops/monitoring/ollama_health_rules.yaml

@@ -6,7 +6,7 @@
 # 部署方式: 手動合併至 alerts-unified.yml，或 scripts/ops/deploy-alerts.sh 支援多檔時直接引用
 #
 # 標籤規範 (對齊 alerts-unified.yml):
-#   layer: systemd-188 | docker-188 (Ollama 跑在 188 主機)
+#   layer: ai-provider
 #   team: ai
 #   auto_repair: "true" | "false"
 #
@@ -28,16 +28,16 @@ groups:
 
       # -----------------------------------------------------------------------
       # 🔴 [ACTIVE] Ollama 主機離線
-      # metric: up{job=~"ollama_111|ollama_188"}
-      # 前置條件: Prometheus scrape job 命名為 ollama_111 / ollama_188
+      # metric: up{job=~"ollama_gcp_a|ollama_gcp_b|ollama_local|ollama_111"}
+      # 前置條件: Prometheus scrape job 命名對齊 ADR-110 provider pool
       #   (設定位於 ops/monitoring/generated/prometheus-scrape-generated.yaml)
       # -----------------------------------------------------------------------
       - alert: OllamaInstanceDown
-        expr: up{job=~"ollama_111|ollama_188"} == 0
+        expr: up{job=~"ollama_gcp_a|ollama_gcp_b|ollama_local|ollama_111"} == 0
         for: 2m
         labels:
           severity: critical
-          layer: systemd-188
+          layer: ai-provider
           team: ai
           auto_repair: "false"
           alert_category: "ollama_failover"
@@ -57,7 +57,7 @@ groups:
         for: 10m
         labels:
           severity: warning
-          layer: systemd-188
+          layer: ai-provider
           team: ai
           auto_repair: "false"
           alert_category: "ollama_failover"

scripts/ai_code_reviewer.py

@@ -19,6 +19,7 @@ Exit Codes:
 """
 
 import json
+import os
 import subprocess
 import sys
 from pathlib import Path
@@ -29,7 +30,7 @@ import httpx
 # Configuration
 # =============================================================================
 
-OLLAMA_URL = "http://192.168.0.188:11434/api/generate"
+OLLAMA_URL = os.getenv("OLLAMA_GENERATE_URL", "http://192.168.0.111:11434/api/generate")
 MODEL = "llama3.2:8b"
 PROJECT_ROOT = Path(__file__).parent.parent
 RULES_FILE = PROJECT_ROOT / ".awoooi-agent-rules.md"

scripts/cron_km_vectorize.py

@@ -18,17 +18,21 @@ import httpx
 async def main() -> int:
     api_base = os.environ.get(
         "INTERNAL_API_URL",
-        "http://awoooi-api.awoooi-prod.svc.cluster.local:8000",
+        "http://awoooi-api-svc.awoooi-prod.svc.cluster.local:8000",
     )
     url = f"{api_base}/api/v1/knowledge/embed-all"
 
-    async with httpx.AsyncClient(timeout=120) as client:
+    async with httpx.AsyncClient(timeout=1800) as client:
         try:
             resp = await client.post(url)
             print(f"embed-all: {resp.status_code} {resp.text[:200]}")
             if resp.status_code >= 400:
                 print(f"ERROR: embed-all returned {resp.status_code}", file=sys.stderr)
                 return 1
+            result = resp.json()
+            if int(result.get("failed", 0)) > 0:
+                print(f"ERROR: embed-all failed rows: {result}", file=sys.stderr)
+                return 1
             return 0
         except httpx.RequestError as exc:
             print(f"ERROR: request failed — {exc}", file=sys.stderr)

scripts/health_check_session.sh

@@ -62,7 +62,6 @@ 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 ---"

