From 09c027692f1aeb0afba9459c75d9db311a18cfce Mon Sep 17 00:00:00 2001
From: Kian Jones <11655409+kianjones9@users.noreply.github.com>
Date: Sun, 7 Dec 2025 14:19:25 -0800
Subject: [PATCH] tests: assistant msg validation error (#6536)

* add regression test for dict content in AssistantMessage

Tests the fix for pydantic validation error when send_message tool
returns dict content like {'tofu': 1, 'mofu': 1, 'bofu': 1}.

The test verifies that dict content is properly serialized to JSON
string before creating AssistantMessage.

* improve type annotation for validate_function_response

Changed return type from Any to str | dict[str, Any] to match actual
behavior. This enables static type checkers (pyright, mypy) to catch
type mismatches like the AssistantMessage bug.

With proper type annotations, pyright would have caught:
  error: Argument of type "str | dict[str, Any]" cannot be assigned
  to parameter "content" of type "str"

This prevents future bugs where dict is passed to string-only fields.

* add regression test for dict content in AssistantMessage

Moved test into existing test_message_manager.py suite alongside other
message conversion tests.

Tests the fix for pydantic validation error when send_message tool
returns dict content like {'tofu': 1, 'mofu': 1, 'bofu': 1}.

The test verifies that dict content is properly serialized to JSON
string before creating AssistantMessage.
---
 letta/utils.py                         |  4 +-
 tests/managers/test_message_manager.py | 67 ++++++++++++++++++++++++++
 2 files changed, 70 insertions(+), 1 deletion(-)

diff --git a/letta/utils.py b/letta/utils.py
index 02c0ff23..7525ca84 100644
--- a/letta/utils.py
+++ b/letta/utils.py
@@ -854,7 +854,9 @@ def parse_json(string) -> dict:
         raise e
 
 
-def validate_function_response(function_response: Any, return_char_limit: int, strict: bool = False, truncate: bool = True) -> Any:
+def validate_function_response(
+    function_response: Any, return_char_limit: int, strict: bool = False, truncate: bool = True
+) -> str | dict[str, Any]:
     """Check to make sure that a function used by Letta returned a valid response. Truncates to return_char_limit if necessary.
 
     This makes sure that we can coerce the function_response into a string or dict that meets our criteria. We handle some soft coercion.
diff --git a/tests/managers/test_message_manager.py b/tests/managers/test_message_manager.py
index 06b4e967..b62b32a8 100644
--- a/tests/managers/test_message_manager.py
+++ b/tests/managers/test_message_manager.py
@@ -1019,3 +1019,70 @@ async def test_convert_tool_calls_only_assistant_tools(server: SyncServer, sarah
     # check assistant messages content (they appear in reverse order)
     assert letta_messages[0].content == "Second message"
     assert letta_messages[1].content == "First message"
+
+
+@pytest.mark.asyncio
+async def test_convert_assistant_message_with_dict_content(server: SyncServer, sarah_agent, default_user):
+    """Test that send_message with dict content is properly serialized to JSON string
+
+    Regression test for bug where dict content like {'tofu': 1, 'mofu': 1, 'bofu': 1}
+    caused pydantic validation error because AssistantMessage.content expects a string.
+    """
+    import json
+
+    # Test case 1: Simple dict as message content
+    tool_calls = [
+        OpenAIToolCall(
+            id="call_1",
+            type="function",
+            function=OpenAIFunction(name="send_message", arguments='{"message": {"tofu": 1, "mofu": 1, "bofu": 1}}'),
+        ),
+    ]
+
+    message = PydanticMessage(
+        agent_id=sarah_agent.id,
+        role=MessageRole.assistant,
+        content=[TextContent(text="Sending structured data...")],
+        tool_calls=tool_calls,
+    )
+
+    # convert with assistant mode - should not raise validation error
+    letta_messages = message.to_letta_messages(use_assistant_message=True)
+
+    assert len(letta_messages) == 2
+    assert letta_messages[0].message_type == "assistant_message"
+    assert letta_messages[1].message_type == "reasoning_message"
+
+    # check that dict was serialized to JSON string
+    assistant_msg = letta_messages[0]
+    assert isinstance(assistant_msg.content, str)
+
+    # verify the JSON-serialized content can be parsed back
+    parsed_content = json.loads(assistant_msg.content)
+    assert parsed_content == {"tofu": 1, "mofu": 1, "bofu": 1}
+
+    # Test case 2: Nested dict with various types
+    tool_calls_nested = [
+        OpenAIToolCall(
+            id="call_2",
+            type="function",
+            function=OpenAIFunction(
+                name="send_message",
+                arguments='{"message": {"status": "success", "data": {"count": 42, "items": ["a", "b"]}, "meta": null}}',
+            ),
+        ),
+    ]
+
+    message_nested = PydanticMessage(
+        agent_id=sarah_agent.id,
+        role=MessageRole.assistant,
+        content=[TextContent(text="Sending complex data...")],
+        tool_calls=tool_calls_nested,
+    )
+
+    letta_messages_nested = message_nested.to_letta_messages(use_assistant_message=True)
+    assistant_msg_nested = letta_messages_nested[0]
+
+    assert isinstance(assistant_msg_nested.content, str)
+    parsed_nested = json.loads(assistant_msg_nested.content)
+    assert parsed_nested == {"status": "success", "data": {"count": 42, "items": ["a", "b"]}, "meta": None}