sparse-autoencoder-training

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly loads and uses community-hosted models, datasets, and feature browsers from public third-party sites (e.g., SAE.from_pretrained from HuggingFace/GitHub, neuronpedia.org, and the "monology/pile-uncopyrighted" dataset), which are untrusted/user-provided content the agent ingests and inspects as part of its workflow.