Fix tool call validation error handling for Groq LLM provider (#10927)

Co-authored-by: openhands <openhands@all-hands.dev>
Co-authored-by: Engel Nyst <enyst@users.noreply.github.com>

openhands/controller/agent_controller.py

@@ -944,6 +944,23 @@ class AgentController:
                         return
                     else:
                         raise LLMContextWindowExceedError()
+                # Check if this is a tool call validation error that should be recoverable
+                elif (
+                    isinstance(e, BadRequestError)
+                    and 'tool call validation failed' in error_str
+                    and (
+                        'missing properties' in error_str
+                        or 'missing required' in error_str
+                    )
+                ):
+                    # Handle tool call validation errors from Groq as recoverable errors
+                    self.event_stream.add_event(
+                        ErrorObservation(
+                            content=f'Tool call validation failed: {str(e)}. Please check the tool parameters and try again.',
+                        ),
+                        EventSource.AGENT,
+                    )
+                    return
                 else:
                     raise e
 

tests/unit/controller/test_agent_controller.py

@@ -299,6 +299,55 @@ async def test_react_to_content_policy_violation(
     await controller.close()
 
 
+@pytest.mark.asyncio
+async def test_tool_call_validation_error_handling(
+    mock_agent_with_stats,
+    test_event_stream,
+):
+    """Test that tool call validation errors from Groq are handled as recoverable errors."""
+    mock_agent, conversation_stats, llm_registry = mock_agent_with_stats
+
+    controller = AgentController(
+        agent=mock_agent,
+        event_stream=test_event_stream,
+        conversation_stats=conversation_stats,
+        iteration_delta=10,
+        sid='test',
+        confirmation_mode=False,
+        headless_mode=True,
+    )
+
+    controller.state.agent_state = AgentState.RUNNING
+
+    # Mock the agent.step method to raise a BadRequestError with tool validation failure
+    def mock_step(state):
+        raise BadRequestError(
+            message='litellm.BadRequestError: GroqException - {"error":{"message":"tool call validation failed: parameters for tool str_replace_editor did not match schema: errors: [missing properties: \'path\']","type":"invalid_request_error","code":"tool_use_failed"}}',
+            model='groq/llama3-8b-8192',
+            llm_provider='groq',
+        )
+
+    mock_agent.step = mock_step
+
+    # Call _step which should handle the tool validation error
+    await controller._step()
+
+    # Verify that the agent state is still RUNNING (not ERROR)
+    assert controller.state.agent_state == AgentState.RUNNING
+
+    # Verify that an ErrorObservation was added to the event stream
+    events = list(test_event_stream.get_events())
+    error_observations = [e for e in events if isinstance(e, ErrorObservation)]
+    assert len(error_observations) == 1
+
+    error_obs = error_observations[0]
+    assert 'tool call validation failed' in error_obs.content
+    assert 'missing properties' in error_obs.content
+    assert 'path' in error_obs.content
+
+    await controller.close()
+
+
 @pytest.mark.asyncio
 async def test_run_controller_with_fatal_error(
     test_event_stream, mock_memory, mock_agent_with_stats