feat(auto-repair): route ssh diagnostics through mcp gateway

apps/api/tests/test_auto_repair_service.py

@@ -19,6 +19,7 @@ from src.models.playbook import (
     RiskLevel,
     SymptomPattern,
 )
+from src.plugins.mcp.interfaces import MCPToolResult
 from src.services.auto_repair_service import AutoRepairService
 from src.utils.timezone import now_taipei
 
@@ -120,6 +121,14 @@ async def _no_cooldown(*args, **kwargs) -> tuple[bool, str]:
     return True, "允許自動修復 (test bypass)"
 
 
+class _DbContext:
+    async def __aenter__(self) -> object:
+        return object()
+
+    async def __aexit__(self, *_args: object) -> None:
+        return None
+
+
 class TestAutoRepairService:
     """Auto Repair Service unit tests"""
 
@@ -415,6 +424,90 @@ class TestAutoRepairService:
 
         assert service._should_escalate_failed_verification(incident, playbook) is False
 
+    def test_legacy_ssh_diagnostic_routes_to_mcp_gateway(self, service):
+        """Legacy YAML_RULE SSH diagnostics become governed read-only MCP calls."""
+        incident = create_test_incident(
+            severity=Severity.P2,
+            alert_category="infrastructure",
+            alert_name="DockerContainerMemoryLimitPressure",
+        )
+        incident.signals[0].labels.update({
+            "host": "110",
+            "container_name": "sentry-self-hosted-clickhouse-1",
+        })
+
+        route = service._route_legacy_ssh_command_to_mcp(
+            incident,
+            'ssh {host} \'echo "=== LOAD ==="; uptime; docker stats --no-stream\'',
+        )
+
+        assert route is not None
+        assert route.tool_name == "ssh_diagnose"
+        assert route.params == {
+            "host": "192.168.0.110",
+            "container_name": "sentry-self-hosted-clickhouse-1",
+        }
+
+    def test_legacy_ssh_write_action_does_not_route_to_read_mcp(self, service):
+        """Write operations must stay out of the read-only diagnostic grant."""
+        incident = create_test_incident(severity=Severity.P2)
+
+        route = service._route_legacy_ssh_command_to_mcp(
+            incident,
+            "ssh {host} 'docker restart minio'",
+        )
+
+        assert route is None
+
+    @pytest.mark.asyncio
+    async def test_execute_legacy_ssh_diagnostic_uses_mcp_gateway(
+        self,
+        service,
+        monkeypatch,
+    ):
+        incident = create_test_incident(
+            severity=Severity.P2,
+            alert_category="infrastructure",
+            alert_name="DockerContainerMemoryLimitPressure",
+        )
+        incident.signals[0].labels.update({
+            "host": "110",
+            "container_name": "momo-scheduler",
+        })
+        step = RepairStep(
+            step_number=1,
+            action_type=ActionType.SSH_COMMAND,
+            command='ssh {host} \'echo "=== LOAD ==="; uptime; docker stats --no-stream\'',
+            risk_level=RiskLevel.LOW,
+        )
+        calls = []
+
+        class FakeGateway:
+            def __init__(self, db: object) -> None:
+                self.db = db
+
+            async def call(self, ctx, params):
+                calls.append({"ctx": ctx, "params": params, "db": self.db})
+                return MCPToolResult(
+                    success=True,
+                    execution_id="gw-ok",
+                    output={"stdout": "=== CPU TOP ==="},
+                )
+
+        monkeypatch.setattr("src.db.base.get_db_context", lambda _project_id: _DbContext())
+        monkeypatch.setattr("src.plugins.mcp.gateway.McpGateway", FakeGateway)
+
+        result = await service._execute_step(incident, step)
+
+        assert result.startswith("SUCCESS: mcp:ssh_diagnose")
+        assert calls
+        assert calls[0]["ctx"].agent_id == "auto_repair_executor"
+        assert calls[0]["ctx"].required_scope == "read"
+        assert calls[0]["ctx"].is_shadow is False
+        assert calls[0]["params"]["host"] == "192.168.0.110"
+        assert calls[0]["params"]["container_name"] == "momo-scheduler"
+        assert calls[0]["params"]["_mcp_audit"]["flywheel_node"] == "execute"
+
     @pytest.mark.asyncio
     async def test_evaluate_low_risk_allowed(self, service, mock_playbook_service):
         """Test that LOW risk actions are allowed"""

apps/api/tests/test_post_execution_verifier.py

@@ -384,6 +384,25 @@ class _FakeRegistry:
         ]
 
