azure-compute

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEDATA_EXFILTRATIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [Network Operations to External Services]: The skill instructions utilize curl -s ifconfig.me to detect the user's public IP address for the purpose of securing Network Security Group (NSG) rules. While this is a common administrative pattern, it involves a request to a third-party service outside the immediate vendor environment.
  • workflows/vm-creator/examples/terraform/README.md: MY_IP=$(curl -s ifconfig.me)/32
  • workflows/vm-creator/references/depth-probe/beginner.md: detect via curl -s ifconfig.me or equivalent
  • [Access to Sensitive File Paths]: The workflow documentation frequently references reading from ~/.ssh/id_rsa.pub to provide SSH public keys for VM provisioning. Although this targets a public key rather than a private secret, access to the .ssh directory is generally monitored for security.
  • workflows/vm-creator/examples/bicep/README.md: adminPublicKey="$(cat ~/.ssh/id_rsa.pub)"
  • workflows/vm-creator/references/output-adapters/az-cli.md: --ssh-key-values @~/.ssh/id_rsa.pub
  • [External Extension Installation]: The skill includes instructions to install Azure CLI extensions at runtime to support specific operations.
  • workflows/capacity-reservation/references/capacity-reservation-overview.md: az extension add --name resource-graph
  • workflows/essential-machine-management/references/emm-enable-flow.md: az extension add --name quota
  • [Remote Repository Operations]: The delivery workflow supports cloning and pushing to GitHub repositories using the gh CLI tool.
  • workflows/vm-creator/references/delivery-options/github-pr.md: gh repo clone {owner/repo} . and git push -u origin {branch}
  • [Indirect Prompt Injection Surface]: The skill processes user-supplied parameters to generate and execute infrastructure commands. While the Plan Card mechanism provides a human-review boundary, the ingestion of external data combined with the ability to execute shell commands and modify repositories represents a vulnerability surface.
  • Ingestion points: User input for VM names, resource groups, and network configurations in vm-creator.md and vm-recommender.md.
  • Boundary markers: The plan-card.md enforces a review step and uses a structured table with a Source column to distinguish user input from inferred values.
  • Capability inventory: The skill can execute az CLI commands, terraform operations, and gh repository management across multiple workflow files.
  • Sanitization: The github-pr.md workflow includes instructions to scrub sensitive passwords from generated files and use environment variables instead.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 12:06 AM
Security Audit — agent-trust-hub — azure-compute