The Agent Skills Directory

[SAFE]: No malicious patterns such as credential theft, obfuscation, or persistence mechanisms were detected. The skill follows legitimate performance marketing workflows and emphasizes the use of grounded inputs to ensure output accuracy.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, which is a standard attack surface for AI agents.
Ingestion points: The agent reads customer reviews, ad comments, and performance data from the inputs/ directory (SKILL.md).
Boundary markers: The instructions lack explicit boundary markers (such as XML tags or specific delimiters) to separate untrusted review text from system instructions.
Capability inventory: The skill utilizes high-capability tools including ad platform CLIs (Google, Meta, TikTok) and system utilities like ffmpeg and Playwright for video rendering (SKILL.md, references/imessage-video-ads.md).
Sanitization: Content is used verbatim from reviews to maintain "real texting voice," and no explicit sanitization step is described for the ingestion phase.
[EXTERNAL_DOWNLOADS]: The skill references external production routes involving the installation of third-party tools like gooseworks and Remotion via npx. These are legitimate developer tools used in the context of the skill's purpose (references/imessage-video-ads.md, references/generative-tools.md).

ad-creative