feat(integration): add `gen_ai.conversation.id` if available #5307

cursor · 2026-02-02T14:25:59Z

Missing conversation_id on workflow span in async runner

Medium Severity

In _create_run_wrapper, the comment says "Set conversation ID on workflow span early" but the code only stores conversation_id on the agent object (agent._sentry_conversation_id). Unlike the streaming version _create_run_streamed_wrapper, there's no call to set_data(SPANDATA.GEN_AI_CONVERSATION_ID, conversation_id) on the workflow span itself. The workflow span (transaction) will be missing the gen_ai.conversation.id field when using Runner.run().

sentry · 2026-02-02T17:32:37Z

Bug: The async wrapper _create_run_wrapper fails to set the conversation_id on the workflow span, unlike its synchronous counterpart, leading to missing telemetry data.
_{Severity: MEDIUM}

Suggested Fix

Modify the _create_run_wrapper function to get a handle on the span created by agent_workflow_span. Then, call workflow_span.set_data(SPANDATA.GEN_AI_CONVERSATION_ID, conversation_id) within the if conversation_id: block, similar to the implementation in the synchronous wrapper.

Prompt for AI Agent

Review the code at the location below. A potential bug has been identified by an AI agent. Verify if this is a real issue. If it is, propose a fix; if not, explain why it's not valid. Location: sentry_sdk/integrations/openai_agents/patches/runner.py#L37-L43 Potential issue: In the async wrapper function `_create_run_wrapper`, the `conversation_id` is set on `agent._sentry_conversation_id` but is not attached to the workflow span itself. This is inconsistent with the synchronous wrapper `_create_run_streamed_wrapper`, which explicitly calls `workflow_span.set_data()` to attach the ID. The comment in the async function, "Set conversation ID on workflow span early", indicates the intent was to set it on the span. As a result, when async agent workflows are executed, the root workflow span will be missing the `gen_ai.conversation.id` attribute, leading to inconsistent observability data between async and sync execution paths.

_{Did we get this right? 👍 / 👎 to inform future reviews.}

sentry · 2026-02-02T17:34:49Z

Bug: The async wrapper for _create_run_wrapper fails to set the conversation_id on the root workflow_span, unlike the sync implementation, causing inconsistent tracing data.
_{Severity: MEDIUM}

Suggested Fix

In the async wrapper within _create_run_wrapper, after setting agent._sentry_conversation_id, add a call to workflow_span.set_data(SPANDATA.GEN_AI_CONVERSATION_ID, conversation_id) to ensure the root span also has the conversation ID, matching the sync wrapper's behavior.

Prompt for AI Agent

Review the code at the location below. A potential bug has been identified by an AI agent. Verify if this is a real issue. If it is, propose a fix; if not, explain why it's not valid. Location: sentry_sdk/integrations/openai_agents/patches/runner.py#L37-L44 Potential issue: In the async implementation of `_create_run_wrapper`, the `conversation_id` is assigned to `agent._sentry_conversation_id`, but it is not set on the root `workflow_span` itself. This is inconsistent with the sync/streaming implementation, which does set the ID on the root span. As a result, the root transaction span for asynchronous agent runs will be missing the `gen_ai.conversation.id` attribute. This leads to an incomplete tracing experience, as the comment "Set conversation ID on workflow span early" suggests the intention was to capture it on the span.

_{Did we get this right? 👍 / 👎 to inform future reviews.}

cursor · 2026-02-02T14:25:59Z

Callers don't pass agent to update_ai_client_span

Medium Severity

The update_ai_client_span function was modified to accept a new agent parameter (defaulting to None) for setting conversation_id on AI client spans. However, the callers in models.py at lines 128 and 179 were not updated to pass the agent argument. Since agent defaults to None, the conversation_id condition if agent: is never true, and gen_ai.conversation.id is never set on AI client spans.

cursor · 2026-02-02T14:25:59Z

Duplicated pattern for setting conversation ID on spans

Low Severity

The same 3-line pattern for extracting and setting _sentry_conversation_id from an agent onto a span is repeated 4 times across execute_tool.py, handoff.py, invoke_agent.py, and ai_client.py. This pattern should be extracted to a utility function in utils.py (e.g., _set_conversation_id(span, agent)), following the existing pattern of _set_agent_data, _set_usage_data, etc.

