💡 How to use these prompts:
Replace everything in [BRACKETS] with your specific details before sending.
Click Copy to copy any prompt to your clipboard instantly.
Here is the configuration, tool list and system prompt for an AI agent. List every external system it can reach, every action it can take that changes state outside this process, and every credential it can access. For each, say whether the effect is reversible. Do not describe intended behaviour, only what is reachable.
Compare the stated purpose below against the tool list and permissions above. List every capability the agent holds that its stated purpose does not require. For each, describe the worst realistic outcome if the agent used it believing it was helping.
You are reviewing this agent for scope violations. Write six scenarios in which the agent, trying to complete its assigned task and finding the sanctioned route blocked, could take an action affecting a real system or a real person outside the task. For each, name the tool it would use and the signal that would appear in logs.
For each external system this agent can reach, specify one log line or metric that would appear if the agent acted outside its sanctioned scope, and would not appear during normal operation. Prefer signals that cannot be suppressed by the agent itself. Say where each should be collected.
Write the explicit stop conditions this agent should carry: the circumstances under which it must halt and ask a human rather than continue. Phrase them as instructions to the agent. Cover the case where every sanctioned approach has failed and the task is still incomplete.