runner-testing

Warn

Audited by Gen Agent Trust Hub on Jun 23, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill facilitates the execution of arbitrary commands on remote hardware hosts using scripts/cf-ssh.sh. This includes frequent use of sudo for sensitive operations such as reading logs in /var/lib/vm0-runner/, checking journalctl, and managing system services.
  • [REMOTE_CODE_EXECUTION]: The pnpm runner:submit command implements a workflow where the user's prompt is directly executed as a bash script within a Firecracker VM sandbox via the mock-claude component.
  • [COMMAND_EXECUTION]: The skill provides the pnpm runner:exec command, which allows for direct command execution within already running VM instances for debugging purposes.
  • [CREDENTIALS_UNSAFE]: The documentation describes an authentication flow using JWTs and Authorization: Bearer tokens passed to VM instances as environment variables (VM0_API_TOKEN). While the flow is described, no hardcoded secrets or sensitive credentials were found in the skill content itself.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 23, 2026, 10:21 PM
Security Audit — agent-trust-hub — runner-testing