Уточнить валидацию guarded MCP ответов агента

scripts/agent_detector_runner.py

@@ -518,12 +518,24 @@ def evaluate_limited_next_action(
     limited_outputs = 0
     for output in outputs:
         text = current_answer_text_for_output(output)
+        step_state_path = output.get("step_state_path")
+        step_state = read_json_object(step_state_path) if isinstance(step_state_path, Path) else {}
+        review = step_state.get("business_first_review") if isinstance(step_state.get("business_first_review"), dict) else {}
+        review_limited = review.get("limited_answer_detected")
+        if review_limited is False:
+            continue
         limited_match = first_pattern_search(limited_patterns, text)
-        if not limited_match:
+        if review_limited is not True and not limited_match:
             continue
         limited_outputs += 1
-        if not first_pattern_search(next_action_patterns, text):
-            failures.append({"path": output["repo_path"], "limited_match": limited_match.group(0)[:240]})
+        next_action_present = review.get("next_action_present") is True or bool(first_pattern_search(next_action_patterns, text))
+        if not next_action_present:
+            evidence = {"path": output["repo_path"]}
+            if limited_match:
+                evidence["limited_match"] = limited_match.group(0)[:240]
+            else:
+                evidence["limited_match"] = "business_first_review.limited_answer_detected"
+            failures.append(evidence)
     status = "fail" if failures else "pass"
     if failures:
         message = "limited answer has no next action"

scripts/domain_case_loop.py

@@ -260,6 +260,10 @@ GUARDED_INSUFFICIENCY_PRIMARY_MARKERS = (
     "\u043d\u0435 \u043f\u043e\u0434\u0442\u0432\u0435\u0440\u0436\u0434\u0430",
     "\u043d\u0435\u043b\u044c\u0437\u044f \u0447\u0435\u0441\u0442\u043d\u043e \u043f\u043e\u0434\u0442\u0432\u0435\u0440\u0434",
     "\u043d\u0435\u043b\u044c\u0437\u044f \u043a\u043e\u0440\u0440\u0435\u043a\u0442\u043d\u043e \u043e\u043f\u0440\u0435\u0434\u0435\u043b",
+    "\u043d\u0435\u043b\u044c\u0437\u044f \u0434\u043e\u043a\u0430\u0437\u0430\u0442\u044c",
+    "\u043d\u0435 \u0434\u043e\u043a\u0430\u0437\u0430\u0442\u044c",
+    "\u043d\u0435 \u0443\u0441\u0442\u0430\u043d\u043e\u0432\u043b\u0435\u043d \u0441\u0440\u043e\u043a \u043e\u043f\u043b\u0430\u0442\u044b",
+    "\u043d\u0435\u0442 \u0443\u0441\u0442\u0430\u043d\u043e\u0432\u043b\u0435\u043d\u043d\u043e\u0433\u043e \u0441\u0440\u043e\u043a\u0430 \u043e\u043f\u043b\u0430\u0442\u044b",
     "\u0447\u0435\u0441\u0442\u043d\u043e \u043f\u043e\u0441\u0447\u0438\u0442\u0430\u0442\u044c \u043d\u0435\u043b\u044c\u0437\u044f",
     "\u043d\u0435\u0442 \u0434\u043e\u0441\u0442\u0430\u0442\u043e\u0447\u043d\u043e\u0439 \u0431\u0430\u0437\u044b",
     "\u043f\u043e\u0434\u0442\u0432\u0435\u0440\u0436\u0434\u0435\u043d\u043d\u043e\u0439 \u0441\u0435\u0431\u0435\u0441\u0442\u043e\u0438\u043c",
@@ -275,6 +279,9 @@ GUARDED_INSUFFICIENCY_LIMITATION_MARKERS = (
     "\u043d\u0435 \u0434\u043e\u043a\u0430\u0437\u044b\u0432\u0430\u0435\u0442 \u043e\u0441\u0442\u0430\u0442\u043e\u043a",
     "\u043d\u0435 \u0444\u0438\u043d\u0430\u043b\u044c\u043d\u044b\u0439 \u0440\u0435\u0435\u0441\u0442\u0440",
     "\u0433\u0440\u0430\u043d\u0438\u0446\u0430 \u043e\u0442\u0432\u0435\u0442\u0430",
+    "due date",
+    "\u0441\u0440\u043e\u043a \u043e\u043f\u043b\u0430\u0442\u044b",
+    "\u0434\u043e\u0433\u043e\u0432\u043e\u0440\u043d\u043e\u0439 \u043f\u0440\u043e\u0441\u0440\u043e\u0447",
     "\u0440\u0430\u0441\u0447\u0435\u0442\u043d\u0443\u044e \u0431\u0430\u0437\u0443",
     "\u0440\u0430\u0441\u0447\u0451\u0442\u043d\u0443\u044e \u0431\u0430\u0437\u0443",
 )
@@ -2798,6 +2805,26 @@ def has_guarded_insufficiency_language(answer_text: str) -> bool:
     return has_unconfirmed_marker and has_limitation_marker
 