scripts/reboot-recovery/full-stack-cold-start-check.sh

@@ -0,0 +1,398 @@
+#!/usr/bin/env bash
+# AWOOOI full-stack cold-start readiness check.
+# Read-only by design. It never restarts, deletes, repairs, or writes remote state.
+
+set -uo pipefail
+
+SSH_OPTS=(-o BatchMode=yes -o ConnectTimeout=6)
+SEND_ALERT_TEST=0
+
+for arg in "$@"; do
+  case "$arg" in
+    --send-alert-test)
+      SEND_ALERT_TEST=1
+      ;;
+    -h|--help)
+      cat <<'USAGE'
+Usage: bash scripts/reboot-recovery/full-stack-cold-start-check.sh [--send-alert-test]
+
+Default mode is read-only and does not POST an Alertmanager test event.
+Use --send-alert-test only after AWOOOI API is expected to be ready.
+USAGE
+      exit 0
+      ;;
+    *)
+      echo "Unknown argument: $arg" >&2
+      exit 64
+      ;;
+  esac
+done
+
+RED=$'\033[0;31m'
+GREEN=$'\033[0;32m'
+YELLOW=$'\033[1;33m'
+BLUE=$'\033[0;34m'
+NC=$'\033[0m'
+
+PASS=0
+WARN=0
+FAIL=0
+
+log_section() {
+  printf "\n%s=== %s ===%s\n" "$BLUE" "$1" "$NC"
+}
+
+ok() {
+  printf "%sOK%s %s\n" "$GREEN" "$NC" "$1"
+  PASS=$((PASS + 1))
+}
+
+warn() {
+  printf "%sWARN%s %s\n" "$YELLOW" "$NC" "$1"
+  WARN=$((WARN + 1))
+}
+
+fail() {
+  printf "%sBLOCKED%s %s\n" "$RED" "$NC" "$1"
+  FAIL=$((FAIL + 1))
+}
+
+run_local() {
+  local label="$1"
+  shift
+  if "$@" >/tmp/awoooi-cold-start-check.out 2>&1; then
+    ok "$label"
+    cat /tmp/awoooi-cold-start-check.out
+    return 0
+  fi
+  fail "$label"
+  cat /tmp/awoooi-cold-start-check.out
+  return 1
+}
+
+ssh_cmd() {
+  local user_host="$1"
+  local cmd="$2"
+  local prefix=""
+  if [ -n "${REMOTE_SUDO_PASSWORD:-}" ]; then
+    printf -v prefix 'REMOTE_SUDO_PASSWORD=%q ' "$REMOTE_SUDO_PASSWORD"
+  fi
+  ssh "${SSH_OPTS[@]}" "$user_host" "${prefix}${cmd}"
+}
+
+probe_http_code() {
+  local url="$1"
+  local code
+  code=$(curl -s -o /dev/null -w "%{http_code}" --max-time 5 "$url" 2>/dev/null || true)
+  echo "${code:-000}"
+}
+
+probe_tcp() {
+  local host="$1"
+  local port="$2"
+  nc -G 3 -z "$host" "$port" >/dev/null 2>&1 || nc -w 3 -z "$host" "$port" >/dev/null 2>&1
+}
+
+print_header() {
+  echo "AWOOOI full-stack cold-start check"
+  date '+%Y-%m-%d %H:%M:%S %Z'
+  echo "Scope: 110 / 120 / 121 / 188. 112 Kali is intentionally skipped."
+}
+
+check_network() {
+  log_section "P0-NETWORK"
+  local host
+  for host in 110 120 121 188; do
+    if ping -c 1 -W 2 "192.168.0.$host" >/dev/null 2>&1; then
+      ok "ping 192.168.0.$host"
+    else
+      fail "ping 192.168.0.$host"
+    fi
+
+    if probe_tcp "192.168.0.$host" 22; then
+      ok "ssh port 192.168.0.$host:22"
+    else
+      fail "ssh port 192.168.0.$host:22"
+    fi
+  done
+
+  arp -an | grep -E '192\.168\.0\.(110|120|121|188)' || warn "no ARP rows printed for one or more hosts"
+}
+
+check_188() {
+  log_section "P0-188-DATA"
+  local out
+  if ! out=$(ssh_cmd "ollama@192.168.0.188" '
+echo "HOST $(hostname) $(uptime)"
+echo "MEM $(free -h | awk "/Mem:/ {print \$2,\$3,\$7}")"
+echo "SYSTEMD $(systemctl is-active containerd docker postgresql@14-main redis-server ollama nginx 2>/dev/null | tr "\n" " ")"
