diff --git a/apps/api/src/services/approval_execution.py b/apps/api/src/services/approval_execution.py
index 809752c47..eaee13e45 100644
--- a/apps/api/src/services/approval_execution.py
+++ b/apps/api/src/services/approval_execution.py
@@ -42,7 +42,7 @@ from src.services.approval_db import get_approval_service, get_timeline_service
 from src.services.awooop_deeplinks import incident_truth_chain_button_row
 from src.services.executor import ExecutionResult, OperationType, get_executor
 from src.services.operation_parser import parse_operation_from_action
-from src.services.operator_outcome import build_operator_outcome
+from src.services.operator_outcome import build_operator_outcome, normalize_operator_outcome
 
 if TYPE_CHECKING:
     from src.services.notifications import ExecutionStatus
@@ -1265,17 +1265,17 @@ class ApprovalExecutionService:
             )
             outcome = truth_chain.get("operator_outcome")
             if isinstance(outcome, dict):
-                return outcome
+                return normalize_operator_outcome(outcome) or outcome
             quality = truth_chain.get("automation_quality")
             if isinstance(quality, dict) and isinstance(quality.get("operator_outcome"), dict):
-                return quality["operator_outcome"]
-            return build_operator_outcome(
+                return normalize_operator_outcome(quality["operator_outcome"]) or quality["operator_outcome"]
+            return normalize_operator_outcome(build_operator_outcome(
                 truth_status=truth_chain.get("truth_status")
                 if isinstance(truth_chain.get("truth_status"), dict)
                 else {},
                 automation_quality=quality if isinstance(quality, dict) else {},
                 source_id=incident_id,
-            )
+            )) or {}
         except Exception as exc:
             logger.warning(
                 "execution_result_outcome_fetch_failed",
@@ -1311,9 +1311,9 @@ class ApprovalExecutionService:
         elif state == "completed_verified":
             icon = "✅"
             title = "修復完成並已驗證"
-        elif state == "diagnostic_only_manual_review":
+        elif state == "diagnostic_only_ai_repair_required":
             icon = "⚠️"
-            title = "已記錄診斷，尚未證明修復"
+            title = "已記錄診斷，AI 修復候選補齊中"
         elif needs_human:
             icon = "⚠️"
             title = "已執行，但需人工確認"
diff --git a/apps/api/src/services/operator_outcome.py b/apps/api/src/services/operator_outcome.py
index 1dc8b0e53..300872b31 100644
--- a/apps/api/src/services/operator_outcome.py
+++ b/apps/api/src/services/operator_outcome.py
@@ -11,6 +11,66 @@ from typing import Any
 
 _ACTION_REQUIRED_CHANNELS = ("telegram_sre_war_room", "awooop_operator_console")
 _RESULT_ONLY_CHANNELS = ("telegram_result_reply", "awooop_operator_console")
+_LEGACY_AI_CONTROLLED_STATE_REWRITE = {
+    "diagnostic_only_manual_review": (
+        "diagnostic_only_ai_repair_required",
+        "auto_generate_repair_candidate_from_diagnostic_evidence",
+        "只完成診斷/觀察；AI 已排入 PlayBook / transport / verifier 修復候選",
+    ),
+    "verification_degraded_manual_required": (
+        "verification_degraded_ai_verifier_required",
+        "run_verifier_or_rollback_candidate",
+        "驗證退化；AI 已排入 verifier / rollback 判定",
+    ),
+    "execution_unverified_manual_required": (
+        "execution_unverified_controlled_verifier_required",
+        "run_post_execution_verifier_or_rollback",
+        "已執行但缺少驗證結果；AI 進入 verifier / rollback 判定",
+    ),
+    "no_action_manual_review": (
+        "no_action_ai_candidate_required",
+        "collect_evidence_or_generate_playbook_candidate",
+        "AI 未找到可執行修復；已排入 evidence / PlayBook 補強與 controlled policy 判定",
+    ),
+    "approval_expired_manual_review": (
+        "approval_expired_ai_retry",
+        "ai_retry_or_rebuild_controlled_packet",
+        "舊審批已逾期；AI 重新建立受控處置包",
+    ),
+    "write_observed_manual_review": (
+        "write_observed_ai_verifier_or_rollback",
+        "auto_verify_or_rollback_observed_write",
+        "補救證據出現寫入旗標；AI 進入 verifier / rollback 判定",
+    ),
+    "blocked_manual_required": (
+        "blocked_ai_repair_required",
+        "auto_repair_blocker_or_connector_then_retry",
+        "自動化流程受阻；AI 已排入 blocker / connector 修復",
+    ),
+}
+_BREAK_GLASS_BLOCKER_TOKENS = (
+    "secret",
+    "token",
+    "private_key",
+    "authorization_header",
+    "drop",
+    "truncate",
+    "restore",
+    "prune",
+    "remote_delete",
+    "reboot",
+    "node_drain",
+    "firewall_cutover",
+    "credentialed_exploit",
+    "external_active_scan",
+    "paid_provider",
+    "cost_limit",
+    "force_push",
+    "repo_delete",
+    "ref_delete",
+    "raw_runtime_secret_volume",
+    "break_glass",
+)
 
 
 def _safe_int(value: Any) -> int:
@@ -27,6 +87,69 @@ def _first_text(values: list[Any]) -> str | None:
     return None
 
