Draught - A provider-agnostic coding-agent CLI and runtime

[webofbits]

My only concern is whether buffering the Markdown could make streamed LLM output feel slower. How does Yoke handle that?