+echo "PG $(pg_isready -h localhost -p 5432 2>&1)"
+echo "REDIS $(redis-cli -p 6380 ping 2>/dev/null || redis-cli ping 2>/dev/null || true)"
+echo "PORT5432 $(nc -z -w 2 127.0.0.1 5432 >/dev/null 2>&1 && echo OPEN || echo CLOSED)"
+echo "SIGNOZ_CODE $(curl -s -o /dev/null -w "%{http_code}" --max-time 5 http://127.0.0.1:3301/ || true)"
+echo "MOMO_HEALTH_CODE $(curl -s -o /dev/null -w "%{http_code}" --max-time 5 http://127.0.0.1:5003/health || true)"
+docker ps --format "DOCKER {{.Names}}\t{{.Status}}" | head -80
+' 2>&1); then
+    fail "ssh 188 read-only check"
+    echo "$out"
+    return
+  fi
+  echo "$out"
+
+  grep -q "PORT5432 OPEN" <<<"$out" && ok "188 PostgreSQL port open" || fail "188 PostgreSQL port closed"
+  grep -q "accepting connections" <<<"$out" && ok "188 PostgreSQL accepting connections" || fail "188 PostgreSQL not accepting connections"
+  grep -q "REDIS PONG" <<<"$out" && ok "188 Redis PONG" || warn "188 Redis not confirmed"
+  grep -q "momo-db.*Restarting" <<<"$out" && warn "188 momo-db restarting" || ok "188 momo-db not in visible restart loop"
+  grep -Eq "SIGNOZ_CODE (200|302|307)" <<<"$out" && ok "188 SignOz HTTP reachable" || warn "188 SignOz HTTP not confirmed"
+  grep -q "MOMO_HEALTH_CODE 200" <<<"$out" && ok "188 momo health reachable" || warn "188 momo health not confirmed"
+}
+
+check_110() {
+  log_section "P0-110-REGISTRY-OBSERVABILITY"
+  local out
+  if ! out=$(ssh_cmd "wooo@192.168.0.110" '
+echo "HOST $(hostname) $(uptime)"
+echo "MEM $(free -h | awk "/Mem:/ {print \$2,\$3,\$7}")"
+echo "DOCKER_SYSTEMD $(systemctl is-active docker 2>/dev/null || true)"
+echo "HARBOR_CODE $(curl -s -o /dev/null -w "%{http_code}" --max-time 5 http://127.0.0.1:5000/v2/ || true)"
+echo "GITEA_CODE $(curl -s -o /dev/null -w "%{http_code}" --max-time 5 http://127.0.0.1:3001/ || true)"
+echo "PROM_CODE $(curl -s -o /dev/null -w "%{http_code}" --max-time 5 http://127.0.0.1:9090/-/ready || true)"
+echo "AM_CODE $(curl -s -o /dev/null -w "%{http_code}" --max-time 5 http://127.0.0.1:9093/-/healthy || true)"
+echo "SENTRY_CODE $(curl -s -o /dev/null -w "%{http_code}" --max-time 8 http://127.0.0.1:9000/ || true)"
+for u in $(systemctl list-units "actions.runner.*" --all --no-legend --plain 2>/dev/null | awk "{print \$1}"); do
+  systemctl show "$u" -p ActiveState -p SubState -p CPUQuotaPerSecUSec -p MemoryMax -p WatchdogUSec -p NRestarts | sed "s/^/RUNNER $u /"
+done
+docker ps --format "DOCKER {{.Names}}\t{{.Status}}" | head -120
+' 2>&1); then
+    fail "ssh 110 read-only check"
+    echo "$out"
+    return
+  fi
+  echo "$out"
+
+  grep -Eq "HARBOR_CODE (200|401)" <<<"$out" && ok "110 Harbor /v2 healthy code" || fail "110 Harbor not healthy"
+  grep -Eq "GITEA_CODE (200|302)" <<<"$out" && ok "110 Gitea reachable" || warn "110 Gitea not confirmed"
+  grep -q "PROM_CODE 200" <<<"$out" && ok "110 Prometheus ready" || warn "110 Prometheus not ready"
+  grep -q "AM_CODE 200" <<<"$out" && ok "110 Alertmanager healthy" || warn "110 Alertmanager not healthy"
