feat(k8s): add Worker HPA + fix non-AI confidence values

Wave 2 Deployment:
- Worker HPA: min:1 max:3, CPU 70%, Memory 80%
- 前置條件: XCLAIM + terminationGracePeriodSeconds:90 (Wave 1 ✅)
- 比 API/Web 更保守的擴縮策略 (120s up, 600s down)

Confidence Fix:
- 非 AI 分析來源 (fallback/playbook/historical/consensus) 設 confidence=0.0
- 避免混淆 AI 信心度與其他指標 (成功率/相似度)
- 涉及: github_webhook, decision_manager, intent_classifier, learning_service

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

apps/api/src/api/v1/github_webhook.py

@@ -1128,7 +1128,7 @@ async def call_openclaw_ci_diagnosis(
                     auto_fixable=False,
                     risk_level="medium",
                     analyzed_by="fallback",
-                    confidence=0.3,
+                    confidence=0.0,  # 🔴 Fallback 不是 AI 分析
                 )
 
     except httpx.TimeoutException:
@@ -1141,7 +1141,7 @@ async def call_openclaw_ci_diagnosis(
             auto_fixable=False,
             risk_level="low",
             analyzed_by="fallback",
-            confidence=0.1,
+            confidence=0.0,  # 🔴 Fallback 不是 AI 分析
         )
     except Exception as e:
         logger.exception("openclaw_ci_diagnosis_error", error=str(e))

apps/api/src/services/decision_manager.py

@@ -677,7 +677,7 @@ class DecisionManager:
                 "description": playbook.description,
                 "risk_level": playbook.repair_steps[0].risk_level.value.lower() if playbook.repair_steps else "medium",
                 "reasoning": f"Playbook 匹配 ({best_match.similarity_score:.0%} 相似度, {playbook.success_rate:.0%} 成功率): {best_match.reason}",
-                "confidence": min(best_match.similarity_score, playbook.success_rate) if playbook.total_executions > 0 else best_match.similarity_score,
+                "confidence": 0.0,  # 🔴 Playbook RAG 匹配不是 AI 分析，信心度設 0
                 "matched_symptoms": best_match.matched_symptoms,
                 "from_cache": False,
             }
@@ -839,7 +839,7 @@ class DecisionManager:
                 "risk_level": consensus_result.risk_level,
                 "kubectl_command": consensus_result.recommended_kubectl,
                 "reasoning": consensus_result.final_reasoning,
-                "confidence": consensus_result.consensus_score,
+                "confidence": 0.0,  # 🔴 Consensus Engine 共識分數不是 AI 信心度，設 0
                 "agent_count": len(consensus_result.opinions),
                 "dissenting_opinions": consensus_result.dissenting_opinions,
                 "from_cache": False,

apps/api/src/services/intent_classifier.py

@@ -518,11 +518,11 @@ class IntentClassifier:
         """
         # TODO: 整合 Ollama qwen2.5:1b (Phase 13.4)
         # 預計使用 text 呼叫 Ollama API 進行分類
-        # 目前先返回低信心度 UNKNOWN，規則引擎已能處理大部分情況
+        # 目前先返回 UNKNOWN，規則引擎已能處理大部分情況
         del text  # 預留給 LLM 分類使用，避免 unused-parameter 警告
         return IntentResult(
             intent=IntentType.UNKNOWN,
-            confidence=0.3,
+            confidence=0.0,  # 🔴 LLM 未啟用，非 AI 分析
             method="llm",
             matched_keywords=[],
             detected_resources=[],

apps/api/src/services/learning_service.py

@@ -544,12 +544,13 @@ class LearningService:
 
         return {
             "action": best["action"],
-            "confidence": best["success_rate"],
+            "confidence": 0.0,  # 🔴 歷史學習不是 AI 分析，信心度設 0
             "tier": best["tier"],
             "based_on": f"{best['total_samples']} 次歷史數據",
             "avg_execution_time": best["avg_execution_time"],
+            "success_rate": best["success_rate"],  # 保留原始成功率作為參考
             "alternatives": [
-                {"action": a["action"], "confidence": a["success_rate"], "tier": a["tier"]}
+                {"action": a["action"], "confidence": 0.0, "success_rate": a["success_rate"], "tier": a["tier"]}
                 for a in alternatives
             ],
         }
@@ -589,12 +590,12 @@ class LearningService:
         """預設推薦 (無歷史數據時)"""
         return {
             "action": "restart_pod",
-            "confidence": 0.3,
+            "confidence": 0.0,  # 🔴 預設推薦不是 AI 分析，信心度設 0
             "tier": 1,
             "based_on": "無歷史數據，使用預設",
             "avg_execution_time": 30.0,
             "alternatives": [
-                {"action": "delete_pod", "confidence": 0.3, "tier": 1},
+                {"action": "delete_pod", "confidence": 0.0, "tier": 1},
             ],
         }
 

k8s/awoooi-prod/12-hpa.yaml

@@ -3,6 +3,7 @@
 # =============================================================================
 # K3.2 2026-03-28: HPA for API/Web (based on VPA recommendations)
 # P3 2026-03-29: maxReplicas 4 → 6 (首席架構師建議)
+# K-Worker 2026-03-29: Worker HPA (min:1 max:3, CPU 指標)
 # Deployed by: Claude Code (首席架構師)
 # VPA 建議: API target 100m, Web target 63m
 # =============================================================================
@@ -94,3 +95,54 @@ spec:
         - type: Pods
           value: 1
           periodSeconds: 120
+---
+# =============================================================================
+# Worker HPA（K-Worker 2026-03-29）
+# =============================================================================
+# 前置條件: ADR-038/039 Wave 1 完成（XCLAIM + terminationGracePeriodSeconds:90）
+# 說明: Worker 消費 Redis Streams，使用 CPU 指標（未來可升級 KEDA）
+# =============================================================================
+apiVersion: autoscaling/v2
+kind: HorizontalPodAutoscaler
+metadata:
+  name: awoooi-worker-hpa
+  namespace: awoooi-prod
+  labels:
+    app.kubernetes.io/name: awoooi
+    app.kubernetes.io/component: worker
+  annotations:
+    description: "Worker 水平自動擴展 (1-3 replicas, 70% CPU)"
+    note: "未來可升級 KEDA Redis Stream 指標按 Queue 長度擴縮"
+spec:
+  scaleTargetRef:
+    apiVersion: apps/v1
+    kind: Deployment
+    name: awoooi-worker
+  minReplicas: 1
+  maxReplicas: 3
+  metrics:
+    - type: Resource
+      resource:
+        name: cpu
+        target:
+          type: Utilization
+          averageUtilization: 70
+    - type: Resource
+      resource:
+        name: memory
+        target:
+          type: Utilization
+          averageUtilization: 80
+  behavior:
+    scaleUp:
+      stabilizationWindowSeconds: 120
+      policies:
+        - type: Pods
+          value: 1
+          periodSeconds: 120
+    scaleDown:
+      stabilizationWindowSeconds: 600
+      policies:
+        - type: Pods
+          value: 1
+          periodSeconds: 300