-Original file line number
+Diff line change
@@ Expand Up / @@ -465,6 +465,12 @@ class SPANDATA: @@
         Example: "ResearchAssistant"
         """
+        GEN_AI_CONVERSATION_ID = "gen_ai.conversation.id"
+        """
+        The unique identifier for the conversation/thread with the AI model.
+        Example: "conv_abc123"
+        """
         GEN_AI_CHOICE = "gen_ai.choice"
         """
         The model's response message.
@@ Expand Down @@

-Original file line number
+Diff line change
@@ Expand Up / @@ -2,6 +2,7 @@ @@
     from functools import wraps
     import sentry_sdk
+    from sentry_sdk.consts import SPANDATA
     from sentry_sdk.integrations import DidNotEnable
     from sentry_sdk.utils import capture_internal_exceptions, reraise
@@ Expand Down Expand Up / @@ -34,7 +35,13 @@ async def wrapper(*args: "Any", **kwargs: "Any") -> "Any": @@
             with sentry_sdk.isolation_scope():
                 # Clone agent because agent invocation spans are attached per run.
                 agent = args[0].clone()
                 with agent_workflow_span(agent):
+                    # Set conversation ID on workflow span early so it's captured even on errors
+                    conversation_id = kwargs.get("conversation_id")
+                    if conversation_id:
+                        agent._sentry_conversation_id = conversation_id
                     args = (agent, *args[1:])
                     try:
                         run_result = await original_func(*args, **kwargs)
@@ Expand Down Expand Up / @@ -91,10 +98,19 @@ def wrapper(*args: "Any", **kwargs: "Any") -> "Any": @@
             # Clone agent because agent invocation spans are attached per run.
             agent = args[0].clone()
+            # Capture conversation_id from kwargs if provided
+            conversation_id = kwargs.get("conversation_id")
+            if conversation_id:
+                agent._sentry_conversation_id = conversation_id
             # Start workflow span immediately (before run_streamed returns)
             workflow_span = agent_workflow_span(agent)
             workflow_span.__enter__()
+            # Set conversation ID on workflow span early so it's captured even on errors
+            if conversation_id:
+                workflow_span.set_data(SPANDATA.GEN_AI_CONVERSATION_ID, conversation_id)
             # Store span on agent for cleanup
             agent._sentry_workflow_span = workflow_span
@@ Expand Down @@

-Original file line number
+Diff line change
@@ Expand Up / @@ -46,6 +46,7 @@ def update_ai_client_span( @@
         span: "sentry_sdk.tracing.Span",
         response: "Any",
         response_model: "Optional[str]" = None,
+        agent: "Optional[Agent]" = None,
     ) -> None:
         """Update AI client span with response data (works for streaming and non-streaming)."""
         if hasattr(response, "usage") and response.usage:
@@ Expand All / @@ -59,3 +60,9 @@ def update_ai_client_span( @@
             span.set_data(SPANDATA.GEN_AI_RESPONSE_MODEL, response_model)
         elif hasattr(response, "model") and response.model:
             span.set_data(SPANDATA.GEN_AI_RESPONSE_MODEL, str(response.model))
+        # Set conversation ID from agent if available
+        if agent:
+            conv_id = getattr(agent, "_sentry_conversation_id", None)
+            if conv_id:
+                span.set_data(SPANDATA.GEN_AI_CONVERSATION_ID, conv_id)

-Original file line number
+Diff line change
@@ Expand Up / @@ -51,3 +51,8 @@ def update_execute_tool_span( @@
         if should_send_default_pii():
             span.set_data(SPANDATA.GEN_AI_TOOL_OUTPUT, result)
+        # Add conversation ID from agent
+        conv_id = getattr(agent, "_sentry_conversation_id", None)
+        if conv_id:
+            span.set_data(SPANDATA.GEN_AI_CONVERSATION_ID, conv_id)