 
+def _requires_break_glass(blockers: list[str]) -> bool:
+    haystack = " ".join(str(item).lower() for item in blockers)
+    return any(token in haystack for token in _BREAK_GLASS_BLOCKER_TOKENS)
+
+
+def _canonical_operator_state(state: str) -> str:
+    rewritten = _LEGACY_AI_CONTROLLED_STATE_REWRITE.get(state)
+    return rewritten[0] if rewritten else state
+
+
+def normalize_operator_outcome(outcome: dict[str, Any] | None) -> dict[str, Any] | None:
+    """Rewrite legacy human-gate outcomes to the current controlled-AI policy.
+
+    Older truth-chain rows can contain states such as ``*_manual_required``.
+    They remain useful as history, but the operator-facing contract must not
+    present them as the live product path unless a break-glass blocker exists.
+    """
+    if not isinstance(outcome, dict):
+        return outcome
+    state = str(outcome.get("state") or "")
+    rewrite = _LEGACY_AI_CONTROLLED_STATE_REWRITE.get(state)
+    if not rewrite:
+        return outcome
+
+    normalized = dict(outcome)
+    new_state, next_action, summary = rewrite
+    blockers = [
+        str(item)
+        for item in normalized.get("blockers", [])
+        if item
+    ] if isinstance(normalized.get("blockers"), list) else []
+    break_glass = _requires_break_glass(blockers)
+    normalized["legacy_state"] = state
+    normalized["state"] = "break_glass_required" if break_glass else new_state
+    normalized["summary_zh"] = (
+        "命中硬阻擋，需 break-glass 或外部授權"
+        if break_glass
+        else summary
+    )
+    normalized["needs_human"] = break_glass
+    normalized["human_action_required"] = break_glass
+    normalized["next_action"] = (
+        "break_glass_or_external_authorization"
+        if break_glass
+        else next_action
+    )
+    normalized["human_action_reason"] = (
+        normalized.get("human_action_reason")
+        if break_glass
+        else "legacy_manual_gate_converted_to_controlled_ai_policy"
+    )
+    notification = (
+        dict(normalized.get("notification"))
+        if isinstance(normalized.get("notification"), dict)
+        else {}
+    )
+    notification["mode"] = "action_required"
+    notification["channels"] = list(_ACTION_REQUIRED_CHANNELS)
+    notification["reason"] = str(normalized["human_action_reason"])
+    normalized["notification"] = notification
+    return normalized
+
+
 def _verification_blocker_code(value: Any) -> str:
     status = str(value or "missing").strip().lower()
     if status in {"", "none", "null", "missing", "--"}:
@@ -96,6 +219,7 @@ def _build_execution_result(
     verification: str,
 ) -> dict[str, Any]:
     """Describe execution completion separately from remediation outcome."""
+    state = _canonical_operator_state(state)
     approval_status = "unknown"
     completion_status = "unknown"
     command_status = "unknown"
@@ -120,29 +244,29 @@ def _build_execution_result(
         failure_status = "command_failed"
         summary = "已失敗：修復指令執行失敗，AI 已排入 rollback / repair 候選"
         terminal = False
-    elif state == "diagnostic_only_manual_review":
-        approval_status = "completed"
-        completion_status = "completed_no_repair"
+    elif state == "diagnostic_only_ai_repair_required":
+        approval_status = "auto_authorized"
+        completion_status = "diagnostic_completed_ai_repair_queued"
         command_status = "diagnostic_completed" if has_nonrepair_operation else "skipped_no_action"
-        repair_status = "not_executed"
+        repair_status = "playbook_or_transport_repair_required"
         failure_status = "no_command_failed"
-        summary = "流程已完成：只完成診斷/觀察，沒有修復指令成功或失敗"
-        terminal = True
-    elif state == "verification_degraded_manual_required":
-        approval_status = "completed"
-        completion_status = "completed_verification_degraded"
-        command_status = "succeeded" if has_repair_execution else "diagnostic_completed"
-        repair_status = "verification_degraded"
-        failure_status = "no_command_failed"
-        summary = "已執行，但驗證結果退化；需人工確認是否真的修復"
+        summary = "只完成診斷/觀察；AI 已排入修復候選補齊"
         terminal = False
-    elif state == "execution_unverified_manual_required":
-        approval_status = "completed"
+    elif state == "verification_degraded_ai_verifier_required":
+        approval_status = "auto_authorized"
+        completion_status = "completed_verification_degraded_ai_recovery"
+        command_status = "succeeded" if has_repair_execution else "diagnostic_completed"
+        repair_status = "ai_verifier_or_rollback_required"
+        failure_status = "no_command_failed"
+        summary = "已執行但驗證退化；AI 已排入 verifier / rollback 判定"
+        terminal = False
+    elif state == "execution_unverified_controlled_verifier_required":
+        approval_status = "auto_authorized"
         completion_status = "completed_unverified"
         command_status = "succeeded"
-        repair_status = "unverified"
+        repair_status = "ai_verifier_or_rollback_required"
         failure_status = "no_command_failed"
-        summary = "已執行成功，但缺少修復驗證結果"
+        summary = "已執行成功但缺少修復驗證結果；AI 進入 verifier / rollback 判定"
         terminal = False
     elif state == "apply_candidate_owner_review_ready":
         approval_status = "auto_authorized_preflight"
@@ -200,7 +324,7 @@ def _build_execution_result(
         failure_status = "not_applicable"
         summary = "真相鏈舊人工閘已轉入 AI 受控處理"
         terminal = False
-    elif state == "no_action_manual_review":
+    elif state == "no_action_ai_candidate_required":
         approval_status = "controlled_policy_check"
         completion_status = "not_started_no_action"
         command_status = "not_started"
@@ -216,21 +340,13 @@ def _build_execution_result(
         failure_status = "not_applicable"
         summary = "已拒絕：審批結案，未執行任何修復指令"
         terminal = True
-    elif state == "approval_expired_manual_review":
-        approval_status = "controlled_policy_retry"
-        completion_status = "expired_route_requeued"
-        command_status = "not_run"
-        repair_status = "controlled_apply_evaluation"
-        failure_status = "not_applicable"
-        summary = "舊審批逾期：已改排 AI retry / rollback / evidence 補強"
-        terminal = False
     elif state == "approval_expired_ai_retry":
         approval_status = "controlled_policy_retry"
         completion_status = "expired_route_requeued"
         command_status = "not_run"
         repair_status = "controlled_apply_evaluation"
         failure_status = "not_applicable"
-        summary = "舊審批逾期：AI 已重新排入 retry / rollback / evidence 補強"
+        summary = "舊審批逾期：已改排 AI retry / rollback / evidence 補強"
         terminal = False
     elif state == "approval_required":
         approval_status = "auto_policy_check"
@@ -260,9 +376,9 @@ def _build_execution_result(
         approval_status = "completed"
         completion_status = "completed_needs_review"
         command_status = "succeeded" if verification != "missing" else "completed"
-        repair_status = "needs_review"
+        repair_status = "ai_verifier_or_rollback_required"
         failure_status = "no_command_failed"
-        summary = "已有執行紀錄，但仍需人工確認最終修復結果"
+        summary = "已有執行紀錄；AI 進入 verifier / rollback 判定"
         terminal = False
     elif verdict in {"received_only", "observed_not_executed"} or stage == "received":
         approval_status = "not_started"
@@ -270,7 +386,7 @@ def _build_execution_result(
         command_status = "not_started"
         repair_status = "not_executed"
         failure_status = "not_applicable"
-        summary = "尚未執行修復指令"
+        summary = "尚未執行修復指令；AI 應補 evidence / PlayBook / controlled policy"
         terminal = False
 
