sagemaker-ai-ops-review
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFE
Full Analysis
- Read-Only Scope Enforcement: The skill instructions explicitly limit the agent to 'List', 'Describe', and 'Get' API calls. It strictly prohibits resource mutation, endpoint invocation, and data-plane access, ensuring the audit process does not affect production workloads.
- Security Posture Evaluation: The skill is designed to detect configuration gaps, such as the absence of customer-managed KMS keys for notebook instances or the use of public internet access for SageMaker Studio domains. These checks are aligned with the skill's core purpose of identifying operational risks.
- Data Integrity Guardrails: The instructions include specific prohibitions against data fabrication, such as 'Never state a number the APIs did not return'. This helps maintain the accuracy and reliability of the generated operational review.
- Information Ingestion Surface: The skill ingests metadata from external sources, including AWS resource names and tags, to populate its findings. While this constitutes an attack surface for indirect instructions, the skill manages the risk by requiring a structured Markdown output and a mandatory AI disclaimer to contextualize the results. (Ingestion points: AWS resource metadata and metrics; Boundary markers: Mandatory report headers and AI Disclaimer blockquote; Capability inventory: Read-only AWS API calls via use_aws; Sanitization: Not explicitly defined beyond structured formatting).
Audit Metadata