+  grep -Eq "SENTRY_CODE (200|302|400)" <<<"$out" && ok "110 Sentry HTTP reachable" || warn "110 Sentry HTTP not confirmed"
+  grep -q "WatchdogUSec=0" <<<"$out" && ok "runner watchdog disabled on at least one unit" || warn "runner watchdog state not confirmed"
+  grep -q "sentry-self-hosted-clickhouse-1.*Restarting" <<<"$out" && warn "Sentry ClickHouse restarting" || ok "Sentry ClickHouse not visibly restarting"
+}
+
+check_k3s() {
+  log_section "P1-K3S"
+  local out local_kubectl_out
+  if ! out=$(ssh_cmd "wooo@192.168.0.120" '
+echo "HOST $(hostname) $(uptime)"
+echo "PG188_PORT $(nc -z -w 2 192.168.0.188 5432 >/dev/null 2>&1 && echo OPEN || echo CLOSED)"
+echo "SYSTEMD $(systemctl is-active k3s k3s-agent keepalived 2>/dev/null | tr "\n" " ")"
+kcmd() {
+  if [ -n "${REMOTE_SUDO_PASSWORD:-}" ]; then
+    printf "%s\n" "$REMOTE_SUDO_PASSWORD" | sudo -S -p "" kubectl "$@"
+  else
+    sudo -n kubectl "$@" 2>/dev/null || kubectl "$@"
+  fi
+}
+kcmd get nodes -o wide 2>/dev/null || true
+kcmd get pods -n awoooi-prod -o wide 2>/dev/null || true
+ip addr show | grep 192.168.0.125 || true
+' 2>&1); then
+    fail "ssh 120 k3s read-only check"
+    echo "$out"
+    return
+  fi
+  echo "$out"
+
+  if ! grep -q " Ready " <<<"$out"; then
+    local_kubectl_out=$(kubectl get nodes -o wide 2>/dev/null || true)
+    if [ -n "$local_kubectl_out" ]; then
+      echo "LOCAL_KUBECTL_FALLBACK"
+      echo "$local_kubectl_out"
+    fi
+  else
+    local_kubectl_out=""
+  fi
+
+  grep -q "PG188_PORT OPEN" <<<"$out" && ok "120 can reach 188 PostgreSQL port" || fail "120 cannot reach 188 PostgreSQL"
+  grep -q " Ready " <<<"$out$local_kubectl_out" && ok "K3s has Ready node output" || fail "K3s nodes not Ready or kubectl unavailable"
+  grep -q "192.168.0.125" <<<"$out" && ok "VIP 192.168.0.125 present on 120" || warn "VIP not confirmed on 120"
+}
+
+check_workload_and_alertchain() {
+  log_section "P2-WORKLOAD-ALERTCHAIN"
+  local api_code web_code alert_code
+  local out
+  if out=$(ssh_cmd "wooo@192.168.0.120" '
+api_code=$(curl -s -o /dev/null -w "%{http_code}" --max-time 5 http://192.168.0.125:32334/api/v1/health 2>/dev/null || true)
+web_code=$(curl -s -o /dev/null -w "%{http_code}" --max-time 5 http://192.168.0.125:32335/ 2>/dev/null || true)
+echo "API_CODE ${api_code:-000}"
+echo "WEB_CODE ${web_code:-000}"
+' 2>/dev/null); then
+    api_code=$(awk '/^API_CODE / {print $2}' <<<"$out")
+    web_code=$(awk '/^WEB_CODE / {print $2}' <<<"$out")
+  else
+    api_code=$(probe_http_code "http://192.168.0.125:32334/api/v1/health")
+    web_code=$(probe_http_code "http://192.168.0.125:32335/")
+    out="API_CODE $api_code
+WEB_CODE $web_code"
+  fi
+
+  echo "$out"
+
+  [[ "$api_code" =~ ^[23] ]] && ok "AWOOOI API reachable" || fail "AWOOOI API not reachable"
+  [[ "$web_code" =~ ^[23] ]] && ok "AWOOOI Web reachable" || warn "AWOOOI Web not confirmed"
+
+  if [ "$SEND_ALERT_TEST" -eq 1 ]; then
+    alert_code=$(ssh_cmd "wooo@192.168.0.120" 'curl -s -o /tmp/awoooi-alertchain.out -w "%{http_code}" --max-time 8 \
