The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest "Grounded Inputs" from untrusted external sources, including customer reviews from platforms like Amazon, Trustpilot, and G2, as well as campaign comments from social media. These inputs are used directly to generate ad copy, which creates an attack surface where malicious instructions embedded in the reviews could influence the agent's behavior during the generation process.
Ingestion points identified in SKILL.md: inputs/reviews/ and inputs/comments/.
Boundary markers: While the skill emphasizes "grounding," it lacks explicit instructions for the sanitization of embedded instructions within the ingested text.
[EXTERNAL_DOWNLOADS]: The documentation in references/generative-tools.md, references/imessage-video-ads.md, and references/motion-video-ads.md directs the user or agent to clone and reference third-party GitHub repositories that are not recognized as trusted vendors. Examples include jamiepine/voicebox, Bomx/super-video-maker-skill, and gooseworks-ai/gooseworks-ads-skills.
[REMOTE_CODE_EXECUTION]: The skill provides instructions for the execution of remote code from unverified sources. Production guides recommend running commands such as npx gooseworks install --all and git clone https://github.com/jamiepine/voicebox.git && make setup, which could lead to the execution of malicious logic if the agent or user follows these guides without manual verification.
[COMMAND_EXECUTION]: The skill relies on and documents the use of local CLI tools (e.g., google-ads.js, meta-ads.js, linkedin-ads.js) and system-level utilities like ffmpeg and Playwright. These tools grant the agent the capability to perform complex network operations and file system manipulations within the user's environment to manage ad campaigns and render creative assets.

ad-creative