cloud-penetration-testing

Fail

Audited by Gen Agent Trust Hub on Jun 14, 2026

Risk Level: HIGHCOMMAND_EXECUTIONCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill provides instructions for importing "stolen" tokens (Import-AzContext), exporting access tokens to local files (Save-AzContext), and extracting secrets from Azure Key Vault and AWS Lambda environment variables.
  • [COMMAND_EXECUTION]: Extensive use of cloud-provider CLI tools (Az, AWS, gcloud) to perform invasive operations, including the creation of backdoor administrative users and service principals with global administrator privileges.
  • [REMOTE_CODE_EXECUTION]: Contains commands to execute arbitrary PowerShell scripts on remote Azure Virtual Machines using the Invoke-AzVMRunCommand utility.
  • [EXTERNAL_DOWNLOADS]: Fetches installers for AWS and GCP CLI tools from official domains (awscli.amazonaws.com and sdk.cloud.google.com). It also references numerous third-party security tools and vulnerable training environments hosted on various GitHub repositories.
  • [DATA_EXFILTRATION]: Includes methods for identifying and bulk-downloading data from cloud storage (S3 buckets, Azure storage accounts) and extracting metadata from sensitive services.
  • [PROMPT_INJECTION]: The skill demonstrates an attack surface for indirect prompt injection by processing external data files (e.g., users.txt, ips.txt, regions.txt) through various scripts and loop constructs without explicit sanitization or boundary markers.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jun 14, 2026, 10:05 PM
Security Audit — agent-trust-hub — cloud-penetration-testing