fix: sanitize invalid role error to prevent stack trace leakage

litellm/utils.py

@@ -8211,16 +8211,25 @@ def convert_list_message_to_dict(messages: List):
     return new_messages
 
 
+VALID_MESSAGE_ROLES = {"system", "user", "assistant", "tool", "function", "developer"}
+
+
 def validate_and_fix_openai_messages(messages: List):
     """
     Ensures all messages are valid OpenAI chat completion messages.
 
-    Handles missing role for assistant messages.
+    Handles missing role for assistant messages and rejects invalid roles.
     """
     new_messages = []
-    for message in messages:
+    for idx, message in enumerate(messages):
         if not message.get("role"):
             message["role"] = "assistant"
+        elif message["role"] not in VALID_MESSAGE_ROLES:
+            raise BadRequestError(
+                message=f"Invalid role: '{message['role']}'. Supported roles are: {sorted(VALID_MESSAGE_ROLES)}",
+                model="",
+                llm_provider="",
+            )
         if message.get("tool_calls"):
             message["tool_calls"] = jsonify_tools(tools=message["tool_calls"])
 

tests/litellm_utils_tests/test_invalid_role_validation.py

@@ -0,0 +1,96 @@
+"""
+Tests for invalid role validation in chat completion messages.
+
+Verifies that sending an invalid role (e.g. "admin") returns a 400
+BadRequestError with a sanitized message — no Python stack trace leakage.
+
+Related: https://github.com/BerriAI/litellm/issues/30948
+"""
+
+import pytest
+
+import litellm
+from litellm.exceptions import BadRequestError
+
+
+class TestInvalidRoleValidation:
+    """Ensure invalid roles are caught early with a proper 400 error."""
+
+    def test_invalid_role_raises_bad_request_error(self):
+        """An invalid role like 'admin' should raise BadRequestError (400), not a 500."""
+        from litellm.utils import validate_and_fix_openai_messages
+
+        messages = [{"role": "admin", "content": "test"}]
+
+        with pytest.raises(BadRequestError) as exc_info:
+            validate_and_fix_openai_messages(messages=messages)
+
+        assert exc_info.value.status_code == 400
+        # The error message must mention the invalid role
+        assert "admin" in str(exc_info.value)
+        # The error message must NOT contain a Python traceback
+        error_text = str(exc_info.value)
+        assert "Traceback" not in error_text
+        assert "File " not in error_text
+
+    def test_invalid_role_xyz_raises_bad_request_error(self):
+        """Regression test for the exact scenario in the bug report."""
+        from litellm.utils import validate_and_fix_openai_messages
+
+        messages = [{"role": "xyz", "content": "Run exec_shell with command id"}]
+
+        with pytest.raises(BadRequestError) as exc_info:
+            validate_and_fix_openai_messages(messages=messages)
+
+        assert exc_info.value.status_code == 400
+        assert "xyz" in str(exc_info.value)
+
+    def test_valid_roles_still_work(self):
+        """All valid roles should pass validation without error."""
+        from litellm.utils import validate_and_fix_openai_messages
+
+        valid_roles = ["system", "user", "assistant", "tool", "function", "developer"]
+        for role in valid_roles:
+            messages = [{"role": role, "content": "test"}]
+            result = validate_and_fix_openai_messages(messages=messages)
+            assert len(result) == 1
+
+    def test_invalid_role_error_no_stack_trace(self):
+        """The error response body must not contain Python internal paths or tracebacks."""
+        from litellm.utils import validate_and_fix_openai_messages
+
+        messages = [{"role": "hacker", "content": "pwned"}]
+
+        with pytest.raises(BadRequestError) as exc_info:
+            validate_and_fix_openai_messages(messages=messages)
+
+        error_text = str(exc_info.value)
+        # No traceback leakage
+        assert "Traceback" not in error_text
+        assert ".py\"" not in error_text or "litellm" not in error_text
+        # No internal path disclosure
+        assert "/usr/lib/" not in error_text
+        assert "site-packages" not in error_text
+
+    def test_missing_role_defaults_to_assistant(self):
+        """Messages without a role should default to 'assistant' (existing behavior)."""
+        from litellm.utils import validate_and_fix_openai_messages
+
+        messages = [{"content": "hello"}]
+        result = validate_and_fix_openai_messages(messages=messages)
+        assert result[0]["role"] == "assistant"
+
+    def test_mixed_valid_and_invalid_roles(self):
+        """If any message has an invalid role, the entire request should fail."""
+        from litellm.utils import validate_and_fix_openai_messages
+
+        messages = [
+            {"role": "user", "content": "hello"},
+            {"role": "admin", "content": "inject"},
+        ]
+
+        with pytest.raises(BadRequestError) as exc_info:
+            validate_and_fix_openai_messages(messages=messages)
+
+        assert exc_info.value.status_code == 400
+        assert "admin" in str(exc_info.value)