Refactor default system prompt to be closer to LLM client creation

generate_conversations/runner.py

@@ -114,9 +114,7 @@ async def run_single_conversation(
         agent = LLMFactory.create_llm(
             model_name=self.agent_model_config["model"],
             name=self.agent_model_config.get("name", "Provider"),
-            system_prompt=self.agent_model_config.get(
-                "system_prompt", "You are a helpful AI assistant."
-            ),
+            system_prompt=self.agent_model_config.get("system_prompt"),
             role=Role.PROVIDER,
             **agent_kwargs,
         )

llm_clients/llm_interface.py

@@ -9,6 +9,9 @@
 
 T = TypeVar("T", bound=BaseModel)
 
+# Default system prompt when none is provided (e.g. for provider/agent LLMs)
+DEFAULT_SYSTEM_PROMPT = "You are a helpful AI assistant."
+
 
 class Role(Enum):
     """Role of the LLM in a conversation."""
@@ -44,7 +47,9 @@ def __init__(
     ):
         self.name = name
         self.role = role
-        self.system_prompt = system_prompt or ""
+        self.system_prompt = (
+            system_prompt if system_prompt is not None else DEFAULT_SYSTEM_PROMPT
+        )
         self.first_message = first_message
         self.start_prompt = (
             start_prompt if start_prompt is not None else DEFAULT_START_PROMPT

tests/integration/test_conversation_runner.py

@@ -16,7 +16,7 @@
 import pytest
 
 from generate_conversations.runner import ConversationRunner
-from llm_clients.llm_interface import Role
+from llm_clients.llm_interface import DEFAULT_SYSTEM_PROMPT, Role
 from tests.mocks.mock_llm import MockLLM
 
 
@@ -236,18 +236,22 @@ async def test_agent_system_prompt_default(
         basic_persona_config: Dict[str, Any],
         mock_llm_factory,
     ) -> None:
-        """When agent config has no system_prompt, create_llm gets the fallback."""
-        default_prompt = "You are a helpful AI assistant."
+        """When agent config has no system_prompt, runner passes None to create_llm;
+        LLMInterface (MockLLM via super().__init__) applies DEFAULT_SYSTEM_PROMPT."""
         agent_config = {
             "model": "mock-agent",
             "name": "test-agent",
         }
-        create_llm_calls = []
+        create_llm_calls: list[dict] = []
+        created_agents: list[MockLLM] = []
         real_create = mock_llm_factory.side_effect
 
         def recording_create_llm(*args: Any, **kwargs: Any) -> MockLLM:
             create_llm_calls.append(copy.deepcopy(kwargs))
-            return real_create(*args, **kwargs)
+            llm = real_create(*args, **kwargs)
+            if kwargs.get("role") == Role.PROVIDER:
+                created_agents.append(llm)
+            return llm
 
         mock_llm_factory.side_effect = recording_create_llm
 
@@ -277,7 +281,11 @@ def recording_create_llm(*args: Any, **kwargs: Any) -> MockLLM:
 
         agent_calls = [c for c in create_llm_calls if c.get("role") == Role.PROVIDER]
         assert len(agent_calls) == 1
-        assert agent_calls[0]["system_prompt"] == default_prompt
+        # Runner passes None when config has no system_prompt
+        assert agent_calls[0]["system_prompt"] is None
+        # LLMInterface applies default; created agent should have DEFAULT_SYSTEM_PROMPT
+        assert len(created_agents) == 1
+        assert created_agents[0].system_prompt == DEFAULT_SYSTEM_PROMPT
 
 
 @pytest.mark.integration

tests/unit/llm_clients/test_azure_llm.py

@@ -8,6 +8,7 @@
 
 from llm_clients import Role
 from llm_clients.azure_llm import AzureLLM
+from llm_clients.llm_interface import DEFAULT_SYSTEM_PROMPT
 
 from .test_base_llm import TestJudgeLLMBase
 from .test_helpers import (
@@ -343,7 +344,9 @@ async def test_generate_response_without_system_prompt(
         mock_llm.ainvoke = AsyncMock(return_value=mock_response)
         mock_azure_model.return_value = mock_llm
 
