replicate

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

• Third-party content exposure detected (high risk: 0.80). The SKILL.md's Fine-Tuning section explicitly shows ingesting training data from an external URL ("https://my-bucket.s3.amazonaws.com/training-images.zip"), which demonstrates the skill can fetch untrusted/third-party content and use it to train models that materially change behavior.