add support for tool call message in output scope

Author: nikhilc-microsoft

libraries/microsoft-agents-a365-observability-core/microsoft_agents_a365/observability/core/execute_tool_scope.py

@@ -23,7 +23,7 @@
     USER_NAME_KEY,
 )
 from .message_utils import serialize_messages
-from .models.messages import ToolInputMessages, ToolOutputMessages
+from .models.messages import ToolOutputMessages
 from .models.user_details import UserDetails
 from .opentelemetry_scope import OpenTelemetryScope
 from .request import Request
@@ -112,7 +112,7 @@ def __init__(
 
         self.set_tag_maybe(GEN_AI_TOOL_NAME_KEY, tool_name)
         if arguments is not None:
-            self.record_tool_input(arguments)
+            self.set_tag_maybe(GEN_AI_TOOL_ARGS_KEY, serialize_messages(arguments))
         self.set_tag_maybe(GEN_AI_TOOL_TYPE_KEY, tool_type)
         self.set_tag_maybe(GEN_AI_TOOL_CALL_ID_KEY, tool_call_id)
         self.set_tag_maybe(GEN_AI_TOOL_DESCRIPTION_KEY, description)
@@ -138,15 +138,7 @@ def __init__(
                 validate_and_normalize_ip(user_details.user_client_ip),
             )
 
-    def record_tool_input(self, messages: ToolInputMessages) -> None:
-        """Record the tool input for telemetry tracking.
-
-        Args:
-            messages: A ToolInputMessages wrapper containing tool call requests
-        """
-        self.set_tag_maybe(GEN_AI_TOOL_ARGS_KEY, serialize_messages(messages))
-
-    def record_tool_output(self, messages: ToolOutputMessages) -> None:
+    def record_response(self, messages: ToolOutputMessages) -> None:
         """Record the tool output for telemetry tracking.
 
         Args:

libraries/microsoft-agents-a365-observability-core/microsoft_agents_a365/observability/core/models/response.py

@@ -4,15 +4,23 @@
 from __future__ import annotations
 
 from dataclasses import dataclass
+from typing import Union
 
-from .messages import OutputMessagesParam
+from .messages import OutputMessagesParam, ToolOutputMessages
+
+ResponseMessagesParam = Union[OutputMessagesParam, ToolOutputMessages]
+"""Accepted type for Response.messages.
+
+Supports plain strings, OutputMessages, or ToolOutputMessages.
+"""
 
 
 @dataclass
 class Response:
     """Response details from agent execution.
 
-    Accepts plain strings (backward compat) or structured OTEL OutputMessages.
+    Accepts plain strings (backward compat), structured OTEL OutputMessages,
+    or ToolOutputMessages.
     """
 
-    messages: OutputMessagesParam
+    messages: ResponseMessagesParam

libraries/microsoft-agents-a365-observability-core/microsoft_agents_a365/observability/core/spans_scopes/output_scope.py

@@ -16,9 +16,9 @@
 from ..models.messages import (
     OutputMessage,
     OutputMessages,
-    OutputMessagesParam,
+    ToolOutputMessages,
 )
-from ..models.response import Response
+from ..models.response import Response, ResponseMessagesParam
 from ..models.user_details import UserDetails
 from ..opentelemetry_scope import OpenTelemetryScope
 from ..request import Request
@@ -94,14 +94,19 @@ def __init__(
 
         self.set_tag_maybe(GEN_AI_CONVERSATION_ID_KEY, request.conversation_id)
 
-        # Normalize response messages and extract inner messages for accumulation
-        normalized = normalize_output_messages(response.messages)
-        self._output_messages: list[OutputMessage] = list(normalized.messages)
-        self._output_messages_dirty = False
-
-        # Set initial output messages attribute as the full versioned wrapper
-        wrapper = OutputMessages(messages=self._output_messages)
-        self.set_tag_maybe(GEN_AI_OUTPUT_MESSAGES_KEY, serialize_messages(wrapper))
+        # Handle tool output messages vs regular output messages
+        if isinstance(response.messages, ToolOutputMessages):
+            # Tool output: serialize directly, no accumulation
+            self._output_messages: list[OutputMessage] = []
+            self._output_messages_dirty = False
+            self.set_tag_maybe(GEN_AI_OUTPUT_MESSAGES_KEY, serialize_messages(response.messages))
+        else:
+            # Regular output: normalize and set up accumulation
+            normalized = normalize_output_messages(response.messages)
+            self._output_messages = list(normalized.messages)
+            self._output_messages_dirty = False
+            wrapper = OutputMessages(messages=self._output_messages)
+            self.set_tag_maybe(GEN_AI_OUTPUT_MESSAGES_KEY, serialize_messages(wrapper))
 
         # Set user details if provided
         if user_details:
@@ -113,18 +118,22 @@ def __init__(
                 validate_and_normalize_ip(user_details.user_client_ip),
             )
 
-    def record_output_messages(self, messages: OutputMessagesParam) -> None:
+    def record_output_messages(self, messages: ResponseMessagesParam) -> None:
         """Records the output messages for telemetry tracking.
 
         Appends the provided messages to the accumulated output messages list.
-        Accepts plain strings (auto-wrapped as OTEL OutputMessage) or a versioned
-        OutputMessages wrapper.
+        Accepts plain strings (auto-wrapped as OTEL OutputMessage), a versioned
+        OutputMessages wrapper, or a ToolOutputMessages wrapper.
         The list is capped at _MAX_OUTPUT_MESSAGES to prevent unbounded memory growth.
         The updated attribute is flushed when the scope is disposed.
 
         Args:
-            messages: List of output message strings or an OutputMessages wrapper to append
+            messages: Plain strings, OutputMessages, or ToolOutputMessages to append
         """
+        if isinstance(messages, ToolOutputMessages):
+            # Tool output: serialize directly, overwrite attribute
+            self.set_tag_maybe(GEN_AI_OUTPUT_MESSAGES_KEY, serialize_messages(messages))
+            return
         normalized = normalize_output_messages(messages)
         self._output_messages.extend(normalized.messages)
         if len(self._output_messages) > self._MAX_OUTPUT_MESSAGES:

tests/observability/core/test_execute_tool_scope.py

@@ -97,22 +97,13 @@ def tearDown(self):
 
         self.span_exporter.clear()
 
-    def test_record_tool_input_method_exists(self):
-        """Test that record_tool_input method exists on ExecuteToolScope."""
+    def test_record_response_method_exists(self):
+        """Test that record_response method exists on ExecuteToolScope."""
         scope = ExecuteToolScope.start(Request(), self.tool_details, self.agent_details)
 
         if scope is not None:
-            self.assertTrue(hasattr(scope, "record_tool_input"))
-            self.assertTrue(callable(scope.record_tool_input))
-            scope.dispose()
-
-    def test_record_tool_output_method_exists(self):
-        """Test that record_tool_output method exists on ExecuteToolScope."""
-        scope = ExecuteToolScope.start(Request(), self.tool_details, self.agent_details)
-
-        if scope is not None:
-            self.assertTrue(hasattr(scope, "record_tool_output"))
-            self.assertTrue(callable(scope.record_tool_output))
+            self.assertTrue(hasattr(scope, "record_response"))
+            self.assertTrue(callable(scope.record_response))
             scope.dispose()
 
     def test_request_metadata_set_on_span(self):