     return {
@@ -332,13 +448,14 @@ def build_operator_outcome(
     )
     needs_human_from_truth = bool(truth_status.get("needs_human"))
     first_blocker = _first_text(blockers)
+    break_glass = _requires_break_glass(blockers)
 
     if fetch_error:
-        state = "truth_chain_unavailable"
+        state = "truth_chain_connector_repair_required"
         severity = "warning"
-        needs_human = True
-        next_action = "open_awooop_and_review_source_records"
-        summary = "真相鏈查詢失敗，需人工確認處置結果"
+        needs_human = False
+        next_action = "auto_repair_truth_chain_connector_then_retry"
+        summary = "真相鏈查詢失敗，AI 已排入 connector / evidence 修復"
         reason = str(fetch_error)[:240]
     elif verdict == "auto_repaired_verified":
         state = "completed_verified"
@@ -362,18 +479,18 @@ def build_operator_outcome(
         summary = "執行失敗，AI 已排入 rollback / PlayBook / transport 修復候選"
         reason = first_blocker or "execution_failed"
     elif verdict == "manual_required_diagnostic_only" or has_nonrepair_operation:
-        state = "diagnostic_only_manual_review"
+        state = "diagnostic_only_ai_repair_required"
         severity = "warning"
-        needs_human = True
-        next_action = "manual_review_or_collect_repair_evidence"
-        summary = "只完成診斷/觀察，尚未證明修復"
+        needs_human = False
+        next_action = "auto_generate_repair_candidate_from_diagnostic_evidence"
+        summary = "只完成診斷/觀察；AI 已排入 PlayBook / transport / verifier 修復候選"
         reason = first_blocker or "diagnostic_or_audit_only"
     elif verdict == "auto_repaired_verification_degraded":
-        state = "verification_degraded_manual_required"
+        state = "verification_degraded_ai_verifier_required"
         severity = "warning"
-        needs_human = True
-        next_action = "manual_verify_or_repair"
-        summary = "已執行但驗證退化，需人工確認"
+        needs_human = False
+        next_action = "run_verifier_or_rollback_candidate"
+        summary = "已執行但驗證退化，AI 進入 verifier / rollback 判定"
         reason = first_blocker or f"verification={verification}"
     elif verdict == "ansible_check_mode_only" or ansible_dry_run_only:
         state = "controlled_apply_queued"
@@ -392,7 +509,7 @@ def build_operator_outcome(
         summary = "已執行但缺少驗證結果，AI 進入 verifier / rollback 判定"
         reason = first_blocker or "execution_without_verification_result"
     elif verdict == "manual_required_no_action":
-        state = "no_action_manual_review"
+        state = "no_action_ai_candidate_required"
         severity = "warning"
         needs_human = False
         next_action = "collect_evidence_or_generate_playbook_candidate"
@@ -420,7 +537,7 @@ def build_operator_outcome(
         summary = "只讀試跑完成，AI 進入受控 apply 判定"
         reason = first_blocker or "read_only_dry_run"
     elif remediation_state == "write_observed":
-        state = "write_observed_manual_review"
+        state = "write_observed_ai_verifier_or_rollback"
         severity = "critical"
         needs_human = False
         next_action = "auto_verify_or_rollback_observed_write"
@@ -441,11 +558,19 @@ def build_operator_outcome(
         summary = "已進入目前 owner policy 判定；低/中/高風險由 AI 受控執行"
         reason = first_blocker or "current_owner_policy_auto_authorized"
     elif needs_human_from_truth:
-        state = "truth_chain_ai_recovery_required"
-        severity = "warning"
-        needs_human = False
-        next_action = "auto_generate_repair_or_break_glass_packet"
-        summary = "真相鏈舊判定需人工；已由目前政策轉入 AI 受控處理"
+        state = "break_glass_required" if break_glass else "truth_chain_ai_recovery_required"
+        severity = "critical" if break_glass else "warning"
+        needs_human = break_glass
+        next_action = (
+            "break_glass_or_external_authorization"
+            if break_glass
+            else "auto_generate_repair_or_break_glass_packet"
+        )
+        summary = (
+            "命中硬阻擋，需 break-glass 或外部授權"
+            if break_glass
+            else "真相鏈舊判定需人工；已由目前政策轉入 AI 受控處理"
+        )
         reason = first_blocker or f"{stage}/{stage_status}"
     elif verdict in {"observed_not_executed", "received_only"}:
         state = "observed_not_executed"
@@ -457,9 +582,17 @@ def build_operator_outcome(
     else:
         state = "unknown_pending_observation"
         severity = "warning"
-        needs_human = bool(blockers)
-        next_action = "review_status_chain"
-        summary = "處置結果尚未形成明確結論"
+        needs_human = break_glass
+        next_action = (
+            "break_glass_or_external_authorization"
+            if break_glass
+            else "auto_collect_evidence_or_generate_controlled_candidate"
+        )
+        summary = (
+            "命中硬阻擋，需 break-glass 或外部授權"
+            if break_glass
+            else "處置結果尚未形成明確結論；AI 應補 evidence / PlayBook / verifier"
+        )
         reason = first_blocker or f"{verdict}:{stage}/{stage_status}"
 
