geniml

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

• Third-party content exposure detected (high risk: 0.70). The skill explicitly fetches and ingests external, potentially user-contributed genomic data and models (e.g., BBClient.load_bed in references/utilities.md which fetches BED files from remote BEDbase/GEO-like repositories, and ScEmbed.from_pretrained / Hugging Face model loading referenced in scembed.md), and those external files/metadata are consumed as part of tokenization, training, and search workflows where their content can materially affect model behavior and search results.