1. To cover a broader range of action types, OpenAIFunctionEval could be expanded to handle more types of OpenAI functions or could even be generalized to accommodate multiple action types. This would likely necessitate restructuring the extract_actions and _filter_actions methods. Additional logic would also be necessary to distinguish between different types of actions.

2. Currently, the OpenAIFunctionEval class may not extract actions properly from other types of messages. It’s specifically designed to handle OpenAI function call actions. To use OpenAIFunctionEval for other types of messages, it might be necessary to add more sophisticated logic to properly parse these messages, or even build new classes to handle different message formats.

3. Currently, OpenAIFunctionEval handles both action extraction and filtering. These concerns could be separated by creating dedicated classes for each task. For example, an ActionExtractor class could handle the extraction of actions while an ActionFilter class could handle the filtration. This would lead to a cleaner design and more flexibility when expanding the capabilities of the system.

Note: Given that this was produced with current context, the actual class, methods and their behaviors might differ in the live environment. The automata.llm.eval.eval_providers or OpenAIFunctionEval and related classes or methods might not even exist. Please refer to official OpenAI documentation for updated and accurate information.