 
+def is_validated_guarded_mcp_discovery_insufficiency(state: dict[str, Any]) -> bool:
+    if state.get("mcp_discovery_response_applied") is not True:
+        return False
+    if str(state.get("mcp_discovery_response_candidate_status") or "").strip() != "ready_for_guarded_use":
+        return False
+    if str(state.get("mcp_discovery_route_candidate_status") or "").strip() != "ready_for_reviewed_execution":
+        return False
+    if str(state.get("mcp_discovery_execution_handoff_can_use_guarded_response") or "").casefold() not in {"true", "1"}:
+        return False
+    if str(state.get("truth_gate_contract_status") or "").strip() != "partial_supported":
+        return False
+    if str(state.get("truth_mode") or "").strip() not in GUARDED_INSUFFICIENCY_TRUTH_MODES:
+        return False
+    if str(state.get("answer_shape") or "").strip() not in GUARDED_INSUFFICIENCY_ANSWER_SHAPES:
+        return False
+    if str(state.get("evidence_grade") or "").strip() not in {"weak", "medium", "strong"}:
+        return False
+    return bool(str(state.get("assistant_text") or "").strip())
+
+
 def is_validated_guarded_insufficiency_answer(
     state: dict[str, Any],
     execution_status: str,
@@ -2808,13 +2835,14 @@ def is_validated_guarded_insufficiency_answer(
         return False
     if violations:
         return False
-    if state.get("mcp_discovery_response_applied") is True:
+    mcp_discovery_guarded = is_validated_guarded_mcp_discovery_insufficiency(state)
+    if state.get("mcp_discovery_response_applied") is True and not mcp_discovery_guarded:
         return False
     if str(state.get("reply_type") or "").strip() not in {"factual", "factual_with_explanation", "partial_coverage"}:
         return False
-    if str(state.get("fallback_type") or "").strip() not in {"", "none"}:
+    if not mcp_discovery_guarded and str(state.get("fallback_type") or "").strip() not in {"", "none"}:
         return False
-    if str(state.get("mcp_call_status") or "").strip() != "matched_non_empty":
+    if not mcp_discovery_guarded and str(state.get("mcp_call_status") or "").strip() != "matched_non_empty":
         return False
     result_mode = str(state.get("result_mode") or "").strip()
     truth_mode = str(state.get("truth_mode") or "").strip()
@@ -2825,7 +2853,7 @@ def is_validated_guarded_insufficiency_answer(
         or answer_shape in GUARDED_INSUFFICIENCY_ANSWER_SHAPES
         or state.get("balance_confirmed") is False
     )
-    if not has_limited_evidence_mode:
+    if not (has_limited_evidence_mode or mcp_discovery_guarded):
         return False
     if not has_guarded_insufficiency_language(str(state.get("assistant_text") or "")):
         return False

scripts/test_agent_detector_runner.py

@@ -160,6 +160,65 @@ class AgentDetectorRunnerTests(unittest.TestCase):
         self.assertEqual(results["summary"]["status"], "pass")
         self.assertEqual(results["results"][0]["message"], "limited answer includes a next action")
 
