Tests whether a counterparty agent's input handling resists prompt-injection and content-trap patterns — a concrete resilience check, not a compliance checkbox.