fix: preserve output type in FunctionSpanData.export()

FunctionSpanData.export() applied str(self.output) if self.output else
None, which corrupted outputs in two ways:

1. Dict/list outputs were converted to Python repr strings (single
   quotes, Python booleans) instead of remaining as dicts that
   json.dumps can serialize correctly.

2. Falsy but valid outputs (0, False, empty string, empty list) were
   silently replaced with None.

This is inconsistent with GenerationSpanData.export(), which passes
self.output through directly.

The fix removes the str() conversion to match the sibling pattern.
Updated MCP tracing snapshots accordingly and added unit tests covering
dict, string, None, falsy, list, and numeric output values.

Author: lawrence3699

src/agents/tracing/span_data.py

@@ -88,7 +88,7 @@ def export(self) -> dict[str, Any]:
             "type": self.type,
             "name": self.name,
             "input": self.input,
-            "output": str(self.output) if self.output else None,
+            "output": self.output,
             "mcp_data": self.mcp_data,
         }
 

tests/mcp/test_mcp_tracing.py

@@ -62,7 +62,7 @@ async def test_mcp_tracing():
                                 "data": {
                                     "name": "test_tool_1",
                                     "input": "",
-                                    "output": "{'type': 'text', 'text': 'result_test_tool_1_{}'}",  # noqa: E501
+                                    "output": {"type": "text", "text": "result_test_tool_1_{}"},  # noqa: E501
                                     "mcp_data": {"server": "fake_mcp_server"},
                                 },
                             },
@@ -133,7 +133,7 @@ async def test_mcp_tracing():
                                 "data": {
                                     "name": "test_tool_2",
                                     "input": "",
-                                    "output": "{'type': 'text', 'text': 'result_test_tool_2_{}'}",  # noqa: E501
+                                    "output": {"type": "text", "text": "result_test_tool_2_{}"},  # noqa: E501
                                     "mcp_data": {"server": "fake_mcp_server"},
                                 },
                             },
@@ -197,7 +197,7 @@ async def test_mcp_tracing():
                                 "data": {
                                     "name": "test_tool_3",
                                     "input": "",
-                                    "output": "{'type': 'text', 'text': 'result_test_tool_3_{}'}",  # noqa: E501
+                                    "output": {"type": "text", "text": "result_test_tool_3_{}"},  # noqa: E501
                                     "mcp_data": {"server": "fake_mcp_server"},
                                 },
                             },

tests/tracing/test_span_data.py

@@ -0,0 +1,67 @@
+"""Tests for span data export methods."""
+
+from __future__ import annotations
+
+import pytest
+
+from agents.tracing.span_data import FunctionSpanData
+
+
+class TestFunctionSpanDataExport:
+    """FunctionSpanData.export() must preserve output values faithfully."""
+
+    def test_dict_output_preserved_as_dict(self) -> None:
+        """Dict outputs should stay as dicts, not be converted to Python repr strings."""
+        span = FunctionSpanData(name="my_tool", input="query", output={"key": "value", "n": 42})
+        exported = span.export()
+        assert exported["output"] == {"key": "value", "n": 42}
+        assert isinstance(exported["output"], dict)
+
+    def test_string_output_preserved(self) -> None:
+        span = FunctionSpanData(name="my_tool", input="query", output="hello world")
+        exported = span.export()
+        assert exported["output"] == "hello world"
+
+    def test_none_output_preserved(self) -> None:
+        span = FunctionSpanData(name="my_tool", input="query", output=None)
+        exported = span.export()
+        assert exported["output"] is None
+
+    @pytest.mark.parametrize(
+        "output",
+        [0, False, "", []],
+        ids=["zero", "false", "empty_str", "empty_list"],
+    )
+    def test_falsy_output_not_converted_to_none(self, output: object) -> None:
+        """Falsy but valid outputs (0, False, '', []) must not become None."""
+        span = FunctionSpanData(name="my_tool", input="query", output=output)
+        exported = span.export()
+        assert exported["output"] is not None
+        assert exported["output"] == output
+
+    def test_list_output_preserved(self) -> None:
+        span = FunctionSpanData(name="my_tool", input="query", output=[1, 2, 3])
+        exported = span.export()
+        assert exported["output"] == [1, 2, 3]
+        assert isinstance(exported["output"], list)
+
+    def test_numeric_output_preserved(self) -> None:
+        span = FunctionSpanData(name="my_tool", input="query", output=42)
+        exported = span.export()
+        assert exported["output"] == 42
+
+    def test_export_includes_all_fields(self) -> None:
+        span = FunctionSpanData(
+            name="my_tool",
+            input="query",
+            output="result",
+            mcp_data={"server": "test"},
+        )
+        exported = span.export()
+        assert exported == {
+            "type": "function",
+            "name": "my_tool",
+            "input": "query",
+            "output": "result",
+            "mcp_data": {"server": "test"},
+        }