Use when the user says “激活红队模式 / 开红队 / 切红队” or wants an authorized attacker-perspective workflow for Web, host, AD, cloud, container, API, or AI-system security work; default to mode activation plus safety-boundary declaration, not immediate prompt-injection…
Use when old sessions, saved notes, or automation still reference the legacy Claude-named prompt-injection robustness skill and you need the backward-compatible alias that points to the canonical model-agnostic workflow.