The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to analyze performance data (provided as CSV, text, or API output) to identify winning themes and generate new ad variations. Since this data is external and potentially attacker-controlled, it could contain hidden instructions intended to override the agent's behavior.
Ingestion points: Performance data ingestion in SKILL.md (Iterating from Performance Data section).
Boundary markers: No delimiters or safety instructions are defined to separate the untrusted data from the agent's internal logic.
Capability inventory: The skill uses local CLI tools (node tools/clis/google-ads.js) and provides examples for making network requests to AI service APIs via curl.
Sanitization: The skill lacks explicit sanitization or verification steps for the ingested metrics and ad copy content.
[REMOTE_CODE_EXECUTION]: The skill provides instructions in references/generative-tools.md for installing 'Voicebox' by cloning a GitHub repository and running make setup. This pattern involves downloading and executing external code that is not from a pre-verified trusted vendor.
[COMMAND_EXECUTION]: The skill documents and instructs the agent on the use of several CLI tools, including node for platform integrations, git and make for repository management, npx for rendering video assets, and ffmpeg for media handling.
[EXTERNAL_DOWNLOADS]: The skill references multiple external services for image, video, and voice generation, including code snippets that use curl to interact with remote APIs (e.g., Gemini, ElevenLabs, Cartesia).

ad-creative