knowledge-distillation

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

• Third-party content exposure detected (high risk: 0.90). The SKILL.md explicitly loads public models via AutoModelForCausalLM.from_pretrained("meta-llama/...") and generates teacher_responses with teacher.generate (see the "Response Distillation" and training sections), so the skill ingests outputs from external, public models/repositories (and even clones a GitHub repo) which are untrusted third‑party content that can be read and used to drive training/behavior.