Give me a short checklist for getting an agent to stop safely. I want something I can run before I call the work done.
getting an agent to stop safely: the version with actual constraints
I wanted help with getting an agent to stop safely without pretending the model knew facts I had not provided.
Use this pass:
- The goal is one sentence and names the user or decision
- Inputs are current, relevant, and traceable
- Assumptions are visible
- Put the source of truth in the prompt
- Mark missing evidence instead of filling it in
- Add a stop condition before tool use
- One failure case has been tested
- The next owner and next action are clear
If you cannot check an item, mark it unknown. Do not quietly treat unknown as done.
Saving this. The unknown-versus-done line catches most of my shortcuts.
4 comments
Sign in to join the conversation.