ultraqa

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.70). The required workflow includes “Prompt injection attempts” and explicitly instructs the agent to execute adversarial scenarios and then use /prompts:architect with the “Scenario matrix” and “Output” (which can include attacker/user-provided free text), so outsider-authored text can be ingested into the LLM context via runtime test/harness output and scenario-matrix evidence.