feat: strict tool calling setting (#8810)

base

tests/managers/test_agent_manager.py

@@ -1851,6 +1851,7 @@ async def test_agent_state_schema_unchanged(server: SyncServer):
         "verbosity",
         "tier",
         "parallel_tool_calls",
+        "strict",
     }
     actual_llm_config_fields = set(llm_config_fields.keys())
     if actual_llm_config_fields != expected_llm_config_fields:

tests/sdk/agents_test.py

@@ -7,9 +7,12 @@ AGENTS_CREATE_PARAMS = [
         {
             # Verify model_settings is populated with config values
             # Note: The 'model' field itself is separate from model_settings
+            # strict defaults to False when no model_settings is explicitly provided
+            # (OpenAIModelSettings defaults to True only when explicitly instantiated)
             "model_settings": {
                 "max_output_tokens": 16384,
                 "parallel_tool_calls": False,
+                "strict": False,
                 "provider_type": "openai",
                 "temperature": 0.7,
                 "reasoning": {"reasoning_effort": "minimal"},
@@ -29,6 +32,7 @@ AGENTS_UPDATE_PARAMS = [
             "model_settings": {
                 "max_output_tokens": 16384,
                 "parallel_tool_calls": False,
+                "strict": False,
                 "provider_type": "openai",
                 "temperature": 0.7,
                 "reasoning": {"reasoning_effort": "minimal"},