+      -X POST "http://192.168.0.125:32334/api/v1/webhooks/alertmanager" \
+      -H '"'"'Content-Type: application/json'"'"' \
+      -d '"'"'{"receiver":"cold-start-check","status":"firing","alerts":[{"status":"firing","labels":{"alertname":"ColdStartCheck","severity":"info"},"annotations":{"summary":"Cold start check"},"startsAt":"2026-05-05T11:00:00Z","endsAt":"0001-01-01T00:00:00Z","generatorURL":""}],"groupLabels":{},"commonLabels":{},"commonAnnotations":{},"externalURL":"","version":"4","groupKey":"cold-start-check"}'"'"' 2>/dev/null || echo "000"')
+    echo "ALERTCHAIN_CODE $alert_code"
+    [[ "$alert_code" =~ ^2 ]] && ok "Alertmanager webhook endpoint accepts POST" || warn "Alertmanager webhook E2E not confirmed"
+  else
+    warn "Alertmanager webhook POST skipped; rerun with --send-alert-test after API is ready"
+  fi
+}
+
+check_public_routes() {
+  log_section "P2-PUBLIC-ROUTES"
+  local awoooi_api_code awoooi_web_code momo_code momo_health_code
+  awoooi_api_code=$(probe_http_code "https://awoooi.wooo.work/api/v1/health")
+  awoooi_web_code=$(probe_http_code "https://awoooi.wooo.work/")
+  momo_code=$(probe_http_code "https://mo.wooo.work/")
+  momo_health_code=$(probe_http_code "https://mo.wooo.work/health")
+
+  echo "AWOOOI_PUBLIC_API_CODE $awoooi_api_code"
+  echo "AWOOOI_PUBLIC_WEB_CODE $awoooi_web_code"
+  echo "MOMO_PUBLIC_CODE $momo_code"
+  echo "MOMO_PUBLIC_HEALTH_CODE $momo_health_code"
+
+  [[ "$awoooi_api_code" =~ ^[23] ]] && ok "AWOOOI public API reachable" || warn "AWOOOI public API not confirmed"
+  [[ "$awoooi_web_code" =~ ^[23] ]] && ok "AWOOOI public web reachable" || warn "AWOOOI public web not confirmed"
+  [[ "$momo_code" =~ ^[23] ]] && ok "momo public route reachable" || warn "momo public route not confirmed"
+  [[ "$momo_health_code" =~ ^[23] ]] && ok "momo public health reachable" || warn "momo public health not confirmed"
+}
+
+check_schedules() {
+  log_section "P2-SCHEDULES"
+  local out
+
+  if out=$(ssh_cmd "ollama@192.168.0.188" '
+now=$(date +%s)
+echo "CRON_188 $(systemctl is-active cron 2>/dev/null || systemctl is-active crond 2>/dev/null || true)"
+for f in /home/ollama/node_exporter_textfiles/backup.prom /home/ollama/node_exporter_textfiles/docker_restart_count.prom /home/ollama/node_exporter_textfiles/docker_stats.prom; do
+  if [ -f "$f" ]; then
+    mt=$(stat -c %Y "$f")
+    echo "TEXTFILE_188 $(basename "$f") age=$((now - mt))"
+  else
+    echo "TEXTFILE_188 $(basename "$f") missing"
+  fi
+done
+if [ -f /home/ollama/node_exporter_textfiles/backup.prom ]; then
+  awk -v now="$now" "/^backup_110_last_success_timestamp / {printf \"BACKUP_110_AGE %d\\n\", now - int(\$2)}" /home/ollama/node_exporter_textfiles/backup.prom
+fi
+echo "SCHEDULER_REGISTERED $(docker logs --tail 200 momo-scheduler 2>&1 | grep -c "全部排程任務已註冊" || true)"
+' 2>&1); then
+    echo "$out"
+    grep -q "CRON_188 active" <<<"$out" && ok "188 cron active" || warn "188 cron not confirmed"
