testing-integration

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

• Third-party content exposure detected (high risk: 0.80). The skill includes a "golden dataset" of public URLs (backend/data/golden_dataset_backup.json) and explicitly documents restore/restore commands that "regenerates embeddings" (scripts/backup_golden_dataset.py restore) and end-to-end analysis workflows that run real LLM-based pipelines against ingested content, which shows the agent ingests untrusted public web content that can influence LLM-driven decisions.