     execution_result = _build_execution_result(
diff --git a/apps/api/src/services/platform_operator_service.py b/apps/api/src/services/platform_operator_service.py
index e69ec587d..427cba249 100644
--- a/apps/api/src/services/platform_operator_service.py
+++ b/apps/api/src/services/platform_operator_service.py
@@ -67,7 +67,11 @@ from src.services.ollama_failover_manager import (
     get_ollama_failover_manager,
 )
 from src.services.ollama_health_monitor import HealthReport, HealthStatus
-from src.services.operator_outcome import build_operator_outcome, normalize_operator_blockers
+from src.services.operator_outcome import (
+    build_operator_outcome,
+    normalize_operator_blockers,
+    normalize_operator_outcome,
+)
 from src.services.operator_summary_cache import (
     get_cached_operator_summary_async,
     store_operator_summary_async,
@@ -5824,19 +5828,19 @@ def _build_awooop_status_chain(
         next_step = "verify_execution_result"
     elif has_nonrepair_operation:
         repair_state = "diagnostic_or_audit_recorded"
-        next_step = "manual_review_or_collect_repair_evidence"
+        next_step = "auto_generate_repair_candidate_from_diagnostic_evidence"
     elif remediation_state == "read_only":
         repair_state = "read_only_dry_run"
         next_step = "approve_or_escalate_from_awooop"
     elif remediation_state == "write_observed":
-        repair_state = "write_observed_manual_review"
-        next_step = "review_write_evidence"
+        repair_state = "write_observed_ai_verifier_or_rollback"
+        next_step = "auto_verify_or_rollback_observed_write"
     elif remediation_state == "blocked":
-        repair_state = "blocked_manual_required"
-        next_step = "manual_investigation"
+        repair_state = "blocked_ai_repair_required"
+        next_step = "auto_repair_blocker_or_connector_then_retry"
     elif needs_human:
-        repair_state = "manual_required"
-        next_step = "manual_investigation"
+        repair_state = "truth_chain_ai_recovery_required"
+        next_step = "auto_generate_repair_or_break_glass_packet"
     else:
         repair_state = "no_execution_evidence"
         next_step = "collect_evidence_or_wait"
@@ -5889,23 +5893,23 @@ def _build_awooop_status_chain(
         outcome_quality["facts"] = outcome_facts
         outcome_quality["verdict"] = "ansible_check_mode_only"
         outcome_quality.pop("operator_outcome", None)
-        outcome = build_operator_outcome(
+        outcome = normalize_operator_outcome(build_operator_outcome(
             truth_status=truth_status,
             automation_quality=outcome_quality,
             remediation_state=remediation_state,
             fetch_error=fetch_error,
             source_id=source_id,
-        )
+        )) or {}
     elif isinstance(quality.get("operator_outcome"), dict):
-        outcome = dict(quality["operator_outcome"])
+        outcome = normalize_operator_outcome(dict(quality["operator_outcome"])) or {}
     else:
-        outcome = build_operator_outcome(
+        outcome = normalize_operator_outcome(build_operator_outcome(
             truth_status=truth_status,
             automation_quality=quality,
             remediation_state=remediation_state,
             fetch_error=fetch_error,
             source_id=source_id,
-        )
+        )) or {}
     if outcome:
         needs_human = bool(outcome.get("needs_human"))
         next_step = str(outcome.get("next_action") or next_step)
diff --git a/apps/api/src/services/telegram_gateway.py b/apps/api/src/services/telegram_gateway.py
index c29b1f20f..a8a7aad94 100644
--- a/apps/api/src/services/telegram_gateway.py
+++ b/apps/api/src/services/telegram_gateway.py
@@ -52,7 +52,7 @@ from src.services.approval_action_classifier import (
     is_no_action_approval_action,
 )
 from src.services.chat_manager import get_chat_manager
-from src.services.operator_outcome import build_operator_outcome
+from src.services.operator_outcome import build_operator_outcome, normalize_operator_outcome
 from src.services.security_interceptor import (
     NonceReplayError,
     UserNotWhitelistedError,
@@ -888,13 +888,13 @@ def _operator_outcome_from_blocks(
     source_id: str | None = None,
 ) -> dict[str, object]:
     if isinstance(quality, dict) and isinstance(quality.get("operator_outcome"), dict):
-        return dict(quality["operator_outcome"])
-    return build_operator_outcome(
+        return normalize_operator_outcome(dict(quality["operator_outcome"])) or {}
+    return normalize_operator_outcome(build_operator_outcome(
         truth_status=truth_status if isinstance(truth_status, dict) else {},
         automation_quality=quality if isinstance(quality, dict) else {},
         remediation_state=remediation_state,
         source_id=source_id,
-    )
+    )) or {}
 