-        llm = AzureLLM(name="TestAzure", role=Role.PERSONA)  # No system prompt
+        llm = AzureLLM(
+            name="TestAzure", role=Role.PERSONA, system_prompt=""
+        )  # No system prompt
         response = await llm.generate_response(conversation_history=mock_system_message)
 
         assert response == "Response without system prompt"
@@ -484,9 +487,8 @@ def test_set_system_prompt(self):
             role=Role.PERSONA,
             model_name="azure-gpt-5.2",
             name="TestAzure",
-            system_prompt="Initial prompt",
         )
-        assert llm.system_prompt == "Initial prompt"
+        assert llm.system_prompt == DEFAULT_SYSTEM_PROMPT
 
         llm.set_system_prompt("Updated prompt")
         assert llm.system_prompt == "Updated prompt"

tests/unit/llm_clients/test_base_llm.py

@@ -33,7 +33,7 @@ def get_mock_patches(self):
 from pydantic import BaseModel, Field
 
 from llm_clients import Role
-from llm_clients.llm_interface import JudgeLLM, LLMInterface
+from llm_clients.llm_interface import DEFAULT_SYSTEM_PROMPT, JudgeLLM, LLMInterface
 
 from .test_helpers import (
     assert_error_metadata,
@@ -125,11 +125,9 @@ def test_init_with_role_and_system_prompt(self):
     def test_set_system_prompt(self):
         """Test setting and updating system prompt."""
         with self.get_mock_patches():  # pyright: ignore[reportGeneralTypeIssues]
-            llm = self.create_llm(
-                role=Role.PERSONA, name="TestLLM", system_prompt="Initial prompt"
-            )
+            llm = self.create_llm(role=Role.PERSONA, name="TestLLM")
 
-            assert llm.system_prompt == "Initial prompt"
+            assert llm.system_prompt == DEFAULT_SYSTEM_PROMPT
 
             llm.set_system_prompt("Updated prompt")
             assert llm.system_prompt == "Updated prompt"

tests/unit/llm_clients/test_claude_llm.py

@@ -7,6 +7,7 @@
 
 from llm_clients import Role
 from llm_clients.claude_llm import ClaudeLLM
+from llm_clients.llm_interface import DEFAULT_SYSTEM_PROMPT
 
 from .test_base_llm import TestJudgeLLMBase
 from .test_helpers import (
@@ -172,7 +173,9 @@ async def test_generate_response_without_system_prompt(
         mock_llm.ainvoke = AsyncMock(return_value=mock_response)
         mock_chat_anthropic.return_value = mock_llm
 
-        llm = ClaudeLLM(name="TestClaude", role=Role.PERSONA)  # No system prompt
+        llm = ClaudeLLM(
+            name="TestClaude", role=Role.PERSONA, system_prompt=""
+        )  # No system prompt
         response = await llm.generate_response(conversation_history=mock_system_message)
 
         assert response == "Response without system prompt"
@@ -285,10 +288,8 @@ def test_last_response_metadata_copy_returns_copy(self):
 
     def test_set_system_prompt(self):
         """Test set_system_prompt method."""
-        llm = ClaudeLLM(
-            name="TestClaude", role=Role.PERSONA, system_prompt="Initial prompt"
-        )
-        assert llm.system_prompt == "Initial prompt"
+        llm = ClaudeLLM(name="TestClaude", role=Role.PERSONA)
+        assert llm.system_prompt == DEFAULT_SYSTEM_PROMPT
 
         llm.set_system_prompt("Updated prompt")
         assert llm.system_prompt == "Updated prompt"

tests/unit/llm_clients/test_endpoint_llm.py

@@ -7,7 +7,7 @@
 
 from llm_clients import Role
 from llm_clients.endpoint_llm import EndpointLLM
-from llm_clients.llm_interface import DEFAULT_START_PROMPT
+from llm_clients.llm_interface import DEFAULT_START_PROMPT, DEFAULT_SYSTEM_PROMPT
 
 from .test_base_llm import TestLLMBase
 from .test_helpers import (
@@ -230,10 +230,8 @@ async def test_generate_response_none_history_delegates_to_start_conversation(
 
     def test_set_system_prompt(self):
         with self.get_mock_patches():
-            llm = self.create_llm(
-                role=Role.PROVIDER, name="TestLLM", system_prompt="Initial"
-            )
-        assert llm.system_prompt == "Initial"
+            llm = self.create_llm(role=Role.PROVIDER, name="TestLLM")
+        assert llm.system_prompt == DEFAULT_SYSTEM_PROMPT
         llm.set_system_prompt("Updated")
         assert llm.system_prompt == "Updated"
 

tests/unit/llm_clients/test_gemini_llm.py

@@ -7,6 +7,7 @@
 
 from llm_clients import Role
 from llm_clients.gemini_llm import GeminiLLM
+from llm_clients.llm_interface import DEFAULT_SYSTEM_PROMPT
 
 from .test_base_llm import TestJudgeLLMBase
 from .test_helpers import (
@@ -169,7 +170,9 @@ async def test_generate_response_without_system_prompt(
         mock_llm.ainvoke = AsyncMock(return_value=mock_response)
         mock_chat_gemini.return_value = mock_llm
 
-        llm = GeminiLLM(name="TestGemini", role=Role.PERSONA)  # No system prompt
+        llm = GeminiLLM(
+            name="TestGemini", role=Role.PERSONA, system_prompt=""
+        )  # No system prompt
         response = await llm.generate_response(conversation_history=mock_system_message)
 
         assert response == "Response without system prompt"
@@ -310,10 +313,8 @@ def test_last_response_metadata_copy_returns_copy(self):
 
     def test_set_system_prompt(self):
         """Test set_system_prompt method."""
-        llm = GeminiLLM(
-            name="TestGemini", role=Role.PERSONA, system_prompt="Initial prompt"
-        )
-        assert llm.system_prompt == "Initial prompt"
+        llm = GeminiLLM(name="TestGemini", role=Role.PERSONA)
+        assert llm.system_prompt == DEFAULT_SYSTEM_PROMPT
 
         llm.set_system_prompt("Updated prompt")
         assert llm.system_prompt == "Updated prompt"

tests/unit/llm_clients/test_llm_interface.py

@@ -5,7 +5,7 @@
 import pytest
 
 from llm_clients import Role
-from llm_clients.llm_interface import LLMInterface
+from llm_clients.llm_interface import DEFAULT_SYSTEM_PROMPT, LLMInterface
 
 
 class ConcreteLLM(LLMInterface):
@@ -42,11 +42,11 @@ class TestLLMInterface:
     """Unit tests for LLMInterface abstract base class."""
 
     def test_init_with_name_only(self):
-        """Test initialization with only name parameter."""
+        """Test initialization with only name parameter uses DEFAULT_SYSTEM_PROMPT."""
         llm = ConcreteLLM(name="TestLLM", role=Role.PROVIDER)
 
         assert llm.name == "TestLLM"
-        assert llm.system_prompt == ""
+        assert llm.system_prompt == DEFAULT_SYSTEM_PROMPT
 
     def test_init_with_name_and_system_prompt(self):
         """Test initialization with name and system prompt."""
@@ -181,12 +181,17 @@ def test_getattr_with_callable_attribute(self):
         assert result == "method result"
         llm.llm.custom_method.assert_called_once()
 
-    def test_system_prompt_default_empty_string(self):
-        """Test that system_prompt defaults to empty string, not None."""
+    def test_system_prompt_default_when_none(self):
+        """When system_prompt is None, base class uses DEFAULT_SYSTEM_PROMPT."""
         llm = ConcreteLLM(name="TestLLM", role=Role.PROVIDER)
-        assert llm.system_prompt == ""
+        assert llm.system_prompt == DEFAULT_SYSTEM_PROMPT
         assert llm.system_prompt is not None
 
+    def test_system_prompt_explicit_empty_string(self):
+        """Explicit empty string is preserved, not replaced by default."""
+        llm = ConcreteLLM(name="TestLLM", role=Role.PROVIDER, system_prompt="")
+        assert llm.system_prompt == ""
+
     def test_getattr_preserves_attribute_type(self):
         """Test that __getattr__ preserves the type of delegated attributes."""
 

tests/unit/llm_clients/test_ollama_llm.py

@@ -5,7 +5,7 @@
 
 import pytest
 
-from llm_clients.llm_interface import Role
+from llm_clients.llm_interface import DEFAULT_SYSTEM_PROMPT, Role
 
 from .test_base_llm import TestLLMBase
 from .test_helpers import (
@@ -215,7 +215,7 @@ async def test_generate_response_without_system_prompt(
         mock_instance.ainvoke = AsyncMock(return_value="This is a test response")
         mock_ollama.return_value = mock_instance
 
-        llm = OllamaLLM(name="test-ollama", role=Role.PROVIDER)
+        llm = OllamaLLM(name="test-ollama", role=Role.PROVIDER, system_prompt="")
         response = await llm.generate_response(conversation_history=mock_system_message)
 
         # Verify message uses Human/Assistant format even without system prompt
@@ -362,7 +362,7 @@ async def test_generate_response_with_none_message(
         mock_instance.ainvoke = AsyncMock(return_value="Default response")
         mock_ollama.return_value = mock_instance
 
-        llm = OllamaLLM(name="test-ollama", role=Role.PROVIDER)
+        llm = OllamaLLM(name="test-ollama", role=Role.PROVIDER, system_prompt="")
         response = await llm.generate_response(None)
 
         # None history: format uses default start_prompt message
@@ -411,20 +411,18 @@ async def test_generate_response_preserves_multiline_messages(self, mock_ollama)
         assert "Line 1\nLine 2\nLine 3" in call_args
 
     @patch("llm_clients.ollama_llm.LangChainOllamaLLM")
-    def test_set_system_prompt_updates_prompt(self, mock_ollama):
+    def test_set_system_prompt(self, mock_ollama):
         """Test that set_system_prompt updates the system_prompt attribute."""
         from llm_clients.ollama_llm import OllamaLLM
 
         llm = OllamaLLM(name="test-ollama", role=Role.PROVIDER)
 
-        # Initially empty string (from LLMInterface base class)
-        assert llm.system_prompt == ""
+        # Initially default prompt (from LLMInterface when system_prompt is None)
+        assert llm.system_prompt == DEFAULT_SYSTEM_PROMPT
 
-        # Set prompt
         llm.set_system_prompt("You are a math tutor")
         assert llm.system_prompt == "You are a math tutor"
 
-        # Update prompt
         llm.set_system_prompt("You are a science tutor")
         assert llm.system_prompt == "You are a science tutor"
 
@@ -438,7 +436,7 @@ async def test_set_system_prompt_affects_subsequent_calls(self, mock_ollama):
         mock_instance.ainvoke = AsyncMock(return_value="Response")
         mock_ollama.return_value = mock_instance
 
-        llm = OllamaLLM(name="test-ollama", role=Role.PROVIDER)
+        llm = OllamaLLM(name="test-ollama", role=Role.PROVIDER, system_prompt="")
 
         # First call without system prompt
         await llm.generate_response(
@@ -517,7 +515,7 @@ async def test_generate_response_with_empty_conversation_history(
         mock_instance.ainvoke = AsyncMock(return_value="Response")
         mock_ollama.return_value = mock_instance
 
-        llm = OllamaLLM(name="test-ollama", role=Role.PROVIDER)
+        llm = OllamaLLM(name="test-ollama", role=Role.PROVIDER, system_prompt="")
 
         response = await llm.generate_response(conversation_history=mock_system_message)
 
@@ -539,7 +537,7 @@ async def test_generate_response_with_none_conversation_history(
         mock_instance.ainvoke = AsyncMock(return_value="Response")
         mock_ollama.return_value = mock_instance
 
-        llm = OllamaLLM(name="test-ollama", role=Role.PROVIDER)
+        llm = OllamaLLM(name="test-ollama", role=Role.PROVIDER, system_prompt="")
 
         response = await llm.generate_response(conversation_history=mock_system_message)
 

tests/unit/llm_clients/test_openai_llm.py

@@ -6,6 +6,7 @@
 import pytest
 
 from llm_clients import Role
+from llm_clients.llm_interface import DEFAULT_SYSTEM_PROMPT
 from llm_clients.openai_llm import OpenAILLM
 
 from .test_base_llm import TestJudgeLLMBase
@@ -179,7 +180,9 @@ async def test_generate_response_without_system_prompt(
         mock_llm.ainvoke = AsyncMock(return_value=mock_response)
         mock_chat_openai.return_value = mock_llm
 
-        llm = OpenAILLM(name="TestOpenAI", role=Role.PERSONA)  # No system prompt
+        llm = OpenAILLM(
+            name="TestOpenAI", role=Role.PERSONA, system_prompt=""
+        )  # No system prompt
         response = await llm.generate_response(conversation_history=mock_system_message)
 
         assert response == "Response without system prompt"
@@ -321,10 +324,8 @@ def test_last_response_metadata_copy_returns_copy(self):
 
     def test_set_system_prompt(self):
         """Test set_system_prompt method."""
-        llm = OpenAILLM(
-            name="TestOpenAI", role=Role.PERSONA, system_prompt="Initial prompt"
-        )
-        assert llm.system_prompt == "Initial prompt"
+        llm = OpenAILLM(name="TestOpenAI", role=Role.PERSONA)
+        assert llm.system_prompt == DEFAULT_SYSTEM_PROMPT
 
         llm.set_system_prompt("Updated prompt")
         assert llm.system_prompt == "Updated prompt"