 
+class _PrometheusRegistry:
+    def __init__(self, provider: _CaptureProvider) -> None:
+        self.provider = provider
+
+    def suggest_tools(self, alertname: str = "", incident_labels: dict | None = None) -> list[RegisteredTool]:
+        return [
+            RegisteredTool(
+                tool=MCPTool(
+                    name="prometheus_query",
+                    description="",
+                    input_schema={},
+                    server_name="capture",
+                ),
+                provider=self.provider,
+                dimensions=[SensorDimension.D3_METRICS],
+            )
+        ]
+
+
 class _DbContext:
     async def __aenter__(self) -> object:
         return object()
@@ -448,3 +467,39 @@ class TestCollectPostStateAuditContext:
         assert ctx.required_scope == "read"
         assert calls[0]["parameters"]["_mcp_audit"]["incident_id"] == "INC-TEST"
         assert calls[0]["parameters"]["_mcp_audit"]["flywheel_node"] == "verify"
+
+    @pytest.mark.asyncio
+    async def test_collect_post_state_sends_prometheus_query_parameter(self):
+        provider = _CaptureProvider()
+        verifier = PostExecutionVerifier()
+        verifier._registry = _PrometheusRegistry(provider)
+        incident = _stub_incident(alertname="DockerContainerMemoryLimitPressure")
+        incident.signals[0].labels.update({
+            "host": "110",
+            "container_name": "momo-scheduler",
+        })
+
+        await verifier._collect_post_state(incident)
+
+        assert provider.seen_parameters is not None
+        query = provider.seen_parameters["query"]
+        assert "docker_container_memory_usage_bytes" in query
+        assert 'host="110"' in query
+        assert 'container_name="momo-scheduler"' in query
+
+
+class TestPrometheusQueryBuilder:
+    def test_docker_memory_alert_query_is_not_empty(self):
+        query = pev_module._build_prometheus_query(
+            "DockerContainerMemoryLimitPressure",
+            {"host": "110", "container_name": "momo-scheduler"},
+        )
+
+        assert "docker_container_memory_usage_bytes" in query
+        assert "docker_container_memory_limit_bytes" in query
+
+    def test_canary_alert_uses_generic_non_empty_query(self):
+        query = pev_module._build_prometheus_query("AwoooPAutoRepairCanaryT16", {})
+
+        assert query
+        assert "up" in query

apps/api/tests/test_ssh_provider_tools.py

@@ -21,6 +21,18 @@ async def test_ssh_diagnose_is_registered_read_only_tool():
     assert "df -h" in command
 
 
+def test_ssh_diagnose_can_include_container_read_only_context():
+    provider = SSHProvider()
+
+    command = provider._build_command(
+        "ssh_diagnose",
+        {"container_name": "sentry-self-hosted-clickhouse-1"},
+    )
+
+    assert "docker stats --no-stream sentry-self-hosted-clickhouse-1" in command
+    assert "docker inspect sentry-self-hosted-clickhouse-1" in command
+
+
 def test_ssh_provider_uses_ollama_user_for_188():
     provider = SSHProvider()
 
@@ -32,6 +44,8 @@ def test_ssh_provider_uses_ollama_user_for_188():
     "raw,expected",
     [
         ("192.168.0.110:9100", "192.168.0.110"),
+        ("110:9100", "192.168.0.110"),
+        ("188", "192.168.0.188"),
         ("wooo@192.168.0.110", "192.168.0.110"),
         ("ssh://wooo@192.168.0.110:22", "192.168.0.110"),
         ("192.168.0.188", "192.168.0.188"),