database-cloud-optimization-cost-optimize
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the cloud environment including resource names, tags, and billing records (e.g., in
_analyze_by_serviceandAutoTagger). - Ingestion points: Cloud billing data (
ce.get_cost_and_usage), CloudTrail events (viaAutoTagger), and resource metadata like names and tags. - Boundary markers: Absent. There are no instructions to ignore embedded commands within the processed resource metadata.
- Capability inventory: The skill contains logic to stop/start EC2 instances, resize instance types (
modify_instance_attribute), and modify S3 bucket lifecycle policies. - Sanitization: The provided code does not show sanitization of resource names or tags before they are processed in reports or used in automation logic.
- [DYNAMIC_EXECUTION]: The implementation playbook includes several methods that return executable code or configuration files as strings to be used by the agent or user.
- Evidence:
create_rightsizing_automationreturns a Python script for EC2 modification,create_spot_configurationreturns Terraform HCL, andcreate_lifecycle_policiesreturns Python code for S3 management. - Pattern: The skill generates scripts from internal templates which are then presented as actionable outputs. While these are intended for legitimate optimization, the generation of executable code is a significant capability.
Audit Metadata