the ACP separation between agent client and agent runtime is the clever design choice - it means Kit isn't locked into one editor's integration, and other harnesses can plug in without custom glue code. shipping as one static binary with the terminal client, ACP server, A2A endpoint, and subagent orchestrator all bundled together also keeps setup simple compared to wiring several separate tools together.
Dial
the ACP separation is the interesting part here - decoupling the client from the runtime so you're not locked into one editor integration. for the subagent orchestrator specifically, when a subagent's output gets merged back into the main session, is there any validation step, or does whatever the subagent produced just flow straight through as if the top-level agent wrote it itself?