The Agent Skills Directory

[SAFE]: Indirect Prompt Injection Surface. The skill is designed to process external, untrusted data to ground its creative output. Ingestion points: reads customer reviews and ad comments from the inputs/ directory and campaign performance metrics from CSV or API sources (referenced in SKILL.md and references/static-ad-templates.md). Boundary markers: while explicit prompt delimiters for untrusted content are not defined in the instructions, the grounding system isolates source material from creative generation instructions. Capability inventory: the skill can execute command-line tools for ad platforms, write dated output files, and generate HTML review pages (assets/creative-review-template.html). Sanitization: the provided HTML review template includes a specific character escaping mechanism for string values and uses JSON.parse for inert data handling to prevent script injection.
[SAFE]: External dependency and tool references. The skill's documentation (references/generative-tools.md and references/imessage-video-ads.md) suggests integrating with well-known third-party services like Google Gemini, OpenAI, and ElevenLabs, as well as specialized tools like gooseworks. These references are provided as optional production routes for video generation and target legitimate technology providers.
[SAFE]: Command execution for media processing. The skill details workflows for batch rendering video and image variations using standard tools like ffmpeg, playwright, and remotion. These operations are performed on local project assets and are strictly aligned with the core functional requirements of producing high-volume ad creative.

ad-creative