You are implementing advanced prompt engineering for a multi-agent procurement solution in Azure. Supplier agents and a compliance agent share context. Red teaming found agents can be tricked into ignoring policy when a user message says “ignore previous instructions.” Which prompt engineering approach best hardens the system?
Select an answer to reveal the explanation.
Short Explanation
C is solid multi-agent prompt hygiene. Defensive guidelines live in trusted system/developer layers; untrusted user text is treated as data, not instructions; few-shot examples show refuse and handoff; and prompts are versioned like code. That addresses prompt injection patterns such as “ignore previous instructions.” A weakens control. B pollutes the untrusted channel with sensitive policy and secrets. D confuses context size with security. In Foundry multi-agent builds, prompt lifecycle management—versioning, review, and regression tests—is part of development, not a one-time paste into a notebook.
Full Explanation
C is correct because developing multi-agent solutions requires advanced prompt engineering: examples, dynamic context injection discipline, defensive guidelines, and prompt lifecycle management. Separating trusted instructions from untrusted user content and encoding refuse/handoff examples reduces instruction-override attacks across collaborating agents. A is incorrect: removing system prompts eliminates policy anchors. B is incorrect: placing secrets and authoritative policy only in user messages makes them easier to override or leak. D is incorrect: larger token limits do not implement defensive instruction hierarchy. Combine prompt defenses with tool-level authorization—prompts alone are insufficient.