The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, specifically customer reviews and ad comments (inputs/reviews/, inputs/comments/), to ground the generation of ad copy. This creates a surface for indirect prompt injection if the ingested content contains malicious instructions designed to influence the agent's behavior.
Ingestion points: The skill reads from local files in inputs/reviews/ and inputs/comments/, as well as performance data provided by the user or fetched via APIs.
Boundary markers: There are no explicit instructions or delimiters defined to isolate ingested data from the skill's own instructions, although it emphasizes using customer language "literally" or "verbatim."
Capability inventory: The skill instructs the agent to execute shell commands (e.g., node tools/clis/google-ads.js, npx remotion render, ffmpeg), perform network requests to external AI service providers (e.g., Google Gemini, ElevenLabs, Ideogram), and write generated assets to the outputs/ directory.
Sanitization: The references/creative-review-page.md document includes guidance for escaping characters when populating the HTML review template, and the template itself (assets/creative-review-template.html) contains a basic escaping function for HTML rendering.
[EXTERNAL_DOWNLOADS]: The skill recommends the installation and use of external software and repositories that are not part of the standard trusted vendor list, including npx gooseworks install --all for the iMessage video pipeline and git clone https://github.com/jamiepine/voicebox.git for local voice synthesis.
[COMMAND_EXECUTION]: The skill provides numerous examples and instructions for executing local shell commands to manage ad data and assemble video assets, such as using node for platform CLIs and ffmpeg for audio/video processing.

ad-creative