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 ingests and processes external, user-provided training data and models (e.g., data.train_files pointing to parquet datasets with "prompt" fields and external reward/model paths like OpenRLHF/... and HuggingFace Hub model paths such as Qwen/...) which are untrusted third‑party/user-generated content the agent will read and interpret during training.