verl-rl-training

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

• Third-party content exposure detected (high risk: 0.90). The SKILL.md explicitly requires loading base models and reward models from public hubs (e.g., "Base model from HuggingFace Hub" and config fields like actor_rollout_ref.model.path: Qwen/... and reward_model: OpenRLHF/Llama-3-8b-rm-700k), which are untrusted third‑party artifacts that the agent will load and use to generate behavior—allowing indirect prompt-injection via those external model/data sources.