The Agent Skills Directory

[COMMAND_EXECUTION]: The Tools/Generate.ts script uses execAsync (a wrapper for Node's exec) to run ImageMagick (magick) commands. It interpolates the imagePath and hexColor variables directly into shell strings without sufficient sanitization. Since imagePath is derived from the --output CLI argument, an attacker could craft a malicious filename containing shell metacharacters (e.g., " ; id ; ") to execute arbitrary commands on the host system. This vulnerability exists in the addBackgroundColor and stampDaSignature functions.- [DATA_EXFILTRATION]: Every workflow within the skill, including the main instructions in SKILL.md, mandates a "Voice Notification" step that uses curl to send a POST request to http://localhost:31337/notify. This transmits workflow metadata and activity details to a local service. While targeting localhost, this automated telemetry behavior is an unexpected network operation for an image generation skill and could be used to monitor agent activity.- [EXTERNAL_DOWNLOADS]: The skill's tools fetch generated image assets from external services including Replicate (replicate.com), Google Generative AI endpoints, and Discord CDN (for Midjourney integration). While these are well-known services, the skill handles remote URLs and downloads them directly to the local filesystem.

Art