AI Agent · Safety

Constraints and guardrails

Automatic Agent prompt to enforce safety and scope guardrails.

All topics / AI Agent prompts

Prompt shell

Replace highlighted brackets, then copy.

Execute [task] but NEVER do [prohibited]. If data is missing, request it. If risk is high, pause and ask for approval.
How to use
  • task - replace the bracketed field with your info
  • prohibited - replace the bracketed field with your info
Provide (Inputs)
  • Task
  • Prohibited actions
  • Risk threshold
Expected output (Outputs)
  • Safe execution
  • Approval request if needed
Tips
  • List prohibited actions explicitly.
  • Define what 'high risk' means.
  • Log any refusal reasons.
Related prompts
More AI Agent prompts Ask the encyclopedia