magic-linguistic-ethics

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

• Third-party content exposure detected (medium risk: 0.65). The skill’s runtime workflow ingests dataset documentation/metadata (e.g., source URLs, license statements, and attribution/lineage fields) that are typically derived from outsider-authored public datasets/pages fetched or provided by other systems, which can include free-form text from those sources into the LLM context for the CARE/FPIC/license/sacred-text checks.