The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process customer-generated content such as reviews and comments from external files (e.g., inputs/reviews/, inputs/comments/). This presents an indirect prompt injection surface where malicious instructions could be embedded in the processed data. The skill mitigates this by providing explicit instructions to escape specific characters (like < to \u003c) when populating the HTML review template.
Ingestion points: SKILL.md (Grounded Inputs section) and references/creative-roadmap.md (Customer/brand signals).
Boundary markers: Present. The skill instructs the agent to isolate data within a specific JSON block in the assets/creative-review-template.html file.
Capability inventory: The skill performs file creation (HTML review page), executes shell commands for advertising platform CLIs (Google, Meta, TikTok), and generates text/code via LLM prompts.
Sanitization: Present. references/creative-review-page.md contains mandatory instructions for the agent to HTML-escape values at render time and specifically escape script tags to prevent injection into the generated output.
[COMMAND_EXECUTION]: The skill documents the use of various command-line interface (CLI) tools (e.g., google-ads, meta-ads, tiktok-ads) to pull performance data and manage campaigns. These operations are essential to the skill's primary function and are intended to be executed in the user's environment.
[EXTERNAL_DOWNLOADS]: The skill references several external tools and APIs for creative production, including code-based video generation (Remotion) and a third-party ad pipeline (Gooseworks). These are presented as production route options for the user rather than automated executions by the skill itself. The referenced generative tools (Gemini, Flux, ElevenLabs, etc.) are well-known and established services.

marketing-ad-creative