Could return shorter props, e.g. “attrs” instead of “attributes”, etc.? Or have a “compact” mode? This becomes important for the OpenAI ChatGPT integration since we need to be efficient with bytes in the LLM context window.