add-benchmark

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

• Third-party content exposure detected (high risk: 0.75). The workflow ingests dataset JSONL (including responses_create_params.input with OpenAI message system/user free text) from GitLab at runtime via the train/validation dataset download/collation step, meaning that outsider-authored dataset content can be fed into the agent’s LLM context.