feat(governance): normalize AI治理告警輸出與元告警解析度

apps/api/src/services/failover_alerter.py

@@ -102,16 +102,46 @@ class FailoverAlerter:
             logger.debug("governance_alert_dedup_skipped", event_type=event_type)
             return
 
-        # 格式化 payload 為可讀字串（key=value，換行分隔）
-        detail_lines = "\n".join(
-            f"{_escape_md(str(k))}：{_escape_md(str(v))}"
-            for k, v in payload.items()
-        )
-        msg = (
-            f"*AI 治理警報*\n\n"
-            f"類型：{_escape_md(event_type)}\n\n"
-            f"{detail_lines}"
-        )
+        status = _escape_md(str(payload.get("status", "warning")))
+        impact = _as_dict(payload.get("impact"))
+        remediation = _as_dict(payload.get("remediation"))
+        actionable = _as_dict(payload.get("actionable"))
+
+        impact_lines = _lines_from_dict(impact, max_items=12, compact=True)
+        remediation_lines = _lines_from_list(remediation.get("items"))
+        remediation_next_action = remediation.get("next_action")
+        remediation_hint = remediation.get("hint")
+        actionable_lines = _lines_from_list(actionable.get("items"))
+
+        next_action_line = ""
+        if remediation_next_action:
+            next_action_line = f"\n  下一步：{_escape_md(str(remediation_next_action))}"
+        if remediation_hint:
+            next_action_line += f"\n  提示：{_escape_md(str(remediation_hint))}"
+
+        sections: list[str] = [
+            "⚠️ *AI 治理警報*",
+            f"\n類型：{_escape_md(event_type)}",
+            f"狀態：{status}",
+        ]
+        if impact_lines:
+            sections.append(f"\n*影響*\n{impact_lines}")
+        if remediation_lines or next_action_line:
+            sections.append(f"\n*修復方向*")
+            if remediation_lines:
+                sections.append(remediation_lines)
+            if next_action_line:
+                sections.append(next_action_line)
+        if actionable_lines:
+            sections.append(f"\n*可直接自動化*\n{actionable_lines}")
+
+        fallback_items = _fallback_pairs(payload, keep={"status", "impact", "remediation", "actionable"})
+        if fallback_items:
+            sections.append(
+                "\n*欄位快覽（備援）*\n" + "\n".join(fallback_items)
+            )
+
+        msg = "\n".join(sections)
         await self._send(msg)
         logger.info("governance_alert_sent", event_type=event_type)
 
@@ -259,6 +289,46 @@ def _escape_md(text: str) -> str:
     return text
 
 
+def _as_dict(value: Any) -> dict[str, Any]:
+    return value if isinstance(value, dict) else {}
+
+
+def _lines_from_dict(data: dict[str, Any], max_items: int = 20, compact: bool = False) -> str:
+    if not data:
+        return ""
+    rows = []
+    idx = 0
+    for k in sorted(data.keys()) if isinstance(data, dict) else []:
+        if idx >= max_items:
+            break
+        rows.append(f"{_escape_md(str(k))}：{_escape_md(str(data.get(k)))}")
+        idx += 1
+    if compact and len(rows) >= max_items:
+        rows.append("...（更多欄位略）")
+    return "\n".join(f"  {line}" for line in rows)
+
+
+def _lines_from_list(value: Any) -> str:
+    if not isinstance(value, list):
+        return ""
+    return "\n".join(
+        f"  {idx + 1}. {_escape_md(str(item))}"
+        for idx, item in enumerate(value)
+    )
+
+
+def _fallback_pairs(payload: dict[str, Any], keep: set[str] | None = None) -> list[str]:
+    if not isinstance(payload, dict):
+        return []
+    keep = set(keep or set())
+    rows = []
+    for key in sorted(payload.keys()):
+        if key in keep:
+            continue
+        rows.append(f"{_escape_md(str(key))}：{_escape_md(str(payload.get(key)))}")
+    return rows
+
+
 # =============================================================================
 # Singleton
 # =============================================================================

apps/api/src/services/governance_agent.py

@@ -142,6 +142,13 @@ class GovernanceAgent:
                         ],
                         "sample_playbook_ids": kept_ids[:10],
                     },
+                    "drifted_count": len(drifted),
+                    "auto_deprecated_count": len(auto_deprecated_ids),
+                    "auto_deprecated_ids": auto_deprecated_ids[:10],
+                    "playbook_ids": kept_ids[:10],
+                    "total_playbooks": total,
+                    "threshold": TRUST_DRIFT_THRESHOLD,
+                    "auto_deprecate_after_days": TRUST_DRIFT_AUTO_DEPRECATE_AFTER_DAYS,
                 },
             )
 
@@ -215,6 +222,11 @@ class GovernanceAgent:
                             "安排至少 2 位 owner 對 stale條目做快速人工審核",
                         ],
                     },
+                    "stale_count": stale,
+                    "total_count": total,
+                    "stale_ratio": round(ratio, 3),
+                    "threshold": KM_STALE_RATIO,
+                    "stale_days": KM_STALE_DAYS,
                 },
             )
 
