[FEATURE] add a custom prompt agent

**Describe the feature you'd like**
I hope to add a custom prompt agent, and even customize the output_ parser, because in non openai llm scenarios, many models do not have good support for langchain's built-in prompt

**Additional context**
Alternatively, have existing agents support customizing prompts according to certain specifications
