fix: handle OpenAI model responses with tool calls and no other assistant content #1562

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open

clareliguori wants to merge 4 commits into strands-agents:main from clareliguori:openai-no-asst-content

src/strands/models/litellm.py

            
                      Original file line number
                      Diff line number
                      Diff line change
                  
    @@ -194,7 +194,7 @@ def format_request_messages(
  
            formatted_messages = cls._format_system_messages(system_prompt, system_prompt_content=system_prompt_content)

            formatted_messages.extend(cls._format_regular_messages(messages))

            return [message for message in formatted_messages if message["content"] or "tool_calls" in message]

            return [message for message in formatted_messages if "content" in message or "tool_calls" in message]

        @override

        def format_chunk(self, event: dict[str, Any], **kwargs: Any) -> StreamEvent:

src/strands/models/openai.py

            
                      Original file line number
                      Diff line number
                      Diff line change
                  
    @@ -360,7 +360,7 @@ def _format_regular_messages(cls, messages: Messages, **kwargs: Any) -> list[dic
  
                formatted_message = {

                    "role": message["role"],

                    "content": formatted_contents,

                    **({"content": formatted_contents} if formatted_contents else {}),

                    **({"tool_calls": formatted_tool_calls} if formatted_tool_calls else {}),

                }

                formatted_messages.append(formatted_message)

    @@ -398,7 +398,7 @@ def format_request_messages(
  
            formatted_messages = cls._format_system_messages(system_prompt, system_prompt_content=system_prompt_content)

            formatted_messages.extend(cls._format_regular_messages(messages))

            return [message for message in formatted_messages if message["content"] or "tool_calls" in message]

            return [message for message in formatted_messages if "content" in message or "tool_calls" in message]

        def format_request(

            self,

tests/strands/models/test_litellm.py

-Original file line number
+Diff line change
@@ Expand Up / @@ -812,3 +812,58 @@ def __init__(self, usage): @@
         assert metadata_events[0]["metadata"]["usage"]["inputTokens"] == 10
         assert metadata_events[0]["metadata"]["usage"]["outputTokens"] == 5
         assert metadata_events[0]["metadata"]["usage"]["totalTokens"] == 15
+    def test_format_request_messages_with_tool_calls_no_content():
+        """Test that messages with tool calls but no content are properly formatted."""
+        messages = [
+            {"role": "user", "content": [{"text": "Use the calculator"}]},
+            {
+                "role": "assistant",
+                "content": [
+                    {
+                        "toolUse": {
+                            "input": {"expression": "2+2"},
+                            "name": "calculator",
+                            "toolUseId": "c1",
+                        },
+                    },
+                ],
+            },
+        ]
+        result = LiteLLMModel.format_request_messages(messages)
+        # Assistant message should have tool_calls but no content field
+        assert len(result) == 2
+        assert result[1]["role"] == "assistant"
+        assert "tool_calls" in result[1]
+        assert "content" not in result[1]
+        assert result[1]["tool_calls"][0]["id"] == "c1"
+    def test_format_request_messages_filters_tool_only_messages():
+        """Test that messages with only tool calls (no content) are included in output."""
+        messages = [
+            {"role": "user", "content": [{"text": "test"}]},
+            {
+                "role": "assistant",
+                "content": [
+                    {
+                        "toolUse": {
+                            "input": {},
+                            "name": "tool1",
+                            "toolUseId": "t1",
+                        },
+                    },
+                ],
+            },
+        ]
+        result = LiteLLMModel.format_request_messages(messages)
+        # Both messages should be included
+        assert len(result) == 2
+        assert result[0]["role"] == "user"
+        assert result[1]["role"] == "assistant"
+        assert "tool_calls" in result[1]