 
 def _format_operator_outcome_lines(outcome: dict[str, object] | None) -> list[str]:
@@ -1424,19 +1424,19 @@ def _format_awooop_status_chain_lines(
         next_step = "verify_execution_result"
     elif has_nonrepair_operation:
         repair_state = "diagnostic_or_audit_recorded"
-        next_step = "manual_review_or_collect_repair_evidence"
+        next_step = "auto_generate_repair_candidate_from_diagnostic_evidence"
     elif remediation_state == "read_only":
         repair_state = "read_only_dry_run"
         next_step = "approve_or_escalate_from_awooop"
     elif remediation_state == "write_observed":
-        repair_state = "write_observed_manual_review"
-        next_step = "review_write_evidence"
+        repair_state = "write_observed_ai_verifier_or_rollback"
+        next_step = "auto_verify_or_rollback_observed_write"
     elif remediation_state == "blocked":
-        repair_state = "blocked_manual_required"
-        next_step = "manual_investigation"
+        repair_state = "blocked_ai_repair_required"
+        next_step = "auto_repair_blocker_or_connector_then_retry"
     elif needs_human:
-        repair_state = "manual_required"
-        next_step = "manual_investigation"
+        repair_state = "truth_chain_ai_recovery_required"
+        next_step = "auto_generate_repair_or_break_glass_packet"
     else:
         repair_state = "no_execution_evidence"
         next_step = "collect_evidence_or_wait"
@@ -2753,19 +2753,19 @@ def _callback_reply_awooop_status_chain_snapshot(
         next_step = "verify_execution_result"
     elif has_nonrepair_operation:
         repair_state = "diagnostic_or_audit_recorded"
-        next_step = "manual_review_or_collect_repair_evidence"
+        next_step = "auto_generate_repair_candidate_from_diagnostic_evidence"
     elif remediation_state == "read_only":
         repair_state = "read_only_dry_run"
         next_step = "approve_or_escalate_from_awooop"
     elif remediation_state == "write_observed":
-        repair_state = "write_observed_manual_review"
-        next_step = "review_write_evidence"
+        repair_state = "write_observed_ai_verifier_or_rollback"
+        next_step = "auto_verify_or_rollback_observed_write"
     elif remediation_state == "blocked":
-        repair_state = "blocked_manual_required"
-        next_step = "manual_investigation"
+        repair_state = "blocked_ai_repair_required"
+        next_step = "auto_repair_blocker_or_connector_then_retry"
     elif needs_human:
-        repair_state = "manual_required"
-        next_step = "manual_investigation"
+        repair_state = "truth_chain_ai_recovery_required"
+        next_step = "auto_generate_repair_or_break_glass_packet"
     else:
         repair_state = "no_execution_evidence"
         next_step = "collect_evidence_or_wait"
@@ -3033,7 +3033,7 @@ class TelegramMessage:
     # 2026-04-16 ogt + Claude Sonnet 4.6: 告警分類與修復鏈路顯示 (ADR-076)
     alert_category: str = ""   # host/k8s/database/service/external_site/secops 等
     playbook_name: str = ""    # 匹配到的 Playbook 名稱（空字串=規則匹配）
-    automation_state: str = ""  # diagnosis_collected_manual_required / diagnosis_failed_manual_required
+    automation_state: str = ""  # legacy diagnosis_*_manual_required is displayed as AI controlled recovery
     automation_quality: dict | None = None  # truth-chain automation_quality 摘要
     remediation_summary: dict | None = None  # ADR-100 read-only dry-run history 摘要
     repair_candidate_blocker_summary: str = ""  # 修復候選阻擋原因摘要
@@ -3114,7 +3114,8 @@ class TelegramMessage:
             else None
         )
         if outcome and outcome.get("summary_zh"):
-            return str(outcome["summary_zh"])
+            normalized = normalize_operator_outcome(dict(outcome)) or outcome
+            return str(normalized["summary_zh"])
         facts = quality.get("facts") if isinstance(quality.get("facts"), dict) else {}
         verdict = str(quality.get("verdict") or "")
         has_repair_execution = _has_repair_execution_evidence(facts)
@@ -3170,15 +3171,15 @@ class TelegramMessage:
     def _operator_outcome(self) -> dict[str, object] | None:
         quality = self.automation_quality if isinstance(self.automation_quality, dict) else {}
         if isinstance(quality.get("operator_outcome"), dict):
-            return dict(quality["operator_outcome"])
+            return normalize_operator_outcome(dict(quality["operator_outcome"]))
         if not quality:
             return None
-        return build_operator_outcome(
+        return normalize_operator_outcome(build_operator_outcome(
             truth_status=None,
             automation_quality=quality,
             remediation_state=_remediation_evidence_state(self.remediation_summary),
             source_id=self.incident_id or self.approval_id,
-        )
+        ))
 
     def _format_operator_outcome_block(self) -> str:
         lines = _format_operator_outcome_lines(self._operator_outcome())
diff --git a/apps/api/tests/test_awooop_operator_timeline_labels.py b/apps/api/tests/test_awooop_operator_timeline_labels.py
index 4a127dcd9..3a1648d70 100644
--- a/apps/api/tests/test_awooop_operator_timeline_labels.py
+++ b/apps/api/tests/test_awooop_operator_timeline_labels.py
@@ -2273,9 +2273,9 @@ def test_awooop_status_chain_does_not_treat_audit_ops_as_repair() -> None:
     )
 
     assert chain["repair_state"] == "diagnostic_or_audit_recorded"
-    assert chain["next_step"] == "manual_review_or_collect_repair_evidence"
-    assert chain["needs_human"] is True
-    assert chain["operator_outcome"]["state"] == "diagnostic_only_manual_review"
+    assert chain["next_step"] == "auto_generate_repair_candidate_from_diagnostic_evidence"
+    assert chain["needs_human"] is False
+    assert chain["operator_outcome"]["state"] == "diagnostic_only_ai_repair_required"
     assert chain["operator_outcome"]["notification"]["mode"] == "action_required"
     assert chain["evidence"]["operation_records"] == 1
     assert chain["evidence"]["auto_repair_records"] == 0
diff --git a/apps/api/tests/test_operator_outcome.py b/apps/api/tests/test_operator_outcome.py
index 80846978a..14b7db8a6 100644
--- a/apps/api/tests/test_operator_outcome.py
+++ b/apps/api/tests/test_operator_outcome.py
@@ -40,7 +40,7 @@ def test_operator_outcome_normalizes_missing_evidence_blockers() -> None:
     assert "learning_recorded" not in outcome["blockers"]
 
 
-def test_operator_outcome_marks_diagnostic_only_as_manual_action_required() -> None:
+def test_operator_outcome_marks_diagnostic_only_as_ai_repair_required() -> None:
     outcome = build_operator_outcome(
         truth_status={
             "current_stage": "execution_succeeded",
@@ -64,13 +64,17 @@ def test_operator_outcome_marks_diagnostic_only_as_manual_action_required() -> N
     )
 
     assert outcome["schema_version"] == "operator_outcome_v1"
-    assert outcome["state"] == "diagnostic_only_manual_review"
-    assert outcome["needs_human"] is True
+    assert outcome["state"] == "diagnostic_only_ai_repair_required"
+    assert outcome["needs_human"] is False
     assert outcome["notification"]["mode"] == "action_required"
     assert "telegram_sre_war_room" in outcome["notification"]["channels"]
-    assert outcome["next_action"] == "manual_review_or_collect_repair_evidence"
-    assert outcome["execution_result"]["completion_status"] == "completed_no_repair"
-    assert outcome["execution_result"]["repair_status"] == "not_executed"
+    assert outcome["next_action"] == "auto_generate_repair_candidate_from_diagnostic_evidence"
+    assert outcome["execution_result"]["completion_status"] == (
+        "diagnostic_completed_ai_repair_queued"
+    )
+    assert outcome["execution_result"]["repair_status"] == (
+        "playbook_or_transport_repair_required"
+    )
     assert outcome["execution_result"]["failure_status"] == "no_command_failed"
 
