sentencepiece

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

• Third-party content exposure detected (high risk: 0.70). The skill's training guide (references/training.md) explicitly shows loading a public Hugging Face dataset via datasets.load_dataset('wikitext', ...) and training from that iterator, which ingests untrusted, user-generated web text as part of the training workflow and can materially influence tokenization/model behavior.