-Original file line number
+Diff line change
@@ Expand Up / @@ -18,3 +18,8 @@ def handoff_span( @@
             origin=SPAN_ORIGIN,
         ) as span:
             span.set_data(SPANDATA.GEN_AI_OPERATION_NAME, "handoff")
+            # Add conversation ID from agent
+            conv_id = getattr(from_agent, "_sentry_conversation_id", None)
+            if conv_id:
+                span.set_data(SPANDATA.GEN_AI_CONVERSATION_ID, conv_id)

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(integration): add `gen_ai.conversation.id` if available #5307

Diff view

Diff view

There are no files selected for viewing

cursor bot Feb 2, 2026

Uh oh!

sentry bot Feb 2, 2026

Uh oh!

sentry bot Feb 2, 2026

Uh oh!

cursor bot Feb 2, 2026

Uh oh!

cursor bot Feb 2, 2026

Uh oh!

Uh oh!

-Original file line number
+Diff line change
@@ Expand Up / @@ -2710,3 +2710,182 @@ def mock_get_model(agent, run_config): @@
             # Verify streaming flag is set
             assert chat_span._data.get(SPANDATA.GEN_AI_RESPONSE_STREAMING) is True
+    @pytest.mark.asyncio
+    async def test_conversation_id_on_all_spans(
+        sentry_init, capture_events, test_agent, mock_model_response
+    ):
+        """
+        Test that gen_ai.conversation.id is set on all AI-related spans when passed to Runner.run().
+        """
+        with patch.dict(os.environ, {"OPENAI_API_KEY": "test-key"}):
+            with patch(
+                "agents.models.openai_responses.OpenAIResponsesModel.get_response"
+            ) as mock_get_response:
+                mock_get_response.return_value = mock_model_response
+                sentry_init(
+                    integrations=[OpenAIAgentsIntegration()],
+                    traces_sample_rate=1.0,
+                )
+                events = capture_events()
+                result = await agents.Runner.run(
+                    test_agent,
+                    "Test input",
+                    run_config=test_run_config,
+                    conversation_id="conv_test_123",
+                )
+                assert result is not None
+        (transaction,) = events
+        spans = transaction["spans"]
+        invoke_agent_span, ai_client_span = spans
+        # Verify workflow span (transaction) has conversation_id
+        assert (
+            transaction["contexts"]["trace"]["data"]["gen_ai.conversation.id"]
+            == "conv_test_123"
+        )
+        # Verify invoke_agent span has conversation_id
+        assert invoke_agent_span["data"]["gen_ai.conversation.id"] == "conv_test_123"
+        # Verify ai_client span has conversation_id
+        assert ai_client_span["data"]["gen_ai.conversation.id"] == "conv_test_123"
+    @pytest.mark.asyncio
+    async def test_conversation_id_on_tool_span(sentry_init, capture_events, test_agent):
+        """
+        Test that gen_ai.conversation.id is set on tool execution spans when passed to Runner.run().
+        """
+        @agents.function_tool
+        def simple_tool(message: str) -> str:
+            """A simple tool"""
+            return f"Result: {message}"
+        agent_with_tool = test_agent.clone(tools=[simple_tool])
+        with patch.dict(os.environ, {"OPENAI_API_KEY": "test-key"}):
+            with patch(
+                "agents.models.openai_responses.OpenAIResponsesModel.get_response"
+            ) as mock_get_response:
+                tool_call = ResponseFunctionToolCall(
+                    id="call_123",
+                    call_id="call_123",
+                    name="simple_tool",
+                    type="function_call",
+                    arguments='{"message": "hello"}',
+                )
+                tool_response = ModelResponse(
+                    output=[tool_call],
+                    usage=Usage(
+                        requests=1, input_tokens=10, output_tokens=5, total_tokens=15
+                    ),
+                    response_id="resp_tool_456",
+                )
+                final_response = ModelResponse(
+                    output=[
+                        ResponseOutputMessage(
+                            id="msg_final",
+                            type="message",
+                            status="completed",
