The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a shell command (mkdir -p ./audit-reports/{target-slug}) where the directory name is derived from user-controlled input ($ARGUMENTS).
[COMMAND_EXECUTION]: Although the instructions state to 'Convert the target argument to a URL-safe folder name', this is a natural language instruction rather than a programmatic constraint. An attacker could potentially provide input containing shell metacharacters (e.g., ;, &, |) to execute arbitrary commands if the agent does not properly sanitize the input before passing it to the shell.
[COMMAND_EXECUTION]: The skill specifies multiple output locations and directory creation steps across different modes (/audit-cost and /plan-cost) that all rely on this unsanitized input pattern.

zerubbabel-cost