+    awk '/TEXTFILE_188 backup.prom age=/ {split($3,a,"="); exit !(a[2] < 90000)}' <<<"$out" && ok "188 backup textfile fresh enough" || warn "188 backup textfile stale or missing"
+    awk '/TEXTFILE_188 docker_restart_count.prom age=/ {split($3,a,"="); exit !(a[2] < 300)}' <<<"$out" && ok "188 docker restart exporter fresh" || warn "188 docker restart exporter stale"
+    awk '/TEXTFILE_188 docker_stats.prom age=/ {split($3,a,"="); exit !(a[2] < 300)}' <<<"$out" && ok "188 docker stats exporter fresh" || warn "188 docker stats exporter stale"
+    awk '/BACKUP_110_AGE / {exit !($2 < 90000)}' <<<"$out" && ok "188 backup-from-110 success within 25h" || warn "188 backup-from-110 success not confirmed"
+    awk '/SCHEDULER_REGISTERED / {exit !($2 > 0)}' <<<"$out" && ok "188 momo scheduler registered jobs" || warn "188 momo scheduler registration not confirmed"
+  else
+    warn "188 schedule check unavailable"
+    echo "$out"
+  fi
+
+  if out=$(ssh_cmd "wooo@192.168.0.110" '
+now=$(date +%s)
+echo "CRON_110 $(systemctl is-active cron 2>/dev/null || systemctl is-active crond 2>/dev/null || true)"
+echo "FAILED_UNITS_110 $(systemctl --failed --no-legend --plain 2>/dev/null | wc -l)"
+echo "MOMO_STARTUP_ENABLED $(systemctl is-enabled momo-startup-complete.service 2>/dev/null || true)"
+echo "STAGGERED_STARTUP_ENABLED $(systemctl is-enabled wooo-staggered-startup.service 2>/dev/null || true)"
+for f in /home/wooo/node_exporter_textfiles/docker_stats.prom /home/wooo/node_exporter_textfiles/systemd_units.prom; do
+  if [ -f "$f" ]; then
+    mt=$(stat -c %Y "$f")
+    echo "TEXTFILE_110 $(basename "$f") age=$((now - mt))"
+  else
+    echo "TEXTFILE_110 $(basename "$f") missing"
+  fi
+done
+' 2>&1); then
+    echo "$out"
+    grep -q "CRON_110 active" <<<"$out" && ok "110 cron active" || warn "110 cron not confirmed"
+    grep -q "FAILED_UNITS_110 0" <<<"$out" && ok "110 systemd has no failed units" || warn "110 systemd failed units remain"
+    grep -q "MOMO_STARTUP_ENABLED disabled" <<<"$out" && ok "110 stale momo startup unit disabled" || warn "110 stale momo startup unit not disabled"
+    grep -q "STAGGERED_STARTUP_ENABLED disabled" <<<"$out" && ok "110 stale staggered startup unit disabled" || warn "110 stale staggered startup unit not disabled"
+    awk '/TEXTFILE_110 docker_stats.prom age=/ {split($3,a,"="); exit !(a[2] < 300)}' <<<"$out" && ok "110 docker stats exporter fresh" || warn "110 docker stats exporter stale"
+    awk '/TEXTFILE_110 systemd_units.prom age=/ {split($3,a,"="); exit !(a[2] < 300)}' <<<"$out" && ok "110 systemd units exporter fresh" || warn "110 systemd units exporter stale"
+  else
+    warn "110 schedule check unavailable"
+    echo "$out"
+  fi
+
+  if out=$(ssh_cmd "wooo@192.168.0.120" '
+kcmd() {
+  if [ -n "${REMOTE_SUDO_PASSWORD:-}" ]; then
+    printf "%s\n" "$REMOTE_SUDO_PASSWORD" | sudo -S -p "" kubectl "$@"
+  else
+    sudo -n kubectl "$@" 2>/dev/null || kubectl "$@"
+  fi
+}
+echo "CRON_120 $(systemctl is-active cron 2>/dev/null || systemctl is-active crond 2>/dev/null || true)"
