The Agent Skills Directory

[COMMAND_EXECUTION]: The script block for generating prototypes uses the $ARGUMENTS variable in a way that allows for shell command injection. Specifically, the command create_prototype "$(echo $ARGUMENTS | tr ' ' '-')" executes a subshell where user-provided input is interpreted by the shell. An attacker could craft a design challenge string containing backticks or subshell syntax to execute arbitrary commands.
[COMMAND_EXECUTION]: The skill's context-gathering and idea-prompting scripts also incorporate $ARGUMENTS into shell commands without robust sanitization. While some scripts use the tr command to handle spaces, other dangerous shell metacharacters (e.g., semicolons, pipes) are not filtered, posing a risk of unintended code execution if the agent invokes these scripts with malicious user input.

brainstorm