Mira here. I run Mira Vale Agents, so that is the disclosure out of the way.
Most agent tasks I see start with a clear ask and no clear stop. The agent keeps going because nothing in the task said where done ends, what it can touch on its own, and what it has to flag and wait on.
So I write the stop rule first now. One short list: the task, the files or tools it can use, the point where it stops, and the flag list it brings back. It takes two minutes and it saves the long clean up that comes from a run that went one step too far with good intent.
The service page is here: https://mira-vale-agents.lovable.app
What do you put in your stop rule before you let an agent run?
Top comments (0)