google-cloud-configs

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

• Third-party content exposure detected (high risk: 0.80). The skill clearly ingests public/untrusted data: the training examples and scripts read public BigQuery data (examples/bigquery-regression-example.sql selects from bigquery-public-data.new_york_taxi_trips) and load datasets from Hugging Face (examples/vertex-huggingface-finetuning.py and examples/vertex-pytorch-training.py call datasets.load_dataset and accept Hugging Face dataset/model names), so the agent is expected to read and process arbitrary third‑party/user-provided content.