Fix #2171: Fix nest handoff history duplication

src/agents/handoffs/history.py

@@ -26,6 +26,20 @@
 _conversation_history_start = _DEFAULT_CONVERSATION_HISTORY_START
 _conversation_history_end = _DEFAULT_CONVERSATION_HISTORY_END
 
+# These item types are summarized inside the generated conversation summary.
+# They should not be forwarded verbatim to the next agent.
+_SUMMARY_ONLY_INPUT_TYPES = {
+    "function_call",
+    "function_call_output",
+    "computer_call",
+    "computer_call_output",
+    "file_search_call",
+    "file_search_call_output",
+    "web_search_call",
+    "web_search_call_output",
+    "reasoning",
+}
+
 
 def set_conversation_history_wrappers(
     *,
@@ -67,23 +81,31 @@ def nest_handoff_history(
 
     normalized_history = _normalize_input_history(handoff_input_data.input_history)
     flattened_history = _flatten_nested_history_messages(normalized_history)
-    pre_items_as_inputs = [
-        _run_item_to_plain_input(item) for item in handoff_input_data.pre_handoff_items
-    ]
-    new_items_as_inputs = [_run_item_to_plain_input(item) for item in handoff_input_data.new_items]
+    pre_items_as_inputs: list[TResponseInputItem] = []
+    filtered_pre_items: list[RunItem] = []
+    for run_item in handoff_input_data.pre_handoff_items:
+        plain_input = _run_item_to_plain_input(run_item)
+        pre_items_as_inputs.append(plain_input)
+        if _should_forward_pre_item(plain_input):
+            filtered_pre_items.append(run_item)
+
+    new_items_as_inputs: list[TResponseInputItem] = []
+    filtered_new_items: list[RunItem] = []
+    for run_item in handoff_input_data.new_items:
+        plain_input = _run_item_to_plain_input(run_item)
+        new_items_as_inputs.append(plain_input)
+        if _should_forward_new_item(plain_input):
+            filtered_new_items.append(run_item)
+
     transcript = flattened_history + pre_items_as_inputs + new_items_as_inputs
 
     mapper = history_mapper or default_handoff_history_mapper
     history_items = mapper(transcript)
-    filtered_pre_items = tuple(
-        item
-        for item in handoff_input_data.pre_handoff_items
-        if _get_run_item_role(item) != "assistant"
-    )
 
     return handoff_input_data.clone(
         input_history=tuple(deepcopy(item) for item in history_items),
-        pre_handoff_items=filtered_pre_items,
+        pre_handoff_items=tuple(filtered_pre_items),
+        new_items=tuple(filtered_new_items),
     )
 
 
@@ -231,6 +253,19 @@ def _split_role_and_name(role_text: str) -> tuple[str, str | None]:
     return (role_text or "developer", None)
 
 
-def _get_run_item_role(run_item: RunItem) -> str | None:
-    role_candidate = run_item.to_input_item().get("role")
-    return role_candidate if isinstance(role_candidate, str) else None
+def _should_forward_pre_item(input_item: TResponseInputItem) -> bool:
+    """Return False when the previous transcript item is represented in the summary."""
+    role_candidate = input_item.get("role")
+    if isinstance(role_candidate, str) and role_candidate == "assistant":
+        return False
+    type_candidate = input_item.get("type")
+    return not (isinstance(type_candidate, str) and type_candidate in _SUMMARY_ONLY_INPUT_TYPES)
+
+
+def _should_forward_new_item(input_item: TResponseInputItem) -> bool:
+    """Return False for tool/side-effect items that the summary already covers."""
+    role_candidate = input_item.get("role")
+    if isinstance(role_candidate, str) and role_candidate:
+        return True
+    type_candidate = input_item.get("type")
+    return not (isinstance(type_candidate, str) and type_candidate in _SUMMARY_ONLY_INPUT_TYPES)

tests/test_extension_filters.py

@@ -1,7 +1,11 @@
 from copy import deepcopy
 from typing import Any, cast
 
-from openai.types.responses import ResponseOutputMessage, ResponseOutputText
+from openai.types.responses import (
+    ResponseFunctionToolCall,
+    ResponseOutputMessage,
+    ResponseOutputText,
+)
 from openai.types.responses.response_reasoning_item import ResponseReasoningItem
 
 from agents import (
@@ -17,6 +21,7 @@
     HandoffOutputItem,
     MessageOutputItem,
     ReasoningItem,
+    ToolCallItem,
     ToolCallOutputItem,
     TResponseInputItem,
 )
@@ -81,6 +86,19 @@ def _get_tool_output_run_item(content: str) -> ToolCallOutputItem:
     )
 
 
+def _get_tool_call_run_item(name: str = "transfer_to_agent") -> ToolCallItem:
+    return ToolCallItem(
+        agent=fake_agent(),
+        raw_item=ResponseFunctionToolCall(
+            id="call-1",
+            arguments="{}",
+            call_id="call-1",
+            name=name,
+            type="function_call",
+        ),
+    )
+
+
 def _get_handoff_input_item(content: str) -> TResponseInputItem:
     return {
         "call_id": "1",
@@ -273,7 +291,36 @@ def test_nest_handoff_history_wraps_transcript() -> None:
     assert "Assist reply" in summary_content
     assert "Hello" in summary_content
     assert len(nested.pre_handoff_items) == 0
-    assert nested.new_items == data.new_items
+    assert len(nested.new_items) == 1
+    assert isinstance(nested.new_items[0], MessageOutputItem)
+
+
+def test_nest_handoff_history_filters_tool_calls_and_outputs() -> None:
+    data = HandoffInputData(
+        input_history=(_get_user_input_item("Hello"),),
+        pre_handoff_items=(
+            _get_tool_call_run_item(),
+            _get_tool_output_run_item("result"),
+        ),
+        new_items=(
+            _get_message_output_run_item("Handoff request"),
+            _get_handoff_output_run_item("transfer"),
+        ),
+        run_context=RunContextWrapper(context=()),
+    )
+
+    nested = nest_handoff_history(data)
+
+    assert isinstance(nested.input_history, tuple)
+    assert len(nested.input_history) == 1
+    summary = _as_message(nested.input_history[0])
+    summary_content = summary["content"]
+    assert isinstance(summary_content, str)
+    assert "function_call" in summary_content
+    assert "function_call_output" in summary_content
+    assert nested.pre_handoff_items == ()
+    assert len(nested.new_items) == 1
+    assert isinstance(nested.new_items[0], MessageOutputItem)
 
 
 def test_nest_handoff_history_handles_missing_user() -> None: