added tests back in

Author: cemde

maseval/interface/agents/langgraph.py

@@ -238,7 +238,6 @@ def gather_config(self) -> dict[str, Any]:
         return base_config
 
     def _run_agent(self, query: str) -> Any:
-
         _check_langgraph_installed()
         from langchain_core.messages import HumanMessage
 

tests/test_contract/test_agent_adapter_contract.py

@@ -444,6 +444,26 @@ def test_adapter_message_history_after_clear_and_run(self, framework):
         agent = create_agent_for_framework(framework, mock_llm)
         adapter = create_adapter_for_framework(framework, agent)
 
+        # First run
+        adapter.run("First query")
+        history_1 = adapter.get_messages()
+        assert len(history_1) > 0
+
+        # Clear and verify empty (or just system message for smolagents)
+        adapter.clear_message_history()
+        history_after_clear = adapter.get_messages()
+        expected_after_clear = 1 if framework == "smolagents" else 0  # smolagents keeps system message
+        assert len(history_after_clear) == expected_after_clear
+
+        # Second run should populate new history
+        adapter.run("Second query")
+        history_2 = adapter.get_messages()
+        assert len(history_2) > expected_after_clear  # Should have more than just system message
+
+        # History should only contain second run's messages
+        # (exact count depends on framework, but should have at least one message)
+        assert any("Second query" in str(msg.get("content", "")) for msg in history_2)
+
     def test_adapter_logs_populated_after_run(self, framework):
         """Test all adapters populate self.logs during execution.
 

tests/test_interface/test_agent_integration/test_langgraph_integration.py

@@ -165,7 +165,7 @@ def test_langgraph_adapter_logs_after_run():
     from maseval.interface.agents.langgraph import LangGraphAgentAdapter
     from langgraph.graph import StateGraph, END
     from typing_extensions import TypedDict
-    from langchain_core.messages import AIMessage, HumanMessage
+    from langchain_core.messages import AIMessage
     from langchain_core.messages.ai import UsageMetadata
     import time
 
@@ -199,7 +199,7 @@ def agent_node(state: State) -> State:
     time_before = time.time()
 
     # Run the agent
-    result = adapter.run("Test query")
+    adapter.run("Test query")
 
     # Capture time after run
     time_after = time.time()