tests/strands/models/test_openai.py

-Original file line number
+Diff line change
@@ Expand Up @@
         assert len(formatted_content) == 1
         assert formatted_content[0]["type"] == "text"
         assert "Location sources are not supported by OpenAI" in caplog.text
+    def test_format_request_messages_with_tool_calls_no_content():
+        """Test that messages with tool calls but no content are properly formatted."""
+        messages = [
+            {"role": "user", "content": [{"text": "Use the calculator"}]},
+            {
+                "role": "assistant",
+                "content": [
+                    {
+                        "toolUse": {
+                            "input": {"expression": "2+2"},
+                            "name": "calculator",
+                            "toolUseId": "c1",
+                        },
+                    },
+                ],
+            },
+        ]
+        result = OpenAIModel.format_request_messages(messages)
+        # Assistant message should have tool_calls but no content field
+        assert len(result) == 2
+        assert result[1]["role"] == "assistant"
+        assert "tool_calls" in result[1]
+        assert "content" not in result[1]
+        assert result[1]["tool_calls"][0]["id"] == "c1"
+    def test_format_request_messages_filters_tool_only_messages():
+        """Test that messages with only tool calls (no content) are included in output."""
+        messages = [
+            {"role": "user", "content": [{"text": "test"}]},
+            {
+                "role": "assistant",
+                "content": [
+                    {
+                        "toolUse": {
+                            "input": {},
+                            "name": "tool1",
+                            "toolUseId": "t1",
+                        },
+                    },
+                ],
+            },
+        ]
+        result = OpenAIModel.format_request_messages(messages)
+        # Both messages should be included
+        assert len(result) == 2
+        assert result[0]["role"] == "user"
+        assert result[1]["role"] == "assistant"
+        assert "tool_calls" in result[1]

tests_integ/models/test_model_openai.py

            
                      Original file line number
                      Diff line number
                      Diff line change
                  
    @@ -1,5 +1,4 @@
  
    import os

    import unittest.mock

    import pydantic

    import pytest

    @@ -205,21 +204,22 @@ def test_rate_limit_throttling_integration_no_retries(model):
  
        the model properly raises a ModelThrottledException. We disable retries

        to avoid waiting for the exponential backoff during testing.

        """

        # Patch the event loop constants to disable retries for this test

        with unittest.mock.patch("strands.event_loop.event_loop.MAX_ATTEMPTS", 1):

            agent = Agent(model=model)

        from strands.event_loop._retry import ModelRetryStrategy

            # Create a message that's very long to trigger token-per-minute rate limits

            # This should be large enough to exceed TPM limits immediately

            very_long_text = "Really long text " * 20000

        # Create agent with retry strategy that doesn't retry

        agent = Agent(model=model, retry_strategy=ModelRetryStrategy(max_attempts=1))

            # This should raise ModelThrottledException without retries

            with pytest.raises(ModelThrottledException) as exc_info:

                agent(very_long_text)

        # Create a message that's very long to trigger token-per-minute rate limits

        # This should be large enough to exceed TPM limits immediately

        very_long_text = "Really long text " * 20000

            # Verify it's a rate limit error

            error_message = str(exc_info.value).lower()

            assert "rate limit" in error_message or "tokens per min" in error_message

        # This should raise ModelThrottledException without retries

        with pytest.raises(ModelThrottledException) as exc_info:

            agent(very_long_text)

        # Verify it's a rate limit error

        error_message = str(exc_info.value).lower()

        assert "rate limit" in error_message or "tokens per min" in error_message

    def test_content_blocks_handling(model):

tests_integ/test_multiagent_swarm.py

-Original file line number
+Diff line change
@@ Expand Up / @@ -113,6 +113,7 @@ def capture_first_node(self, event): @@
         return VerifyHook()
+    @pytest.mark.timeout(120)
     def test_swarm_execution_with_string(researcher_agent, analyst_agent, writer_agent, hook_provider):
         """Test swarm execution with string input."""
         # Create the swarm
@@ Expand Down @@

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix: handle OpenAI model responses with tool calls and no other assistant content #1562

Uh oh!

Diff view

Diff view

There are no files selected for viewing

Uh oh!

fix: handle OpenAI model responses with tool calls and no other assistant content #1562

Are you sure you want to change the base?

Uh oh!

fix: handle OpenAI model responses with tool calls and no other assistant content #1562

Uh oh!

Uh oh!

Diff view

Diff view

There are no files selected for viewing

Uh oh!