aws-bedrock-evals

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill requires loading a pre-computed inference dataset from S3 at runtime (e.g., s3://YOUR_BUCKET/datasets/collected-responses.jsonl), and that externally-hosted file directly supplies the prompts/responses the Bedrock evaluator consumes—meaning remote content can control the evaluation inputs.