It has particularly good observability (the ability to see the full content of every prompt and response and tool call) compared to other harnesses, that's the main thing that stood out to me.
Weird. I don't use the official harnesses by OpenAI or Anthropic but with all others i tried there wasn't a single one that hides any of that. What harnesses did you compare to?
Have you tried pressing ctrl+o?
I think what he meant was that it gives you good overall observabiity, over your entire context, kind of how langfuse or these platforms do, but it does so locally. A good philosophy i picked up is how they treat the transcript as the backbone of the product.