Lab · Prompts · 02
5 prompts for designing, testing, and auditing human-in-the-loop workflows. HITL is effective only when the reviewer has clear criteria, real authority, and a process that cannot be bypassed.
Design the HITL checkpoints for an AI workflow
Map where human review must occur in an AI-assisted process, specifying who reviews, what they assess, and what authority they have.
Write a reviewer brief for a HITL checkpoint
Produce the written brief that a human reviewer receives before assessing an AI output at a defined checkpoint.
Audit an existing HITL workflow for failure modes
Identify ways the human review step in a workflow could fail to catch harmful AI outputs, even when nominally in place.
Define escalation triggers for an AI system
Specify the conditions that trigger escalation from automated processing to human review, written as operational rules.
Write a HITL log entry for a completed review
Produce a standardised log entry after a human reviewer has completed a HITL checkpoint, for inclusion in the governance audit trail.