feat: add responses input items api (#2239)

# What does this PR do?
TSIA

## Test Plan
added integration and unit tests

tests/integration/agents/test_openai_responses.py

@@ -3,11 +3,7 @@
 #
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
-
-from urllib.parse import urljoin
-
 import pytest
-import requests
 from openai import OpenAI
 
 from llama_stack.distribution.library_client import LlamaStackAsLibraryClient
@@ -80,11 +76,10 @@ def test_responses_store(openai_client, client_with_models, text_model_id, strea
         if not tools:
             content = response.output[0].content[0].text
 
-    # list responses is not available in the SDK
-    url = urljoin(str(client.base_url), "responses")
-    response = requests.get(url, headers={"Authorization": f"Bearer {client.api_key}"})
-    assert response.status_code == 200
-    data = response.json()["data"]
+    # list responses - use the underlying HTTP client for endpoints not in SDK
+    list_response = client._client.get("/responses")
+    assert list_response.status_code == 200
+    data = list_response.json()["data"]
     assert response_id in [r["id"] for r in data]
 
     # test retrieve response
@@ -95,3 +90,133 @@ def test_responses_store(openai_client, client_with_models, text_model_id, strea
         assert retrieved_response.output[0].type == "function_call"
     else:
         assert retrieved_response.output[0].content[0].text == content
+
+
+def test_list_response_input_items(openai_client, client_with_models, text_model_id):
+    """Test the new list_openai_response_input_items endpoint."""
+    if isinstance(client_with_models, LlamaStackAsLibraryClient):
+        pytest.skip("OpenAI responses are not supported when testing with library client yet.")
+
+    client = openai_client
+    message = "What is the capital of France?"
+
+    # Create a response first
+    response = client.responses.create(
+        model=text_model_id,
+        input=[
+            {
+                "role": "user",
+                "content": message,
+            }
+        ],
+        stream=False,
+    )
+
+    response_id = response.id
+
+    # Test the new list input items endpoint
+    input_items_response = client.responses.input_items.list(response_id=response_id)
+
+    # Verify the structure follows OpenAI API spec
+    assert input_items_response.object == "list"
+    assert hasattr(input_items_response, "data")
+    assert isinstance(input_items_response.data, list)
+    assert len(input_items_response.data) > 0
+
+    # Verify the input item contains our message
+    input_item = input_items_response.data[0]
+    assert input_item.type == "message"
+    assert input_item.role == "user"
+    assert message in str(input_item.content)
+
+
+def test_list_response_input_items_with_limit_and_order(openai_client, client_with_models, text_model_id):
+    """Test the list input items endpoint with limit and order parameters."""
+    if isinstance(client_with_models, LlamaStackAsLibraryClient):
+        pytest.skip("OpenAI responses are not supported when testing with library client yet.")
+
+    client = openai_client
+
+    # Create a response with multiple input messages to test limit and order
+    # Use distinctive content to make order verification more reliable
+    messages = [
+        {"role": "user", "content": "Message A: What is the capital of France?"},
+        {"role": "assistant", "content": "The capital of France is Paris."},
+        {"role": "user", "content": "Message B: What about Spain?"},
+        {"role": "assistant", "content": "The capital of Spain is Madrid."},
+        {"role": "user", "content": "Message C: And Italy?"},
+    ]
+
+    response = client.responses.create(
