The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs extensive system audits by executing shell commands such as sshd -T, ufw status, and ss -tlnH to inspect host and network security.
[EXTERNAL_DOWNLOADS]: The skill installs standard security packages including fail2ban and unattended-upgrades from official repositories and uses xcaddy to build plugins from the well-known Caddy community GitHub repository.
[REMOTE_CODE_EXECUTION]: Conducts remote security audits over SSH on targets specified by the user using existing credentials. This is a core feature for server hardening.
[SAFE]: The skill includes an attack surface for indirect prompt injection via command output processing. 1. Ingestion points: scripts/audit.sh reads output from ss, sshd, and ufw tools. 2. Boundary markers: The skill mandates a human-approval gate for all non-read-only actions and clear separation between audit and apply phases. 3. Capability inventory: The skill possesses the capability to modify system files in /etc/ssh/, manage firewall rules via sudo, and execute remote commands over SSH. 4. Sanitization: The skill uses basic filtering and relies on operator review to validate audit findings before application. The overall design effectively mitigates risk through these human-in-the-loop gates.

infra-security-audit