medchem

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime path scripts/filter_molecules.py → apply_query() → mc.query.QueryFilter(query) ingests OUTSIDER-authored free text from the user-provided --query CLI argument (arbitrary query string), which is then parsed/evaluated and can indirectly influence the LLM context if the agent includes that text in prompts.