moe-training

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly recommends using public training data (e.g., "The Pile" in references/training.md and a --data-path for training) and examples that load models from public hubs (e.g., mistralai/Mixtral via vLLM), so the workflow expects ingesting open, user-generated third‑party content which could carry indirect prompt injections.