Set Boundaries for an Agent
Decide which process steps an agent can handle, which should stay with a person, and where human approval is needed; identify likely failure situations and safeguards.
Prompt template
- Format
- Text Parameterized
- Inputs
- Process_steps (text, markdown) · Required
- Outputs
- Agent_boundary_plan (text, markdown)
Here are the process steps: ${process-steps}.
Say which steps should be handled by an agent, which should remain with a person, and where a person must confirm an action before it happens. Separately list five situations in which the agent could make a mistake and what should happen instead. Be specific.
Replace the ${…} placeholders with your values
before using this prompt. See parameters
and defaults below
Parameters
Replace each placeholder with your own value, or use its default when one is provided
${process-steps}
process-steps
Required
What you'll create
The analysis assigns responsibility and identifies actions that need human confirmation. Five concrete failure situations include the behavior that should replace a mistaken action.
Frequently asked questions
Are failure cases included?
Yes. Five specific situations.
Does every process step automatically go to the agent?
No. Human-owned steps are identified separately.
Comments
Tried this prompt? Share what you made, compare results, or ask a question