The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill utilizes a 'Grounded Inputs' strategy that ingests untrusted external data (customer reviews and ad comments) to drive creative generation.
Ingestion points: External text data is read from the inputs/reviews/ and inputs/comments/ directories and processed directly into the agent's context.
Boundary markers: While the skill mandates a 'grounding' disclosure in its outputs to distinguish real from illustrative data, it lacks strict technical delimiters to isolate external text from instructions.
Capability inventory: The skill is capable of executing campaign management actions through external ad platform tools and generating local HTML artifacts.
Sanitization: The creative-review-template.html asset includes defense-in-depth measures, such as character escaping (\u003c) and the use of JSON.parse for data handling, to prevent XSS or script breakout within the generated review pages.
[EXTERNAL_DOWNLOADS]: The skill provides detailed guidance and API examples for several well-known generative AI and cloud services.
Services referenced: Google Gemini/Veo, Replicate (Flux), Ideogram, ElevenLabs, OpenAI, and Runway.
Context: These references are for established technology services and are used exclusively to support the skill's intended purpose of producing ad visuals and audio.
[COMMAND_EXECUTION]: The skill documents the use of local command-line tools for ad platform operations.
Evidence: SKILL.md contains example shell commands (e.g., node tools/clis/google-ads.js) for fetching campaign reports and managing ad data through local script execution.

ad-creative