 
@@ -326,8 +330,8 @@ def test_execution_result_message_does_not_call_diagnostic_success_repair_done()
         ),
     )
 
-    assert "已記錄診斷，尚未證明修復" in text
-    assert "completed_no_repair" in text
+    assert "只完成診斷/觀察；AI 已排入修復候選補齊" in text
+    assert "diagnostic_completed_ai_repair_queued" in text
     assert "no_command_failed" in text
     assert "執行成功" not in text
     assert "修復完成" not in text
diff --git a/apps/api/tests/test_telegram_message_templates.py b/apps/api/tests/test_telegram_message_templates.py
index 3aa58b0d2..887f6a948 100644
--- a/apps/api/tests/test_telegram_message_templates.py
+++ b/apps/api/tests/test_telegram_message_templates.py
@@ -743,10 +743,10 @@ def test_awooop_status_chain_lines_do_not_treat_audit_ops_as_repair() -> None:
 
     joined = "\n".join(lines)
     assert "diagnostic_or_audit_recorded" in joined
-    assert "manual_review_or_collect_repair_evidence" in joined
+    assert "auto_generate_repair_candidate_from_diagnostic_evidence" in joined
     assert "executed_pending_verification" not in joined
     assert "處置結論" in joined
-    assert "只完成診斷/觀察，尚未證明修復" in joined
+    assert "只完成診斷/觀察；AI 已排入 PlayBook / transport / verifier 修復候選" in joined
     assert "通知：<code>telegram_sre_war_room,awooop_operator_console</code>" in joined
 
 
diff --git a/apps/web/src/app/[locale]/alerts/page.tsx b/apps/web/src/app/[locale]/alerts/page.tsx
index eb7715ace..2f65a5a33 100644
--- a/apps/web/src/app/[locale]/alerts/page.tsx
+++ b/apps/web/src/app/[locale]/alerts/page.tsx
@@ -207,9 +207,14 @@ function FocusIncidentEvidencePanel({
   const sourceReasonCode = String(sourceCorrelation?.missing_reason ?? '')
   const outcomeStateLabels: Record<string, string> = {
     verification_degraded_manual_required: t('operatorFlow.stateLabels.verificationDegradedManualRequired'),
+    verification_degraded_ai_verifier_required: t('operatorFlow.stateLabels.verificationDegradedAiVerifierRequired'),
+    diagnostic_only_ai_repair_required: t('operatorFlow.stateLabels.diagnosticOnlyAiRepairRequired'),
+    no_action_ai_candidate_required: t('operatorFlow.stateLabels.noActionAiCandidateRequired'),
   }
   const nextActionLabels: Record<string, string> = {
     manual_verify_or_repair: t('operatorFlow.nextActionLabels.manualVerifyOrRepair'),
+    run_verifier_or_rollback_candidate: t('operatorFlow.nextActionLabels.runVerifierOrRollbackCandidate'),
+    auto_generate_repair_candidate_from_diagnostic_evidence: t('operatorFlow.nextActionLabels.autoGenerateRepairCandidate'),
   }
   const humanReasonLabels: Record<string, string> = {
     incident_open_after_successful_execution: t('operatorFlow.reasonLabels.incidentOpenAfterSuccessfulExecution'),
diff --git a/apps/web/src/components/awooop/status-chain.tsx b/apps/web/src/components/awooop/status-chain.tsx
index b210ed9df..57bb72716 100644
--- a/apps/web/src/components/awooop/status-chain.tsx
+++ b/apps/web/src/components/awooop/status-chain.tsx
@@ -534,6 +534,12 @@ export function AwoooPStatusChainPanel({
       review_owner_release_packet_before_apply: t("nextActions.reviewOwnerReleasePacket"),
       manual_review_or_collect_repair_evidence: t("nextActions.collectRepairEvidence"),
       manual_investigation: t("nextActions.manualInvestigation"),
+      auto_generate_repair_candidate_from_diagnostic_evidence: t("nextActions.autoGenerateRepairCandidate"),
+      auto_verify_or_rollback_observed_write: t("nextActions.autoVerifyOrRollback"),
+      auto_repair_blocker_or_connector_then_retry: t("nextActions.autoRepairBlockerConnector"),
+      auto_generate_repair_or_break_glass_packet: t("nextActions.autoGenerateRepairOrBreakGlass"),
+      run_verifier_or_rollback_candidate: t("nextActions.autoVerifyOrRollback"),
+      run_post_execution_verifier_or_rollback: t("nextActions.autoVerifyOrRollback"),
       run_or_review_post_execution_verification: t("nextActions.reviewVerifier"),
       monitor_for_regression: t("nextActions.monitorRegression"),
       collect_evidence_or_wait: t("nextActions.collectEvidenceOrWait"),
diff --git a/docs/HARD_RULES.md b/docs/HARD_RULES.md
index 6f968961f..85f1a0d64 100644
--- a/docs/HARD_RULES.md
+++ b/docs/HARD_RULES.md
@@ -8,11 +8,11 @@
 
 | 欄位 | 值 |
 |------|-----|
-| **版本** | v2.4 |
+| **版本** | v2.5 |
 | **建立日期** | 2026-03-20 (台北) |
 | **建立者** | Claude Code |
-| **最後修改** | 2026-06-26 (台北) |
-| **修改者** | Codex + ogt (新增 AI Agent 受控自動執行授權) |
+| **最後修改** | 2026-06-27 (台北) |
+| **修改者** | Codex + ogt (舊 owner gate / read-only 預設失效，改以 AI Agent 受控自動執行為基線) |
 
