database-cloud-optimization-cost-optimize

Pass

Audited by Gen Agent Trust Hub on Jun 22, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted data from cloud resource metadata (such as instance tags, names, and CloudTrail event details) via boto3 calls in resources/implementation-playbook.md. This data is used to drive rightsizing recommendations and automated tagging without sanitization.
  • Ingestion points: Resource tags and event logs retrieved via AWS SDK in resources/implementation-playbook.md.
  • Boundary markers: None identified; metadata values are processed as trusted input for optimization logic.
  • Capability inventory: The skill contains scripts to modify and stop EC2 instances, change storage policies, and create VPC endpoints (ec2.stop_instances, ec2.modify_instance_attribute, s3.put_bucket_lifecycle_configuration).
  • Sanitization: None; resource names and tag values are used directly in generated reports and tagging automation.
  • [COMMAND_EXECUTION]: The skill generates and recommends executable scripts in Python, Terraform, and Kubernetes YAML to implement cost-saving measures. Although intended for implementation by the user, the broad instruction in SKILL.md to 'Implement budgets, alerts, and ongoing optimization' could lead an agent to execute these powerful infrastructure-altering commands autonomously.
  • [EXTERNAL_DOWNLOADS]: The generated HTML dashboard templates in resources/implementation-playbook.md include references to external libraries from well-known services, specifically cdn.jsdelivr.net and d3js.org. These are standard dependencies for rendering charts and visualizations.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 22, 2026, 09:02 PM
Security Audit — agent-trust-hub — database-cloud-optimization-cost-optimize