Audit an LLM application for indirect prompt injection - compose objective x technique payloads, deliver them through the channels the agent actually reads, and prove impact with an out-of-band callback. Use to assess an AI agent's resistance to injected instructions.