strands-agents-evals 0.1.2__py3-none-any.whl → 0.1.4__py3-none-any.whl

{strands_agents_evals-0.1.2.dist-info → strands_agents_evals-0.1.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: strands-agents-evals
-Version: 0.1.2
+Version: 0.1.4
 Summary: Evaluation framework for Strands
 Author-email: AWS <opensource@amazon.com>
 License: Apache-2.0

{strands_agents_evals-0.1.2.dist-info → strands_agents_evals-0.1.4.dist-info}/RECORD

@@ -30,8 +30,8 @@ strands_evals/evaluators/prompt_templates/tool_selection_accuracy/tool_selection
 strands_evals/extractors/__init__.py,sha256=Jmlrk-m8sSS_LwmCVSloIkg3BjOgRzNEezjaAGMw5rw,74
 strands_evals/extractors/graph_extractor.py,sha256=TeT-58JB9roqSvy2ELz1kg8WF5YO-cfLlGZTO0F9s_4,1105
 strands_evals/extractors/swarm_extractor.py,sha256=Sm1XFCkAGVdF3XDyO3iF-20I8C6sAQ8JPNP5fgotOFU,2682
-strands_evals/extractors/tools_use_extractor.py,sha256=o2e9ZuPqQ_hdcrDkl1Rw9h7Ipfc-CsLNkWemxcRvglg,6409
-strands_evals/extractors/trace_extractor.py,sha256=l7gk5rUFoUcxQduPJz49OX66SdgeK1MLt81aF1yr4Lc,6653
+strands_evals/extractors/tools_use_extractor.py,sha256=emLL63LKldL2IA2u5wZL0ZhklZJqX0KLr5xFRt-S4i4,6600
+strands_evals/extractors/trace_extractor.py,sha256=TJKl0OdjFhh-htlV1Wxzem8TQdb0rxa-efkq_e0pAdo,7287
 strands_evals/generators/__init__.py,sha256=B1F30DAIf0kPyBdE4PAZvSby-dTelqb_7hFJoATqVb0,89
 strands_evals/generators/experiment_generator.py,sha256=6wLTL0iG2b0YAiu0w8dDiaBxOIy7p_Fs7l3hCjgQc0w,22655
 strands_evals/generators/topic_planner.py,sha256=FtgTVDlV9hWJyO8E4Z__nEWvvrOJzmTW4y6yZ9Alx1A,2436
@@ -61,8 +61,8 @@ strands_evals/types/evaluation_report.py,sha256=vT86zO4Qn9CQbULo3aziGMdG-1qWLdcB
 strands_evals/types/trace.py,sha256=BFoEylzAlENyPH702T5MDz-_H21-Wfx-FFTSXX1tDfY,4844
 strands_evals/types/simulation/__init__.py,sha256=-mz5lW6qFfIMm4dJGaP9pXY3xeiefLbB0XevjdFykkU,133
 strands_evals/types/simulation/actor.py,sha256=ESTV8165c3Ad5QT4yYmjm-A-oZdwZ0Rf0Lq7zokjTPo,1163
-strands_agents_evals-0.1.2.dist-info/METADATA,sha256=eO-nDiGzAJxDmmwBwPVvNfTdds0408trSL4cKqLR8b4,17721
-strands_agents_evals-0.1.2.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-strands_agents_evals-0.1.2.dist-info/licenses/LICENSE,sha256=yIWKWnZEC7ykhOE0z330Y4XQkN0YssWOZQ-TkliALls,10141
-strands_agents_evals-0.1.2.dist-info/licenses/NOTICE,sha256=Eg13ogOmcI7JpMjxniFnKG81vwU3X8X7P_IlpvVg5RU,66
-strands_agents_evals-0.1.2.dist-info/RECORD,,
+strands_agents_evals-0.1.4.dist-info/METADATA,sha256=VQm_tm1Umm3fi_HfujW0Ovm_XyvQQCjEJrAL4-dGjKQ,17721
+strands_agents_evals-0.1.4.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+strands_agents_evals-0.1.4.dist-info/licenses/LICENSE,sha256=yIWKWnZEC7ykhOE0z330Y4XQkN0YssWOZQ-TkliALls,10141
+strands_agents_evals-0.1.4.dist-info/licenses/NOTICE,sha256=Eg13ogOmcI7JpMjxniFnKG81vwU3X8X7P_IlpvVg5RU,66
+strands_agents_evals-0.1.4.dist-info/RECORD,,

