compact-hypercube-embeddings-fast

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime workflow ingests user-provided paired data (e.g., “iNaturalist images with … captions” or “text-observation pairs”) and trains on those text fields, which are outsider-authored free-form content not authored by the operating user.