Free tool
How do you write an operating brief for an AI agent?
このツールは英語のみで公開しています。
Give the agent a job title, not a prompt, then state five things: what it owns, what good work looks like, which sources it may use, what it must never do without approval, and where it delivers. Fill in the fields below and copy the brief into your agent's Training tab.
Your brief
You are the Add a job titleOwn: Add what it ownsGood work: Add what good work looks likeSources: Add the sources it may useStop: Add what it must never do without approvalOutput: Add where it deliversWhat makes a good brief?
Clear limits. One agent should own one domain, so if the "owns" field needs an "and", you probably need two agents.
What should an agent never do without approval?
The test is reversibility, not task size. An agent can draft, summarize, research, classify, tag, prepare and reconcile on its own. It should park for approval before it sends an external message, publishes, spends money, accepts terms, changes production data or deletes material data.
What happens after I paste it in?
Give the agent one small errand with a result you can check quickly, and review the first output before you add more work. Every saved instruction has a version, so you can restore an earlier brief.