agent-platform-eval-flywheel

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

• Third-party content exposure detected (medium risk: 0.30). SKILL.md’s required runtime workflow ingests user-authored evaluation dataset text (e.g., Stage 2 client.evals.run_inference(..., src=dataset/df) and endpoint/MaaS scripts reading --dataset JSONL prompts/responses), so an outsider can poison the dataset via the queued GCS dataset source the workflow consumes.