The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent on the use of significant system networking tools, including tcpdump for packet capturing, mtr and dig for network diagnostics, and ip route for routing table manipulation. These tools provide extensive visibility into or control over the host system's network stack.
[INDIRECT_PROMPT_INJECTION]: The skill outlines workflows where hostnames, IP addresses, and CIDR blocks provided in the troubleshooting context are used as arguments for system commands. This pattern presents a vulnerability surface if the input is not strictly validated before execution.
Ingestion points: External network identifiers (hostnames, IP addresses, interface names) used in the triage steps in SKILL.md and references/tunnels.md.
Boundary markers: There are no specified delimiters or instructions provided to isolate or ignore potentially malicious instructions embedded in the network metadata being processed.
Capability inventory: Includes execution of tcpdump, mtr, ss, ip, ping, dig, and WireGuard (wg) command-line utilities.
Sanitization: The instructions do not describe sanitization or escaping procedures for handling user-supplied identifiers before their interpolation into command strings.
[PERSISTENCE]: The documentation in references/routers.md recommends the use of automated configuration exports via the RouterOS scheduler or system cron jobs. While these are standard practices for system administration and backup, they represent a persistence mechanism on the device.

z-net-core