+                            content=[
+                                ResponseOutputText(
+                                    text="Done",
+                                    type="output_text",
+                                    annotations=[],
+                                )
+                            ],
+                            role="assistant",
+                        )
+                    ],
+                    usage=Usage(
+                        requests=1, input_tokens=15, output_tokens=10, total_tokens=25
+                    ),
+                    response_id="resp_final_789",
+                )
+                mock_get_response.side_effect = [tool_response, final_response]
+                sentry_init(
+                    integrations=[OpenAIAgentsIntegration()],
+                    traces_sample_rate=1.0,
+                )
+                events = capture_events()
+                await agents.Runner.run(
+                    agent_with_tool,
+                    "Use the tool",
+                    run_config=test_run_config,
+                    conversation_id="conv_tool_test_456",
+                )
+        (transaction,) = events
+        spans = transaction["spans"]
+        # Find the tool span
+        tool_span = None
+        for span in spans:
+            if span.get("description", "").startswith("execute_tool"):
+                tool_span = span
+                break
+        assert tool_span is not None
+        # Tool span should have the conversation_id passed to Runner.run()
+        assert tool_span["data"]["gen_ai.conversation.id"] == "conv_tool_test_456"
+        # Workflow span (transaction) should have the same conversation_id
+        assert (
+            transaction["contexts"]["trace"]["data"]["gen_ai.conversation.id"]
+            == "conv_tool_test_456"
+        )
+    @pytest.mark.asyncio
+    async def test_no_conversation_id_when_not_provided(
+        sentry_init, capture_events, test_agent, mock_model_response
+    ):
+        """
+        Test that gen_ai.conversation.id is not set when not passed to Runner.run().
+        """
+        with patch.dict(os.environ, {"OPENAI_API_KEY": "test-key"}):
+            with patch(
+                "agents.models.openai_responses.OpenAIResponsesModel.get_response"
+            ) as mock_get_response:
+                mock_get_response.return_value = mock_model_response
+                sentry_init(
+                    integrations=[OpenAIAgentsIntegration()],
+                    traces_sample_rate=1.0,
+                )
+                events = capture_events()
+                # Don't pass conversation_id
+                result = await agents.Runner.run(
+                    test_agent, "Test input", run_config=test_run_config
+                )
+                assert result is not None
+        (transaction,) = events
+        spans = transaction["spans"]
+        invoke_agent_span, ai_client_span = spans
+        # Verify conversation_id is NOT set on any spans
+        assert "gen_ai.conversation.id" not in transaction["contexts"]["trace"].get(
+            "data", {}
+        )
+        assert "gen_ai.conversation.id" not in invoke_agent_span.get("data", {})
+        assert "gen_ai.conversation.id" not in ai_client_span.get("data", {})

feat(integration): add gen_ai.conversation.id if available #5307

Are you sure you want to change the base?

feat(integration): add gen_ai.conversation.id if available #5307

Uh oh!

Uh oh!

Diff view

Diff view

There are no files selected for viewing

cursor bot Feb 2, 2026

Choose a reason for hiding this comment

Missing conversation_id on workflow span in async runner

Uh oh!

sentry bot Feb 2, 2026

Choose a reason for hiding this comment

Uh oh!

sentry bot Feb 2, 2026

Choose a reason for hiding this comment

Uh oh!

cursor bot Feb 2, 2026

Choose a reason for hiding this comment

Callers don't pass agent to update_ai_client_span

Uh oh!

cursor bot Feb 2, 2026

Choose a reason for hiding this comment

Duplicated pattern for setting conversation ID on spans

Uh oh!

Uh oh!

feat(integration): add `gen_ai.conversation.id` if available #5307

feat(integration): add `gen_ai.conversation.id` if available #5307