Stakeholders want a refund agent that can draft refund recommendations but must never execute refunds above $500 without human approval. How should you specify the agent’s design?
Select an answer to reveal the explanation.
Short Explanation
You need to spell out the agent’s persona, scope, autonomy, and rules: it can draft refund recommendations, but anything over $500 needs human approval and the tools must block direct high-value refunds without an approval token. That keeps automation useful while keeping control where it belongs.
Full Explanation
Design skills include specifying agent personas, scopes, boundaries, autonomy levels, and behavioral guidelines, plus human-in-the-loop for high-risk actions. The correct design captures those elements and ties them to tool permission boundaries: draft refunds, require approval over $500, and deny high-value tools without approval tokens. Full production DB admin rights and full autonomy over-privilege the agent. A threshold hidden only in source-code comments is not an operable control. Document the autonomy matrix, make autonomy levels testable and visible in HAX flows, and implement approval workflows that emit audit events.