data-pipeline-manager

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

• Third-party content exposure detected (high risk: 0.80). The skill's contamination audit (references/contamination-guide.md) explicitly instructs the agent to "Download or access the benchmark's test split" and to compare test examples against public pretraining sources and published instruction‑tuning datasets, which requires ingesting open, untrusted third‑party content that can change decisions about contamination and next actions.