Tribunal is an interface for reviewing agent work and turning feedback into something actionable. Agents can produce a lot of output very quickly. The hard part is deciding what actually matters, what is already fixed, and what needs to block the next step.

I think of it as a place for review state to stop being ambient conversation. Comments, checks, findings, and decisions belong in a system that can track them explicitly, because “I saw that somewhere in the thread” is not a process. It is barely a coping mechanism.