fix(alerts): correct telegram execution truth

apps/api/src/services/approval_execution.py

@@ -36,6 +36,7 @@ from src.db.base import get_db_context
 from src.models.approval import ApprovalRequest
 from src.plugins.mcp.gateway import GatewayContext, McpGateway, McpGatewayError
 from src.plugins.mcp.interfaces import MCPToolResult
+from src.services.approval_action_classifier import is_no_action_approval_action
 from src.services.approval_db import get_approval_service, get_timeline_service
 from src.services.executor import ExecutionResult, OperationType, get_executor
 from src.services.operation_parser import parse_operation_from_action
@@ -165,6 +166,7 @@ class ApprovalExecutionService:
         # ADR-090 § 自動化動作回灌 (2026-04-19): 主流程開始即在 aol 留痕,
         # 結束時 update。不依賴 fire-and-forget,確保 33 件/7d approval 全部可觀測。
         _aol_op_id = await self._log_aol_started(approval)
+        await self._log_alert_execution_started(approval, aol_op_id=_aol_op_id)
         _aol_started_ms = time.time()
 
         service = get_approval_service()
@@ -228,15 +230,7 @@ class ApprovalExecutionService:
             # 2026-04-19 ogt + Claude Opus 4.7: 區分 NO_ACTION vs 真解析失敗
             # NO_ACTION 是 AI 刻意選的「純調查不破壞」,不該誤標 EXECUTION_FAILED
             # 污染 auto_execute 成功率 KPI (MASTER §7.1 #11)
-            _action_upper = (approval.action or "").upper()
-            _is_no_action = (
-                "NO_ACTION" in _action_upper
-                or "NO-ACTION" in _action_upper
-                or "NOACTION" in _action_upper
-                or "(未設)" in approval.action
-                or _action_upper.startswith("OBSERVE")
-                or _action_upper.startswith("INVESTIGATE")
-            )
+            _is_no_action = is_no_action_approval_action(approval.action)
 
             if _is_no_action:
                 logger.info(
@@ -246,13 +240,17 @@ class ApprovalExecutionService:
                     reason="NO_ACTION - 純調查/觀察類,不執行破壞動作",
                     path="no_action",
                 )
-                # 標為 SUCCESS (觀察/調查本身就是成功完成)
-                await service.update_execution_status(approval.id, success=True)
+                # 仍以 terminal success 關閉簽核，但 metadata 明確標記未執行修復。
+                await service.update_execution_status(
+                    approval.id,
+                    success=True,
+                    execution_kind="no_action",
+                )
                 await timeline.add_event(
                     event_type="exec",
                     status="success",
-                    title="✅ 純觀察類動作完成 (NO_ACTION)",
-                    description=f"Action: {approval.action[:120]}",
+                    title="ℹ️ 純觀察類動作已記錄（未執行修復）",
+                    description=f"Action: {(approval.action or '')[:120]}",
                     actor="leWOOOgo",
                     actor_role="executor",
                     approval_id=str(approval.id),
@@ -269,7 +267,22 @@ class ApprovalExecutionService:
                     op_id=_aol_op_id,
                     status="success",
                     duration_ms=int((time.time() - _aol_started_ms) * 1000),
-                    output={"reason": "NO_ACTION", "action": approval.action[:200]},
+                    output={
+                        "reason": "NO_ACTION",
+                        "execution_kind": "no_action",
+                        "repair_executed": False,
+                        "action": (approval.action or "")[:200],
+                    },
+                )
+                await self._log_alert_execution_completed(
+                    approval,
+                    success=True,
+                    execution_kind="no_action",
+                    duration_ms=int((time.time() - _aol_started_ms) * 1000),
+                    output={
+                        "reason": "NO_ACTION",
+                        "repair_executed": False,
+                    },
                 )
                 # F2 (2026-05-07 ogt + Claude Sonnet 4.6 + Codex):
                 # NO_ACTION 路徑要把 incident 推到 RESOLVED，否則 incident 永遠卡
@@ -336,6 +349,13 @@ class ApprovalExecutionService:
                 duration_ms=int((time.time() - _aol_started_ms) * 1000),
                 error=f"parse_fail: {approval.action[:300]}",
             )
+            await self._log_alert_execution_completed(
+                approval,
+                success=False,
+                execution_kind="parse_failed",
+                duration_ms=int((time.time() - _aol_started_ms) * 1000),
+                error_message=f"Could not parse operation type from action: {approval.action[:150]}",
+            )
             return False  # 解析失敗 → 執行未發生
 
         executor = get_executor()
@@ -553,6 +573,20 @@ class ApprovalExecutionService:
                     "total_attempts": total_attempts,
                 },
             )
+            await self._log_alert_execution_completed(
+                approval,
+                success=True,
+                execution_kind=operation_type.value,
+                duration_ms=int((time.time() - _aol_started_ms) * 1000),
+                output={
+                    "operation_type": operation_type.value,
+                    "resource_name": resource_name,
+                    "namespace": namespace,
+                    "executor_duration_ms": result.duration_ms,
+                    "total_attempts": total_attempts,
+                    "repair_executed": True,
+                },
+            )
             return True  # K8s 執行成功
 
