fix(auto-repair): prefer exact playbooks and fail failed steps

apps/api/tests/test_auto_repair_service.py

@@ -277,6 +277,49 @@ class TestAutoRepairService:
         assert decision.playbook.playbook_id == playbook.playbook_id
         assert decision.blocked_by is None
 
+    @pytest.mark.asyncio
+    async def test_exact_alert_match_wins_over_higher_fuzzy_similarity(
+        self,
+        service,
+        mock_playbook_service,
+    ):
+        """Exact alert/service playbooks must outrank unrelated fuzzy matches."""
+        fuzzy_playbook = create_high_quality_playbook(
+            playbook_id="PB-FUZZY-HOST",
+            risk_level=RiskLevel.LOW,
+        )
+        fuzzy_playbook.symptom_pattern = SymptomPattern(
+            alert_names=["HostCPUHigh"],
+            affected_services=["node-exporter"],
+            severity_range=["P2"],
+        )
+        exact_playbook = create_high_quality_playbook(
+            playbook_id="PB-EXACT-CANARY",
+            risk_level=RiskLevel.LOW,
+        )
+        exact_playbook.symptom_pattern = SymptomPattern(
+            alert_names=["AwoooPAutoRepairCanaryT16"],
+            affected_services=["awoooi-auto-repair-canary"],
+            severity_range=["P3"],
+        )
+        mock_playbook_service.add_playbook(fuzzy_playbook)
+        mock_playbook_service.add_playbook(exact_playbook)
+        mock_playbook_service.set_recommendations([
+            MockPlaybookRecommendation(fuzzy_playbook, similarity_score=0.95),
+            MockPlaybookRecommendation(exact_playbook, similarity_score=0.45),
+        ])
+
+        incident = create_test_incident(
+            severity=Severity.P3,
+            alert_name="AwoooPAutoRepairCanaryT16",
+        )
+        incident.affected_services = ["awoooi-auto-repair-canary"]
+        decision = await service.evaluate_auto_repair(incident)
+
+        assert decision.can_auto_repair is True
+        assert decision.playbook is not None
+        assert decision.playbook.playbook_id == "PB-EXACT-CANARY"
+
     @pytest.mark.asyncio
     async def test_backup_failure_blocks_k8s_playbook(self, service, mock_playbook_service):
         """Backup/host incidents must not execute K8s rollout playbooks."""

apps/api/tests/test_learning_chain_e2e.py

@@ -298,9 +298,6 @@ async def test_auto_repair_failure_does_not_call_verifier(monkeypatch):
     pb_service = FailingPlaybookService()
     pb_service.add_playbook(playbook)
 
-    # 讓 _execute_step 拋例外以觸發失敗路徑
-    original_execute_step = AutoRepairService._execute_step
-
     async def _always_fail(self_inner, incident_arg, step_arg) -> str:
         raise RuntimeError("強制測試失敗")
 
@@ -323,6 +320,40 @@ async def test_auto_repair_failure_does_not_call_verifier(monkeypatch):
     assert len(stub_learning.verification_calls) == 0, "執行失敗時不應呼叫 record_verification_result"
 
 
+@pytest.mark.asyncio
+async def test_auto_repair_failed_step_string_marks_execution_failure(monkeypatch):
+    """Executor returned FAILED text must not be stored as successful repair."""
+    stub_verifier = StubVerifier(result="success")
+    stub_learning = StubLearningService()
+
+    import src.services.post_execution_verifier as _pev_mod
+    monkeypatch.setattr(_pev_mod, "_verifier", stub_verifier)
+
+    import src.services.learning_service as _ls_mod
+    monkeypatch.setattr(_ls_mod, "_learning_service", stub_learning)
+
+    playbook = _make_playbook()
+    pb_service = StubPlaybookService()
+    pb_service.add_playbook(playbook)
+
+    async def _returns_failed(self_inner, incident_arg, step_arg) -> str:
+        return "FAILED: simulated executor failure"
+
+    monkeypatch.setattr(AutoRepairService, "_execute_step", _returns_failed)
+
+    service = AutoRepairService(
+        playbook_service=pb_service,
+        cooldown_checker=_no_cooldown,
+    )
+
+    result = await service.execute_auto_repair(_make_incident(), playbook)
+
+    assert result.success is False
+    assert "simulated executor failure" in (result.error or "")
+    assert len(stub_verifier.calls) == 0
+    assert len(stub_learning.verification_calls) == 0
+
+
 @pytest.mark.asyncio
 async def test_record_verification_result_no_playbook_id_does_not_crash():
     """
@@ -330,7 +361,6 @@ async def test_record_verification_result_no_playbook_id_does_not_crash():
     驗證 learning_service 對 None playbook_id 的防禦性。
     """
     from src.services.learning_service import LearningService
-    from src.repositories.interfaces import ILearningRepository, ITrustRepository
 
     class NullLearningRepo:
         async def record_repair(self, **kwargs) -> bool: