a1-consumer-psy

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow ingests the outsider-authored “User Instruction” and “Input” free text to produce diagnoses (e.g., cases like evals/cases/psy-belenf-... feed user-supplied copy into the model via the required workflow steps that read the diagnostic spine, then produce the diagnosis for the user-facing output).