torchforge-rl-training

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly ingests public, user-contributed data and models (e.g., dataset.path: "openai/gsm8k" and models referenced from the HuggingFace Hub) as part of its training workflow, so the agent will read and interpret untrusted third‑party content from the open web.