fix(responses-api): use list format with input_text for tool results (#18257)

The Responses API expects tool results to use input_text/input_image types,
not output_text. This fix ensures consistent list format for all tool results:
- String content → [{"type": "input_text", "text": "..."}]
- Image content → [{"type": "input_image", "image_url": "..."}]

This resolves the conflict between tests that expected different formats
and aligns with OpenAI's Responses API requirements.

Fixes the regression introduced in #18226.

litellm/completion_extras/litellm_responses_transformation/transformation.py

@@ -167,28 +167,27 @@ class LiteLLMResponsesTransformationHandler(CompletionTransformationBridge):
                     )
             elif role == "tool":
                 # Convert tool message to function call output format
-                # The Responses API expects 'output' to be a string, not a list
+                # The Responses API expects 'output' to be a list with input_text/input_image types
+                # Using list format for consistency across text and multimodal content
+                tool_output: List[Dict[str, Any]]
                 if content is None:
-                    output_str = ""
+                    tool_output = []
                 elif isinstance(content, str):
-                    output_str = content
+                    # Convert string to list with input_text
+                    tool_output = [{"type": "input_text", "text": content}]
                 elif isinstance(content, list):
-                    # If content is a list, extract text parts and join them
-                    text_parts = []
-                    for item in content:
-                        if isinstance(item, str):
-                            text_parts.append(item)
-                        elif isinstance(item, dict) and item.get("type") == "text":
-                            text_parts.append(item.get("text", ""))
-                    output_str = " ".join(text_parts) if text_parts else str(content)
+                    # Transform list content to Responses API format
+                    tool_output = self._convert_content_to_responses_format(
+                        content, "user"  # Use "user" role to get input_* types
+                    )
                 else:
-                    # Fallback: convert unexpected types to string
-                    output_str = str(content)
+                    # Fallback: convert unexpected types to input_text
+                    tool_output = [{"type": "input_text", "text": str(content)}]
                 input_items.append(
                     {
                         "type": "function_call_output",
                         "call_id": tool_call_id,
-                        "output": output_str,
+                        "output": tool_output,
                     }
                 )
             elif role == "assistant" and tool_calls and isinstance(tool_calls, list):

tests/test_litellm/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py

@@ -802,15 +802,15 @@ def test_text_plus_tool_calls_sequence():
 # =============================================================================
 
 
-def test_tool_message_output_is_string_not_list():
+def test_tool_message_output_uses_input_text_not_output_text():
     """
-    Test that tool message content is converted to a string, not a list.
+    Test that tool message content uses input_text type, not output_text.
 
     This is a regression test for a bug where tool results were transformed to:
         {"type": "function_call_output", "output": [{"type": "output_text", "text": "..."}]}
 
-    But the Responses API expects:
-        {"type": "function_call_output", "output": "..."}
+    But the Responses API expects input_text for tool results:
+        {"type": "function_call_output", "output": [{"type": "input_text", "text": "..."}]}
 
     The incorrect format caused OpenAI to reject with:
         "Invalid value: 'output_text'. Supported values are: 'input_text', 'input_image', and 'input_file'."
@@ -856,12 +856,14 @@ def test_tool_message_output_is_string_not_list():
     assert function_call_output is not None, "function_call_output not found"
     assert function_call_output["call_id"] == "call_abc123"
 
-    # The output should be a string, NOT a list
+    # The output should be a list with input_text type
     output = function_call_output["output"]
-    assert isinstance(output, str), f"output should be a string, got {type(output)}"
-    assert output == '{"temperature": 15, "condition": "sunny"}'
+    assert isinstance(output, list), f"output should be a list, got {type(output)}"
+    assert len(output) == 1
+    assert output[0]["type"] == "input_text", f"Expected input_text, got {output[0].get('type')}"
+    assert output[0]["text"] == '{"temperature": 15, "condition": "sunny"}'
 
-    print("✓ Tool message output is correctly a string, not a list")
+    print("✓ Tool message output correctly uses input_text type")
 
 
 def test_multiple_tool_calls_in_single_choice():