agentic-vps

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The provision script performs network fetch-and-execute at runtime (high-confidence examples: curl -fsSL https://tailscale.com/install.sh | bash, curl --proto "=https" --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y, curl -fsSL https://bun.sh/install | bash, curl -fsSL https://deb.nodesource.com/setup_22.x | bash, and runs npm install -g @anthropic-ai/claude-code @anthropic-ai/sandbox-runtime), which fetches and executes remote code the skill relies on during provisioning.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The prompt explicitly instructs running root provisioning scripts, creating a non-root agent user with passwordless sudo, and changing firewall/SSH access (UFW/edge rules, removing public :22), i.e. directed privileged system modifications and account creation.