+        model=text_model_id,
+        input=messages,
+        stream=False,
+    )
+
+    response_id = response.id
+
+    # First get all items to establish baseline
+    all_items_response = client.responses.input_items.list(response_id=response_id)
+    assert all_items_response.object == "list"
+    total_items = len(all_items_response.data)
+    assert total_items == 5  # Should have all 5 input messages
+
+    # Test 1: Limit parameter - request only 2 items
+    limited_response = client.responses.input_items.list(response_id=response_id, limit=2)
+    assert limited_response.object == "list"
+    assert len(limited_response.data) == min(2, total_items)  # Should be exactly 2 or total if less
+
+    # Test 2: Edge case - limit larger than available items
+    large_limit_response = client.responses.input_items.list(response_id=response_id, limit=10)
+    assert large_limit_response.object == "list"
+    assert len(large_limit_response.data) == total_items  # Should return all available items
+
+    # Test 3: Edge case - limit of 1
+    single_item_response = client.responses.input_items.list(response_id=response_id, limit=1)
+    assert single_item_response.object == "list"
+    assert len(single_item_response.data) == 1
+
+    # Test 4: Order parameter - ascending vs descending
+    asc_response = client.responses.input_items.list(response_id=response_id, order="asc")
+    desc_response = client.responses.input_items.list(response_id=response_id, order="desc")
+
+    assert asc_response.object == "list"
+    assert desc_response.object == "list"
+    assert len(asc_response.data) == len(desc_response.data) == total_items
+
+    # Verify order is actually different (if we have multiple items)
+    if total_items > 1:
+        # First item in asc should be last item in desc (reversed order)
+        first_asc_content = str(asc_response.data[0].content)
+        first_desc_content = str(desc_response.data[0].content)
+        last_asc_content = str(asc_response.data[-1].content)
+        last_desc_content = str(desc_response.data[-1].content)
+
+        # The first item in asc should be the last item in desc (and vice versa)
+        assert first_asc_content == last_desc_content, (
+            f"Expected first asc ({first_asc_content}) to equal last desc ({last_desc_content})"
+        )
+        assert last_asc_content == first_desc_content, (
+            f"Expected last asc ({last_asc_content}) to equal first desc ({first_desc_content})"
+        )
+
+        # Verify the distinctive content markers are in the right positions
+        assert "Message A" in first_asc_content, "First item in ascending order should contain 'Message A'"
+        assert "Message C" in first_desc_content, "First item in descending order should contain 'Message C'"
+
+    # Test 5: Combined limit and order
+    combined_response = client.responses.input_items.list(response_id=response_id, limit=3, order="desc")
+    assert combined_response.object == "list"
+    assert len(combined_response.data) == min(3, total_items)
+
+    # Test 6: Verify combined response has correct order for first few items
+    if total_items >= 3:
+        # Should get the last 3 items in descending order (most recent first)
+        assert "Message C" in str(combined_response.data[0].content), "First item should be most recent (Message C)"
+        # The exact second and third items depend on the implementation, but let's verify structure
+        for item in combined_response.data:
+            assert hasattr(item, "content")
+            assert hasattr(item, "role")
+            assert hasattr(item, "type")
+            assert item.type == "message"
+            assert item.role in ["user", "assistant"]