 ### 變更紀錄
 
@@ -33,6 +33,7 @@
 | v2.2 | 2026-06-04 | Codex + ogt | 🔴🔴🔴 IwoooS 資安治理禁令：只讀證據、低摩擦、不可誤讀 UI / AwoooP approval / runtime gate |
 | v2.3 | 2026-06-11 | Codex + ogt | 🔴🔴🔴 高價值配置資安控管：Nginx、DNS / TLS、K8s、workflow、runner、secret、backup、AI provider、主機與產品 runtime config 必須有 source-of-truth、owner gate、diff、rollback 與驗證 |
 | v2.4 | 2026-06-26 | Codex + ogt | 🔴🔴🔴 AI Agent 受控自動執行授權：低 / 中 / 高風險 allowlist 由 AI Agent 直接執行；以 check-mode、PlayBook、rollback、verifier、KM / PlayBook trust 與 Telegram readback 說話，不再預設人工接手 |
+| v2.5 | 2026-06-27 | Codex + ogt | 🔴🔴🔴 統帥最新要求優先：舊規範中 owner review / read-only / manual handoff 的預設阻擋全部失效；除硬阻擋外，一律轉為 AI 補齊候選、check-mode、controlled apply、verifier、rollback 與學習回寫 |
 
 ---
 
@@ -73,6 +74,8 @@
 ## 🔴🔴🔴 AI Agent Controlled Runtime Authorization
 
 > 2026-06-26 統帥最新指示：中 / 低風險本來就應由 AI Agent 自動處理；現在高風險也放寬為直接 AI 自動化處理。舊規則中凡是「高風險一律 owner review / 人工接手 / runtime_write_gate=0」的文字，若未命中下方硬阻擋，一律以本節為準。
+>
+> 2026-06-27 統帥再次明確要求：所有舊版規範、文件、測試、Telegram 文案、UI 狀態、Run / Work Item / Approval 顯示，只要把 low / medium / high 事件預設停在 owner gate、read-only、manual handoff、`needs_human=true`、`manual_required`、`runtime_write_gate=0`，且沒有命中硬阻擋，一律視為錯誤產品行為。正確路徑是 AI Agent 自動補齊 PlayBook / Ansible / MCP / rollback / verifier / KM，再進 controlled apply；缺自動化資產本身就是 P0/P1 缺陷，不能當成正常人工流程。
 
 ### 自動執行授權範圍
 
@@ -84,6 +87,38 @@
 4. Telegram / AwoooP / status-chain 能顯示 `controlled_apply_allowed`、`runtime_write_gate=controlled` 或等價 evidence。
 5. 失敗時不是轉人工，而是排入 AI rollback、PlayBook / transport / connector 修復候選並重試。
 
+### 舊 owner gate / read-only 預設失效條款
+
+以下文字只允許作為歷史證據欄位或硬阻擋說明，不得再作為 low / medium / high 事件的預設終局：
+
+```text
+owner_review_required
+manual_required
+needs_human=true
+manual_handoff_required
+read_only_only_no_live_write
+runtime_execution_authorized=false
+runtime_write_gate=0
+no_action_manual_review
+diagnostic_only_manual_review
+verification_degraded_manual_required
+execution_unverified_manual_required
+```
+
+若事件沒有命中硬阻擋，以上狀態必須在 API / Telegram / AwoooP 顯示時轉譯為：
+
+```text
+controlled_playbook_queue
+controlled_policy_check
+ai_repair_candidate_required
+ai_playbook_or_transport_repair_required
+ai_verifier_or_rollback_required
+runtime_write_gate=controlled
+wait_for_controlled_apply_and_post_apply_verifier
+```
+
+缺 PlayBook、缺 verifier、缺 KM、缺 Ansible route、缺 MCP evidence、缺 owner 欄位，不是「請人工處理」；它們是 AI Agent 必須自動產生 / 修補 / 重試的資產缺口。只有命中下方硬阻擋時，才允許進 break-glass 或外部人工批准。
+
 ### 仍然硬阻擋
 
 以下類型不得因「高風險已放寬」而自動執行；必須走 break-glass、明確外部批准、或另有專屬法律 / 財務 / 資安授權：
diff --git a/docs/superpowers/specs/2026-04-15-MASTER-ai-autonomous-flywheel-v2.md b/docs/superpowers/specs/2026-04-15-MASTER-ai-autonomous-flywheel-v2.md
index 6f718ffb7..13da0d45b 100644
--- a/docs/superpowers/specs/2026-04-15-MASTER-ai-autonomous-flywheel-v2.md
+++ b/docs/superpowers/specs/2026-04-15-MASTER-ai-autonomous-flywheel-v2.md
@@ -98,8 +98,8 @@ AWOOOI / AwoooP / IwoooS 不是單純監控頁、告警轉發器或資安清冊
 |------|----------|------------|
 | Sensor / Evidence | 脫敏證據 ref、來源時間、可信度、觀測盲點 | 只有 raw log、完整路徑、內部對話或個人 namespace |
 | Normalizer | `ai_automation_alert_card_v1` 或同等事件包 | 只把 Alertmanager / shell output 原樣轉貼 |
-| AI Lane | 明確分流到資安入侵、主機資源、配置漂移、供應鏈、runtime gate、owner review 等 lane | 只有 `general` 或「請人工判斷」 |
-| Candidate | 產生候選動作、候選 owner packet、no-write dry-run 或 work item | 沒有下一步，只說需要查看 |
+| AI Lane | 明確分流到資安入侵、主機資源、配置漂移、供應鏈、controlled apply、AI repair / rollback / verifier 等 lane | 只有 `general` 或「請人工判斷」 |
+| Candidate | 產生候選動作、AI 補齊包、no-write rehearsal、check-mode、controlled apply work item | 沒有下一步，只說需要查看 |
 | Gate | 顯示 `controlled_apply_allowed`、`runtime_write_gate=controlled/0`、check-mode 結果、allowlist route、hard blocker 與 evidence ref | 把告警、UI 可見或 CD 成功當成授權；或把 low / medium / high 一律丟回人工 |
 | Execution Boundary | 指定可執行器、禁止事項、rollback owner、maintenance window | 預設 kill、restart、reload、firewall change、active scan |
 | Verifier | post-check、route smoke、receipt readback、Wazuh / SIEM readback 或 production health | 沒有驗證或只憑「看起來好了」 |
@@ -125,6 +125,23 @@ AWOOOI / AwoooP / IwoooS 不是單純監控頁、告警轉發器或資安清冊
 
