This is a synthetic design exercise for an administrative reminder, not a deployed or tested workflow. DEMO-42 is a fictional task, with no patient or customer data.
Documented starting point
n8n’s human-review documentation describes pausing selected AI Agent tools before execution. The reviewer sees the proposed tool and parameters. Approval runs that tool with the AI-specified inputs; denial cancels it. The documentation also recommends telling the agent how to handle rejection.
$tool.name and $tool.parameters can supply review context. I would still check which inputs reach the review message rather than assume it displays every final tool setting.
A proposed review card
For a fictional reminder, I would show:
- task ID and version;
- destination and exact message;
- scheduled time and timezone;
- missing information that prevents sending.
My proposed application rule: a change to the destination, message or time invalidates the previous approval and requires a new review. This is a design requirement to implement and test, not a built-in guarantee established by the documentation.
Three cases to test
- Approve the unchanged draft: does the outgoing action match the reviewed values?
- Deny it: does the workflow avoid sending and request clarification?
- Edit it after approval: can the old approval still trigger a different action?
Would you bind approval directly to the tool call, or to a versioned task that a separate deterministic sender executes?
Top comments (0)