+    def test_limited_next_action_ignores_non_limited_off_domain_can_not_phrase(self) -> None:
+        with tempfile.TemporaryDirectory() as tmp:
+            root = Path(tmp)
+            artifact_dir = root / "run"
+            step_dir = artifact_dir / "scenarios" / "chat" / "steps" / "s01"
+            write_text(step_dir / "output.md", "Нет, пингвины не могут летать.")
+            write_json(
+                step_dir / "step_state.json",
+                {
+                    "business_first_review": {
+                        "limited_answer_detected": False,
+                        "next_action_present": False,
+                    }
+                },
+            )
+            registry_path = root / "detector_registry.json"
+            issue_catalog_path = root / "issue_catalog.json"
+            write_json(
+                registry_path,
+                {
+                    "schema_version": "agent_detector_registry_v1",
+                    "detectors": {
+                        "limited_answer_without_next_action": {
+                            "kind": "answer_text_required_when_limited",
+                            "automation_level": "semi_automatic",
+                            "description": "Limited answers need a next action.",
+                            "issue_codes": ["business_next_step_missing"],
+                            "inputs": ["output.md"],
+                            "check": {
+                                "limited_patterns": ["(?i)не могу"],
+                                "required_next_action_patterns_any": ["(?i)что проверить дальше"],
+                            },
+                        }
+                    },
+                },
+            )
+            write_json(
+                issue_catalog_path,
+                {
+                    "schema_version": "agent_issue_catalog_v1",
+                    "issues": {
+                        "business_next_step_missing": {
+                            "detectors": ["limited_answer_without_next_action"],
+                        }
+                    },
+                },
+            )
+
+            results = runner.build_detector_results(
+                artifact_dir,
+                issue_codes=["business_next_step_missing"],
+                registry_path=registry_path,
+                issue_catalog_path=issue_catalog_path,
+                include_default_global=False,
+            )
+
+        self.assertEqual(results["summary"]["status"], "pass")
+        self.assertEqual(results["results"][0]["message"], "answer is not limited; next action requirement not triggered")
+
     def test_candidate_forbidden_regex_is_limited_to_evidence_path(self) -> None:
         with tempfile.TemporaryDirectory() as tmp:
             root = Path(tmp)

scripts/test_domain_case_loop_step_state.py

@@ -1617,6 +1617,46 @@ class DomainCaseLoopStepStateTests(unittest.TestCase):
         self.assertTrue(step_state["guarded_insufficiency_validated"])
         self.assertEqual(step_state["acceptance_status"], "validated")
 
+    def test_guarded_mcp_discovery_debt_boundary_followup_validates(self) -> None:
+        answer_text = (
+            "Да, доказать просрочку по компании ООО Альтернатива Плюс на 2020-12-31 нельзя, "
+            "потому что в подтвержденных данных не установлен срок оплаты.\n"
+            "Открытый долг на 35 472 380,36 руб. виден, но без due date это задолженность "
+            "без доказанной договорной просрочки."
+        )
+        step_state = dcl.validate_step_contract(
+            {
+                "execution_status": "partial",
+                "reply_type": "partial_coverage",
+                "required_answer_shape": "direct_answer_first",
+                "required_answer_patterns_all": [
+                    "(?i)срок.*оплат|due[- ]?date|договор",
+                    "(?i)не установлен|не хватает|не подтвержд|не доказ",
+                    "(?i)просроч",
+                    "(?i)2020|2020-12-31|конец 2020",
+                ],
+                "fallback_type": "partial",
+                "truth_gate_contract_status": "partial_supported",
+                "truth_mode": "limited",
+                "answer_shape": "limited_with_reason",
+                "evidence_grade": "medium",
+                "mcp_discovery_response_applied": True,
+                "mcp_discovery_response_candidate_status": "ready_for_guarded_use",
+                "mcp_discovery_route_candidate_status": "ready_for_reviewed_execution",
+                "mcp_discovery_execution_handoff_can_use_guarded_response": True,
+                "assistant_text": answer_text,
+                "actual_direct_answer": (
+                    "Да, доказать просрочку по компании ООО Альтернатива Плюс на 2020-12-31 нельзя, "
+                    "потому что в подтвержденных данных не установлен срок оплаты."
+                ),
+                "top_non_empty_lines": answer_text.splitlines(),
+            }
+        )
+
+        self.assertNotIn("required_answer_patterns_all_missing", step_state["violated_invariants"])
+        self.assertTrue(step_state["guarded_insufficiency_validated"])
+        self.assertEqual(step_state["acceptance_status"], "validated")
+
     def test_expected_clarification_partial_answer_validates(self) -> None:
         answer_text = (
             "Для рейтинга прибыльности номенклатуры нужен период.\n\n"