編碼 Agent 的任務何時才算真正完成?
When is a coding agent's task actually done?
編碼 Agent 停止輸出並不代表任務完成;作者會先檢視修改,並確認變更符合要求,再把任務標記為完成並準備合併。作者開發的工具 Styx 會在已完成任務滿七天後自動封存,以免等待中的任務清單不斷累積。
An agent stops typing and the spinner goes away. Is the task finished?
It's tempting to say yes. It isn't: the agent has only run out of things to do and is waiting for my next message. I haven't read the diff, the checks haven't run on the merged result, and nothing is on main.
So I treat 'done' as something I decide, not something the agent signals. When I'm happy with the changes, I mark the task done, and that's the point where it becomes ready to merge. Until then, an agent that went quiet is just a quiet agent.
The other side has a cost. If nothing ever closes a task for you, the 'waiting' ones pile up and you forget what's in them. In the tool I build, Styx, finished tasks archive themselves after seven days, which keeps the list from growing without end.
So what's your rule? Do you call it done when the agent stops, when the tests pass, when you've read the diff, or something else?
來源:DEV Community · dev.to