Use this prompt to turn a vague agent idea into a workflow you can review. It does not grant the agent permission to act.
Copy the prompt
Help me map this agent workflow before implementation.
Goal: [the user outcome]
User: [who initiates the task]
Inputs: [what the user provides]
Tools: [what the agent may access]
List the steps from input to result. For each step identify:
1. What the system can observe or verify.
2. What permission it needs.
3. Whether the action is reversible.
4. What the user sees while it runs.
5. The failure and recovery path.
Mark any action that sends data, spends money, changes access, or deletes work.
Do not infer permission for those actions from the goal alone.
Call out missing requirements before suggesting an implementation.
Finish with acceptance criteria that can be tested.
Review the output
Look for actions hidden inside vague verbs such as “sync,” “clean,” or “publish.” Split them until their permissions and effects are clear. Validate provider limits against current official documentation.
Use the map
Turn each agreed outcome into a testable requirement. A workflow is ready for implementation when its failure paths and user controls are as clear as its successful result.