strands_evals/extractors/tools_use_extractor.py

@@ -22,37 +22,39 @@ def extract_agent_tools_used_from_messages(agent_messages):
         if message.get("role") == "assistant":
             message_info = message.get("content")
             if len(message_info) > 0:
-                tool = None
+                tools = []
                 for message in message_info:
                     if "toolUse" in message:
+                        tools.append(message.get("toolUse"))
+
+                for tool in tools:
+                    if tool:
+                        tool_name = tool.get("name")
+                        tool_input = tool.get("input")
+                        tool_id = tool.get("toolUseId")
+                        # get the tool result from the next message
+                        tool_result = None
+                        is_error = False
+                        next_message_i = i + 1
+                        while next_message_i < len(agent_messages):
+                            next_message = agent_messages[next_message_i]
+                            next_message_i += 1
+
+                            if next_message.get("role") == "user":
+                                content = next_message.get("content")
+                                if content:
+                                    tool_result_dict = content[0].get("toolResult")
+                                    if tool_result_dict and tool_result_dict.get("toolUseId") == tool_id:
+                                        tool_result_content = tool_result_dict.get("content", [])
+                                        if len(tool_result_content) > 0:
+                                            tool_result = tool_result_content[0].get("text")
+                                            is_error = tool_result_dict.get("status") == "error"
+                                            break
+
+                        tools_used.append(
+                            {"name": tool_name, "input": tool_input, "tool_result": tool_result, "is_error": is_error}
+                        )
                         tool = message.get("toolUse")
-
-                if tool:
-                    tool_name = tool.get("name")
-                    tool_input = tool.get("input")
-                    tool_id = tool.get("toolUseId")
-                    # get the tool result from the next message
-                    tool_result = None
-                    is_error = False
-                    next_message_i = i + 1
-                    while next_message_i < len(agent_messages):
-                        next_message = agent_messages[next_message_i]
-                        next_message_i += 1
-
-                        if next_message.get("role") == "user":
-                            content = next_message.get("content")
-                            if content:
-                                tool_result_dict = content[0].get("toolResult")
-                                if tool_result_dict and tool_result_dict.get("toolUseId") == tool_id:
-                                    tool_result_content = tool_result_dict.get("content", [])
-                                    if len(tool_result_content) > 0:
-                                        tool_result = tool_result_content[0].get("text")
-                                        is_error = tool_result_dict.get("status") == "error"
-                                        break
-
-                    tools_used.append(
-                        {"name": tool_name, "input": tool_input, "tool_result": tool_result, "is_error": is_error}
-                    )
     return tools_used
 
 

strands_evals/extractors/trace_extractor.py

@@ -45,9 +45,11 @@ class TraceExtractor:
     def _extract_trace_level(self, session: Session) -> list[TraceLevelInput]:
         """Extract trace-level inputs with session history up to each turn."""
         evaluation_inputs: list[TraceLevelInput] = []
-        previous_turns: list[Union[UserMessage, AssistantMessage]] = []
+        previous_turns: list[Union[UserMessage, list[ToolExecution], AssistantMessage]] = []
 
         for trace in session.traces:
+            tool_spans = self._find_tool_execution_spans(trace)
+
             for span in trace.spans:
                 if not isinstance(span, AgentInvocationSpan):
                     continue
@@ -59,6 +61,17 @@ class TraceExtractor:
                     logger.warning(f"Failed to create user message: {e}")
                     continue
 
+                # Include tool executions in session history
+                if tool_spans:
+                    try:
+                        tool_executions = [
+                            ToolExecution(tool_call=ts.tool_call, tool_result=ts.tool_result)
+                            for ts in tool_spans
+                        ]
+                        previous_turns.append(tool_executions)
+                    except (AttributeError, TypeError, ValueError) as e:
+                        logger.warning(f"Failed to create tool executions: {e}")
+
                 trace_input = TraceLevelInput(
                     span_info=span.span_info,
                     agent_response=TextContent(text=span.agent_response),