         else:
@@ -654,6 +688,22 @@ class ApprovalExecutionService:
                 error=result.error,
                 stderr=result.error,  # E6 stderr 回灌 — 給 retry/Playbook 負向強化用
             )
+            await self._log_alert_execution_completed(
+                approval,
+                success=False,
+                execution_kind=operation_type.value,
+                duration_ms=int((time.time() - _aol_started_ms) * 1000),
+                output={
+                    "operation_type": operation_type.value,
+                    "resource_name": resource_name,
+                    "namespace": namespace,
+                    "executor_duration_ms": result.duration_ms,
+                    "total_attempts": total_attempts,
+                    "repair_attempted": True,
+                    "repair_executed": False,
+                },
+                error_message=result.error,
+            )
             return False  # K8s 執行失敗
 
     async def _execute_ssh_host_action(
@@ -919,7 +969,14 @@ class ApprovalExecutionService:
             except Exception:
                 pass
 
-            if success:
+            no_action = success and is_no_action_approval_action(approval.action)
+            if no_action:
+                text = (
+                    f"ℹ️ <b>已記錄觀察，未執行修復</b>\n"
+                    f"<code>{(approval.action or '')[:180]}</code>"
+                    f"{km_info}"
+                )
+            elif success:
                 text = (
                     f"✅ <b>執行成功</b>\n"
                     f"<code>{(approval.action or '')[:180]}</code>"
@@ -948,8 +1005,34 @@ class ApprovalExecutionService:
                 incident_id=approval.incident_id,
                 approval_id=str(approval.id),
                 success=success,
+                no_action=no_action,
                 orig_msg_id=orig_msg_id,
             )
+            try:
+                from src.repositories.alert_operation_log_repository import (
+                    get_alert_operation_log_repository,
+                )
+
+                await get_alert_operation_log_repository().append(
+                    "TELEGRAM_RESULT_SENT",
+                    incident_id=approval.incident_id,
+                    approval_id=str(approval.id),
+                    actor="approval_execution",
+                    action_detail="telegram_execution_result_sent",
+                    success=success,
+                    error_message=error,
+                    context={
+                        "reply_to_message_id": orig_msg_id,
+                        "execution_kind": "no_action" if no_action else "execution",
+                        "repair_executed": not no_action and success,
+                    },
+                )
+            except Exception as _log_e:
+                logger.warning(
+                    "alert_op_telegram_result_write_failed",
+                    approval_id=str(approval.id),
+                    error=str(_log_e),
+                )
         except Exception as e:
             logger.warning(
                 "push_execution_result_failed",
@@ -1592,6 +1675,85 @@ class ApprovalExecutionService:
     # 22 筆 notification_formatted。修復後每次執行都留痕。
     # =========================================================================
 
+    async def _log_alert_execution_started(
+        self,
+        approval: ApprovalRequest,
+        *,
+        aol_op_id: str | None,
+    ) -> None:
+        """Append immutable alert_operation_log start event for manual execution."""
+        try:
+            from src.repositories.alert_operation_log_repository import (
+                get_alert_operation_log_repository,
+            )
+
+            await get_alert_operation_log_repository().append(
+                "EXECUTION_STARTED",
+                incident_id=approval.incident_id,
+                approval_id=str(approval.id),
+                actor="approval_execution",
+                action_detail="approval_execution_started",
+                success=None,
+                context={
+                    "action": (approval.action or "")[:500],
+                    "automation_operation_id": aol_op_id,
+                    "execution_kind": (
+                        "no_action"
+                        if is_no_action_approval_action(approval.action)
+                        else "executable"
+                    ),
+                    "repair_attempted": False,
+                    "repair_executed": False,
+                },
+            )
+        except Exception as e:
+            logger.warning(
+                "alert_op_execution_started_write_failed",
+                approval_id=str(approval.id),
+                incident_id=approval.incident_id,
+                error=str(e),
+            )
+
+    async def _log_alert_execution_completed(
+        self,
+        approval: ApprovalRequest,
+        *,
+        success: bool,
+        execution_kind: str,
+        duration_ms: int,
+        output: dict | None = None,
+        error_message: str | None = None,
+    ) -> None:
+        """Append immutable alert_operation_log completion event for manual execution."""
+        try:
+            from src.repositories.alert_operation_log_repository import (
+                get_alert_operation_log_repository,
+            )
+
+            context = {
+                "action": (approval.action or "")[:500],
+                "duration_ms": duration_ms,
+                "execution_kind": execution_kind,
+                **(output or {}),
+            }
+            await get_alert_operation_log_repository().append(
+                "EXECUTION_COMPLETED",
+                incident_id=approval.incident_id,
+                approval_id=str(approval.id),
+                actor="approval_execution",
+                action_detail=f"approval_execution_{execution_kind}",
+                success=success,
+                error_message=(error_message or "")[:2000] if error_message else None,
+                context=context,
+            )
+        except Exception as e:
+            logger.warning(
+                "alert_op_execution_completed_write_failed",
+                approval_id=str(approval.id),
+                incident_id=approval.incident_id,
+                error=str(e),
+            )
+
     async def _log_aol_started(self, approval: ApprovalRequest) -> str | None:
         """
         在 automation_operation_log 寫一筆 'pending' 紀錄,回傳 op_id 供 _log_aol_completed 更新。