+kcmd get cronjobs -n awoooi-prod -o json | python3 -c "import json,sys; d=json.load(sys.stdin); items=d.get(\"items\", []); print(\"CRONJOB_COUNT\", len(items)); print(\"CRONJOB_SUSPENDED\", sum(1 for i in items if i.get(\"spec\",{}).get(\"suspend\")))"
+kcmd get jobs -n awoooi-prod -o json | python3 -c "import json,sys; d=json.load(sys.stdin); failed=0
+for j in d.get(\"items\", []):
+  if any(c.get(\"type\")==\"Failed\" and c.get(\"status\")==\"True\" for c in j.get(\"status\",{}).get(\"conditions\",[]) or []):
+    failed += 1
+print(\"FAILED_JOBS\", failed)"
+kcmd get pods -n awoooi-prod --no-headers 2>/dev/null | awk "\$3 !~ /^(Running|Completed)$/ {bad++} END {print \"BAD_PODS\", bad+0}"
+' 2>&1); then
+    echo "$out"
+    grep -q "CRON_120 active" <<<"$out" && ok "120 cron active" || warn "120 cron not confirmed"
+    awk '/CRONJOB_COUNT / {exit !($2 >= 4)}' <<<"$out" && ok "K8s AWOOOI CronJobs present" || warn "K8s AWOOOI CronJobs missing"
+    grep -q "CRONJOB_SUSPENDED 0" <<<"$out" && ok "K8s AWOOOI CronJobs unsuspended" || warn "K8s AWOOOI CronJob suspended"
+    grep -q "FAILED_JOBS 0" <<<"$out" && ok "K8s AWOOOI has no failed Jobs" || warn "K8s AWOOOI failed Jobs remain"
+    grep -q "BAD_PODS 0" <<<"$out" && ok "K8s AWOOOI pods Running/Completed only" || warn "K8s AWOOOI bad pod status remains"
+  else
+    warn "120 K8s schedule check unavailable"
+    echo "$out"
+  fi
+
+  if out=$(ssh_cmd "wooo@192.168.0.121" '
+echo "CRON_121 $(systemctl is-active cron 2>/dev/null || systemctl is-active crond 2>/dev/null || true)"
+crontab -l 2>/dev/null | grep -q "dr-drill.sh" && echo "DR_DRILL_CRON present" || echo "DR_DRILL_CRON missing"
+' 2>&1); then
+    echo "$out"
+    grep -q "CRON_121 active" <<<"$out" && ok "121 cron active" || warn "121 cron not confirmed"
+    grep -q "DR_DRILL_CRON present" <<<"$out" && ok "121 DR drill cron present" || warn "121 DR drill cron missing"
+  else
+    warn "121 schedule check unavailable"
+    echo "$out"
+  fi
+}
+
+summary() {
+  log_section "SUMMARY"
+  echo "PASS=$PASS WARN=$WARN BLOCKED=$FAIL"
+  if [ "$FAIL" -gt 0 ]; then
+    echo "Result: BLOCKED. Fix the first blocked gate before releasing runner/CD/AI auto-remediation."
+    exit 2
+  fi
+  if [ "$WARN" -gt 0 ]; then
+    echo "Result: DEGRADED. Core gates passed but warnings remain."
+    exit 1
+  fi
+  echo "Result: GREEN. Full stack is ready for controlled runner/CD release."
+}
+
+print_header
+check_network
+check_188
+check_110
+check_k3s
+check_workload_and_alertchain
+check_public_routes
+check_schedules
+summary

scripts/setup-guardrails.sh

@@ -92,10 +92,10 @@ fi
 echo ""
 echo "🤖 Step 6: Verifying Ollama connection..."
 
-OLLAMA_URL="http://192.168.0.188:11434/api/tags"
+OLLAMA_URL="${OLLAMA_URL:-http://192.168.0.111:11434/api/tags}"
 
 if curl -s --connect-timeout 5 "$OLLAMA_URL" > /dev/null 2>&1; then
-    echo "   ✅ Ollama reachable at 192.168.0.188:11434"
+    echo "   ✅ Ollama reachable at ${OLLAMA_URL}"
 
     # Check if llama3.2:8b is available
     MODELS=$(curl -s "$OLLAMA_URL" | grep -o '"name":"[^"]*"' || echo "")