What's this about?
Chinese AI writer Baoyu (dotey) shared practical guidance for using an agent's long-running `/goal` feature: give the agent a goal and let it run until it's done. His three keys: (1) state the goal clearly, (2) define how to verify the result, and (3) define the stop conditions.
What is a "goal mode" for agents?
Goal mode is where you hand an agent an objective ("migrate this codebase," "research and draft this report") and it works autonomously — planning, acting, and checking itself — for a long stretch without you steering each step. The skill is in the setup: a vague goal or missing verification means the agent wanders or declares false victory.
Why does it matter?
As agents get more autonomous (see Auto mode above), the human's job shifts from doing the task to *specifying* it well — clear goals, checkable success, and explicit stop conditions become the core operator skill.