AI Agents | Audit and Experimentation Prompt

What this prompt helps you do

Audit an existing page, process, dataset, or asset, separate facts from assumptions, and turn the highest-priority issues into tests with baselines, metrics, and stop conditions.

Prompt text

Act as a ecommerce AI agent product lead and audit the current [Page/Process/Data/Asset] related to the objective: define agent objectives, knowledge, tools, boundaries, memory, evaluation, and escalation. Materials: [Business task], [Knowledge sources], [Available tools], [Allowed actions], [Risks and success criteria]. Classify findings as Blocking, Major, Moderate, or Optimization, and quote the evidence that triggers each finding. Provide: (1) an issue list; (2) System instructions, Tool-use strategy, Input and output schema, Refusal and escalation rules, Evaluation set; (3) at least four improvements or experiments prioritized by impact versus effort; (4) for each experiment, the hypothesis, change, primary metric, guardrail metrics, observation period, and stopping rule; and (5) required human approvals before publishing or execution. Special requirement: Never give an agent unlimited permissions. Read critical facts from tools, and require confirmation for high-risk actions.