The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user to execute unverified third-party software using npx gooseworks install --all in references/imessage-video-ads.md. The use of npx for package installation without version pinning or source verification allows for the execution of arbitrary remote code.\n- [EXTERNAL_DOWNLOADS]: External links in SKILL.md (e.g., https://github.com/mbzadev/skills/blob/main/tools/REGISTRY.md) point to a GitHub organization mbzadev which differs from the author's specified handle mabzadev by a single character. This presents a potential typosquatting risk where users could be directed to an untrusted repository.\n- [PROMPT_INJECTION]: The skill exhibits a significant surface for indirect prompt injection. \n * Ingestion points: Reads untrusted customer feedback from inputs/reviews/ and inputs/comments/ (SKILL.md). \n * Boundary markers: Absent. The skill lacks instructions to delimit or ignore potentially malicious instructions embedded in these external text files. \n * Capability inventory: The skill has capabilities to write files to the outputs/ directory and execute shell commands via local CLI tools (node tools/clis/google-ads.js). \n * Sanitization: Absent. The instructions explicitly direct the agent to copy user language verbatim into its output, which could lead to the execution of embedded instructions by the agent or downstream systems. (Severity: LOW).\n- [COMMAND_EXECUTION]: The skill relies on the execution of local scripts and CLI tools (e.g., google-ads, meta-ads, ffmpeg, Playwright) to manage ad campaigns and render video content. While these are presented as integration tools, they represent a broad surface for command injection if inputs are not properly sanitized.

ad-creative