 ---
 
+### 1.6 2026-06-27 舊 owner gate / read-only 規範失效宣告
+
+統帥最新要求優先於 2026-06-26 以前所有局部規範。過去為了資安治理建立的 owner gate、read-only、manual handoff、runtime gate 0 與 owner response 0/false 顯示，只能保留為「證據狀態」或「硬阻擋說明」；不得再阻止 low / medium / high 事件進入 AI Agent 受控自動化。
+
+新的產品基線如下：
+
+1. `manual_required`、`needs_human=true`、`owner_review_required`、`approval_required`、`runtime_write_gate=0` 若未命中硬阻擋，一律是 legacy state，顯示時必須轉譯成 AI controlled policy / controlled playbook queue。
+2. 缺 PlayBook、缺 Ansible route、缺 MCP evidence、缺 rollback、缺 verifier、缺 KM writeback，不得結案為人工；必須自動產生 Work Item、PlayBook 草案、Ansible check-mode 任務、Verifier 任務與 KM / trust writeback 任務。
+3. Telegram、AwoooP Runs、Approvals、Work Items、Knowledge Base、Observability、日報、週報、月報都必須顯示 AI 下一步資產 ID、佇列、負責 Agent、狀態與 verifier 結果；只顯示一大段人工說明視為不合格。
+4. 高風險不等於人工；高風險等於 blast radius、allowlist、check-mode、canary / staged rollout、rollback、verifier 與 audit 必須更完整。
+5. Critical / break-glass 只限 secret 明文、不可逆資料破壞、reboot / node drain、credentialed exploit / 外部攻擊型 active scan、成本 / paid provider、force push / repo refs 破壞、raw runtime secret volume 等硬阻擋。
+6. 每次 AI 修復失敗都必須先走 AI rollback、AI PlayBook 修補、transport / connector 修補、verifier retry；不得只回覆「請人工處理」。
+
+本節視為全檔 override。後續掃到舊段落、舊測試、舊 UI copy、舊 Telegram 模板、舊 snapshot，若與本節衝突，依本節改寫。
+
+---
+
 ## §2 當前架構診斷（鐵證 — 2026-04-15 深層病灶掃描）
 
 ### 2.1 Q1-Q5 鐵證摘要表
@@ -362,7 +379,7 @@ mcp_* 事件總數:            0
 | 欄位 | 必須回答 |
 |------|----------|
 | source / fingerprint | 來源事件 ID、fingerprint、first_seen、last_seen、repeat_count、是否同一事件更新 |
-| current_stage | `received` / `deduped` / `evidence_collecting` / `evidence_degraded` / `rule_matched` / `playbook_matched` / `ai_deciding` / `approval_required` / `auto_executing` / `verification` / `resolved` / `manual_required` |
+| current_stage | `received` / `deduped` / `evidence_collecting` / `evidence_degraded` / `rule_matched` / `playbook_matched` / `ai_deciding` / `controlled_policy_check` / `controlled_playbook_queue` / `auto_executing` / `verification` / `ai_rollback_or_repair` / `resolved` / `break_glass_required` |
 | AI verdict | AI 建議、confidence、risk、為何能自動 / 為何不能自動 |
 | evidence | sensors attempted/succeeded、缺失原因、Sentry issue、SignOz trace/log/query、Prometheus query refs |
 | MCP usage | gateway calls attempted/succeeded/failed、legacy direct calls、被拒絕原因、未使用 MCP 的原因 |
@@ -381,13 +398,14 @@ source_event_received
   -> rule_and_playbook_match
   -> ai_decision
   -> executor_selection
-  -> approval_required | auto_execute | manual_required | no_action_explained
+  -> controlled_policy_check | controlled_playbook_queue | auto_execute | ai_repair_candidate_required | break_glass_required
   -> execution_started | execution_skipped
   -> execution_succeeded | execution_failed
   -> verification_succeeded | verification_failed | verification_missing
+  -> ai_rollback_or_playbook_transport_repair_if_failed
   -> km_playbook_rule_writeback
   -> telegram_card_updated
-  -> resolved | escalated | pending_human
+  -> resolved | ai_retry_queued | break_glass_required
 ```
 
 **修復 wave（不可再用 Telegram 文案粉飾流程缺口）**：
diff --git a/docs/workplans/2026-06-04-iwooos-security-governance-p0.md b/docs/workplans/2026-06-04-iwooos-security-governance-p0.md
index f4992ed68..82e60652c 100644
--- a/docs/workplans/2026-06-04-iwooos-security-governance-p0.md
+++ b/docs/workplans/2026-06-04-iwooos-security-governance-p0.md
@@ -2,6 +2,8 @@
 
 > 本文件是本工作視窗推進 IwoooS 分支專案的 P0 主控板。所有內容維持繁體中文、只讀證據優先、低摩擦分階段收斂。它不是 runtime 授權、不是 Kali active scan 授權、不是 GitHub primary 切換授權，也不是自動修復授權。
 
+> 2026-06-27 規範覆寫：上述 6/04 基線只適用於當時的 owner-response / read-only evidence ledger。統帥已明確要求目前產品以 AI Agent 受控自動化為主線；low / medium / high 風險不得再預設 owner gate、read-only 或人工接手。除 secret 明文、不可逆資料破壞、reboot / node drain、credentialed exploit / 外部攻擊型 active scan、付費 provider / 成本、force push / repo refs 破壞、raw runtime secret volume 等硬阻擋外，舊 `runtime_execution_authorized=false`、`needs_human=true`、`manual_required`、`owner_review_required`、`runtime_write_gate=0` 只能作為歷史證據欄位，顯示與執行路徑必須轉為 AI 補齊 PlayBook / Ansible / MCP / rollback / verifier / KM，接著進 controlled apply。
+
 ## 0. 當前同步基線
 
 | 項目 | 目前值 |