create-skill-autoresearch

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

• Third-party content exposure detected (high risk: 0.75). Outsider-authored free text can enter the LLM context via Phase 2/4/5 evaluation and judging: the factory ingests user-provided input/ gold standards/study materials (which may include outsider-authored text) and then feeds them to LLMs as system/user messages in evaluate.sh (LLM-as-judge) and to panel verifiers (skill + gold standards + rubric + premortem risks).