tests/unit/providers/agents/meta_reference/test_openai_responses.py

@@ -15,7 +15,9 @@ from openai.types.chat.chat_completion_chunk import (
     ChoiceDeltaToolCallFunction,
 )
 
+from llama_stack.apis.agents import Order
 from llama_stack.apis.agents.openai_responses import (
+    ListOpenAIResponseInputItem,
     OpenAIResponseInputMessageContentText,
     OpenAIResponseInputToolFunction,
     OpenAIResponseInputToolWebSearch,
@@ -504,3 +506,117 @@ async def test_create_openai_response_with_instructions_and_previous_response(
     assert sent_messages[2].content == "Galway, Longford, Sligo"
     assert sent_messages[3].role == "user"
     assert sent_messages[3].content == "Which is the largest?"
+
+
+@pytest.mark.asyncio
+async def test_list_openai_response_input_items_delegation(openai_responses_impl, mock_responses_store):
+    """Test that list_openai_response_input_items properly delegates to responses_store with correct parameters."""
+    # Setup
+    response_id = "resp_123"
+    after = "msg_after"
+    before = "msg_before"
+    include = ["metadata"]
+    limit = 5
+    order = Order.asc
+
+    input_message = OpenAIResponseMessage(
+        id="msg_123",
+        content="Test message",
+        role="user",
+    )
+
+    expected_result = ListOpenAIResponseInputItem(data=[input_message])
+    mock_responses_store.list_response_input_items.return_value = expected_result
+
+    # Execute with all parameters to test delegation
+    result = await openai_responses_impl.list_openai_response_input_items(
+        response_id, after=after, before=before, include=include, limit=limit, order=order
+    )
+
+    # Verify all parameters are passed through correctly to the store
+    mock_responses_store.list_response_input_items.assert_called_once_with(
+        response_id, after, before, include, limit, order
+    )
+
+    # Verify the result is returned as-is from the store
+    assert result.object == "list"
+    assert len(result.data) == 1
+    assert result.data[0].id == "msg_123"
+
+
+@pytest.mark.asyncio
+async def test_responses_store_list_input_items_logic():
+    """Test ResponsesStore list_response_input_items logic - mocks get_response_object to test actual ordering/limiting."""
+
+    # Create mock store and response store
+    mock_sql_store = AsyncMock()
+    responses_store = ResponsesStore(sql_store_config=None)
+    responses_store.sql_store = mock_sql_store
+
+    # Setup test data - multiple input items
+    input_items = [
+        OpenAIResponseMessage(id="msg_1", content="First message", role="user"),
+        OpenAIResponseMessage(id="msg_2", content="Second message", role="user"),
+        OpenAIResponseMessage(id="msg_3", content="Third message", role="user"),
+        OpenAIResponseMessage(id="msg_4", content="Fourth message", role="user"),
+    ]
+
+    response_with_input = OpenAIResponseObjectWithInput(
+        id="resp_123",
+        model="test_model",
+        created_at=1234567890,
+        object="response",
+        status="completed",
+        output=[],
+        input=input_items,
+    )
+
+    # Mock the get_response_object method to return our test data
+    mock_sql_store.fetch_one.return_value = {"response_object": response_with_input.model_dump()}
+
+    # Test 1: Default behavior (no limit, desc order)
+    result = await responses_store.list_response_input_items("resp_123")
+    assert result.object == "list"
+    assert len(result.data) == 4
+    # Should be reversed for desc order
+    assert result.data[0].id == "msg_4"
+    assert result.data[1].id == "msg_3"
+    assert result.data[2].id == "msg_2"
+    assert result.data[3].id == "msg_1"
+
+    # Test 2: With limit=2, desc order
+    result = await responses_store.list_response_input_items("resp_123", limit=2, order=Order.desc)
+    assert result.object == "list"
+    assert len(result.data) == 2
+    # Should be first 2 items in desc order
+    assert result.data[0].id == "msg_4"
+    assert result.data[1].id == "msg_3"
+
+    # Test 3: With limit=2, asc order
+    result = await responses_store.list_response_input_items("resp_123", limit=2, order=Order.asc)
+    assert result.object == "list"
+    assert len(result.data) == 2
+    # Should be first 2 items in original order (asc)
+    assert result.data[0].id == "msg_1"
+    assert result.data[1].id == "msg_2"
+
+    # Test 4: Asc order without limit
+    result = await responses_store.list_response_input_items("resp_123", order=Order.asc)
+    assert result.object == "list"
+    assert len(result.data) == 4
+    # Should be in original order (asc)
+    assert result.data[0].id == "msg_1"
+    assert result.data[1].id == "msg_2"
+    assert result.data[2].id == "msg_3"
+    assert result.data[3].id == "msg_4"
+
+    # Test 5: Large limit (larger than available items)
+    result = await responses_store.list_response_input_items("resp_123", limit=10, order=Order.desc)
+    assert result.object == "list"
+    assert len(result.data) == 4  # Should return all available items
+    assert result.data[0].id == "msg_4"
+
+    # Test 6: Zero limit edge case
+    result = await responses_store.list_response_input_items("resp_123", limit=0, order=Order.asc)
+    assert result.object == "list"
+    assert len(result.data) == 0  # Should return no items