@@ -260,6 +272,27 @@ class GovernanceAgent:
             await self._alert(
                 "llm_hallucination",
                 {
+                    "status": "warning",
+                    "impact": {
+                        "failed_count": failed,
+                        "total_checked": total,
+                        "hallucination_rate": round(rate, 3),
+                        "threshold": HALLUCINATION_RATE_THRESHOLD,
+                    },
+                    "remediation": {
+                        "items": [
+                            "檢核 AI 建議來源與 evidence snapshot 一致性",
+                            "檢視最近 incident 的 verifier 輸入欄位是否缺失關鍵上下文",
+                        ],
+                        "next_action": "run_knowledge_gap_audit",
+                        "hint": "高失敗率通常表示 evidence 收斂流程退化或資料欄位解讀偏差",
+                    },
+                    "actionable": {
+                        "items": [
+                            "啟動 `playbook_evidence` 對齊補償流程",
+                            "調整 verify timeout 與降級策略，避免過度信任低品質證據",
+                        ],
+                    },
                     "failed_count": failed,
                     "total_checked": total,
                     "hallucination_rate": round(rate, 3),
@@ -304,6 +337,27 @@ class GovernanceAgent:
             await self._alert(
                 "execution_blast_radius",
                 {
+                    "status": "warning",
+                    "impact": {
+                        "failed_count": failed,
+                        "total_executions": total,
+                        "failure_rate": round(rate, 3),
+                        "threshold": EXECUTION_FAIL_RATE_THRESHOLD,
+                    },
+                    "remediation": {
+                        "items": [
+                            "鎖定失敗 playbook 清單，關閉高風險自動執行",
+                            "比對 incident evidence 與 post_execution_verification 失敗原因",
+                        ],
+                        "next_action": "pause_auto_repair_for_top_failing_playbooks",
+                        "hint": "可能是 auto_repair_playbook 與 runtime 版本/環境脫節",
+                    },
+                    "actionable": {
+                        "items": [
+                            "跑 `run_self_check` 快照與失敗 playbook 熱點報表",
+                            "必要時啟用 emergency fallback 路由進人工審核",
+                        ],
+                    },
                     "failed_count": failed,
                     "total_executions": total,
                     "failure_rate": round(rate, 3),
@@ -548,9 +602,25 @@ class GovernanceAgent:
                 await self._alert(
                     "governance_self_failure",
                     {
-                        "failed_checks": failed_checks,
-                        "total_checks": 5,  # 2026-04-27 P3.4 by Claude — 加入 slo_compliance 後共 5 項
-                        "errors": {k: results[k].get("error") for k in failed_checks},
+                        "status": "critical",
+                        "impact": {
+                            "failed_checks": failed_checks,
+                            "total_checks": 5,  # 2026-04-27 P3.4 by Claude — 加入 slo_compliance 後共 5 項
+                            "errors": {k: results[k].get("error") for k in failed_checks},
+                        },
+                        "remediation": {
+                            "items": [
+                                "暫停非關鍵治理自動化接收鏈路",
+                                "聚焦治理執行路徑錯誤並補齊 fallback",
+                            ],
+                            "next_action": "investigate_governance_pipeline_health",
+                        },
+                        "actionable": {
+                            "items": [
+                                "檢查 GovernanceAgent run loop 是否完整執行 5 個項目",
+                                "確認 DB 寫入與 Prometheus fetch 未被上游干擾",
+                            ],
+                        },
                     },
                 )
             except Exception:

apps/api/src/services/telegram_gateway.py

@@ -2603,12 +2603,12 @@ class TelegramGateway:
             f"━━━━━━━━━━━━━━━━━━━\n"
             f"📋 <code>{html.escape(incident_id)}</code>\n"
             f"🚨 異常元件：<b>{html.escape(alertname)}</b>\n"
-            f"🎯 診斷結果：{html.escape(diagnosis[:100])}\n"
+            f"🎯 診斷結果：{html.escape(_smart_truncate(diagnosis, 320))}\n"
         )
         if system_impact:
-            text += f"\n🧠 <b>系統影響</b>\n{html.escape(system_impact[:150])}\n"
+            text += f"\n🧠 <b>系統影響</b>\n{html.escape(_smart_truncate(system_impact, 320))}\n"
         if probable_cause:
-            text += f"└─ 可能根因：{html.escape(probable_cause[:100])}\n"
+            text += f"└─ 可能根因：{html.escape(_smart_truncate(probable_cause, 320))}\n"
 
         # 2026-04-16 ogt: 移除 flywheel_diag / flywheel_dashboard (3-part ghost button，無 handler)
         # 鐵律: 寧可沒按鈕，不可有死按鈕 (feedback_no_ghost_buttons.md)