优化语音输入/普通文本输入

src/cognitive_agent/application/seat_service.py

@@ -133,7 +133,9 @@ CONVERSATION_ID_PATTERN = re.compile(r"^conv-[a-z0-9-]{6,80}$")
 MODE_EDITOR_REQUEST_ID_PATTERN = re.compile(r"^req-[a-z0-9-]{6,100}$")
 SEAT_SCOPE_PATTERN = re.compile(r"\bseats?\b|座椅|坐椅", re.IGNORECASE)
 DRIVER_SEAT_SCOPE_PATTERN = re.compile(
-    r"\bdriver\b|\bfront[-_ ]?left\b|\bleft[-_ ]?front\b|主驾|驾驶位|司机位|左前",
+    r"\bdriver\b|\bfront[-_ ]?left\b|\bleft[-_ ]?front\b|"
+    r"主驾|主驾驶(?:座|座椅|席)?|驾驶位|驾驶席|驾驶座(?:椅)?|"
+    r"司机位|司机座(?:椅)?|左前",
     re.IGNORECASE,
 )
 PASSENGER_SEAT_SCOPE_PATTERN = re.compile(
@@ -2257,13 +2259,9 @@ class SeatDemoService:
                 local_preview=local_preview,
                 model_parse_source=model_parse_source,
             )
-            if self._is_dynamic_mode_stop_request(utterance):
+            dynamic_mode_stop = self._is_dynamic_mode_stop_request(utterance)
+            if dynamic_mode_stop:
                 parsed = self._apply_dynamic_mode_stop_request(parsed, utterance)
-            else:
-                parsed = self._apply_mode_profiles(
-                    parsed, matched_modes, context, vehicle_profile
-                )
-            parsed = self._mark_control_parse_source(parsed, "local_rule")
             relative_resolution = (
                 self.native_can_relative_resolver.resolve(utterance, context)
                 if self.native_can_relative_resolver.matches(utterance)
@@ -2282,6 +2280,19 @@ class SeatDemoService:
                     relative_resolution.claimed_controls if relative_resolution else None
                 ),
             )
+            if not dynamic_mode_stop:
+                # Apply repository-backed modes after the final native-CAN
+                # merge.  Otherwise that merge can discard a valid passenger
+                # mode operation while enforcing its generic unscoped-seat
+                # CH1 default.
+                parsed = self._apply_mode_profiles(
+                    parsed,
+                    matched_modes,
+                    context,
+                    vehicle_profile,
+                    utterance=utterance,
+                )
+            parsed = self._mark_control_parse_source(parsed, "local_rule")
         if direct_mode_profile is not None:
             parsed = self._mark_control_parse_source(
                 parsed,
@@ -3327,6 +3338,8 @@ class SeatDemoService:
         matched_modes: list[ModeProfile],
         context: dict[str, Any],
         vehicle_profile: VehicleProfile,
+        *,
+        utterance: str,
     ) -> dict[str, Any]:
         if not matched_modes:
             return parsed
@@ -3354,6 +3367,12 @@ class SeatDemoService:
             ]
             return rejected
 
+        parsed = self._apply_implicit_mode_target_policy(
+            parsed,
+            matched_modes,
+            utterance,
+        )
+
         safety_ambiguities = [
             item
             for item in parsed.get("ambiguities", [])
@@ -3430,6 +3449,38 @@ class SeatDemoService:
         )
         return merged
 
+    @staticmethod
+    def _apply_implicit_mode_target_policy(
+        parsed: dict[str, Any],
+        matched_modes: list[ModeProfile],
+        utterance: str,
+    ) -> dict[str, Any]:
+        """Let one matched mode own only an omitted seat target.
+
+        The native CH1 parser can add an implicit-driver ambiguity before or
+        after mode injection.  A unique repository-backed mode already carries
+        authoritative target scopes on every operation, so that one tagged
+        default is redundant.  Explicit scopes and all other safety questions
+        remain untouched.
+        """
+
+        if len(matched_modes) != 1 or (
+            DRIVER_SEAT_SCOPE_PATTERN.search(utterance)
+            or PASSENGER_SEAT_SCOPE_PATTERN.search(utterance)
+        ):
+            return parsed
+        ambiguities = list(parsed.get("ambiguities", []))
+        filtered = [
+            item
+            for item in ambiguities
+            if item.get("reason") != "implicit_driver_default_unresolved"
+        ]
+        if len(filtered) == len(ambiguities):
+            return parsed
+        resolved = deepcopy(parsed)
+        resolved["ambiguities"] = filtered
+        return resolved
+
     @staticmethod
     def _is_dynamic_mode_stop_request(utterance: str) -> bool:
         compact = re.sub(r"\s+", "", utterance)