The Agent Skills Directory

[PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface because it processes untrusted metadata harvested from external URLs to drive creative generation.
Ingestion points: Untrusted content from product and app store URLs is ingested via scripts/classify-ad-source.mjs and scripts/harvest-ecommerce-assets.mjs (referenced in SKILL.md and references/asset-harvest.md).
Boundary markers: The skill employs structured JSON outputs and explicit creative gates to minimize the impact of external content on agent behavior.
Capability inventory: The skill executes local Node.js scripts, performs file writes (e.g., ad-brief.json, concepts.json), and triggers video rendering processes.
Sanitization: Input properties are validated against a strict Zod schema located in assets/remotion-template/src/schema.ts.
[EXTERNAL_DOWNLOADS]: The skill fetches assets and libraries from established well-known technology services.
Evidence: The Remotion template downloads typography via @remotion/google-fonts (Google Fonts), and the Hyperframes template fetches the GSAP library from the JSDelivr CDN (https://cdn.jsdelivr.net/npm/gsap@3.14.2/dist/gsap.min.js).
[COMMAND_EXECUTION]: The skill provides instructions for executing local utility and build scripts included with the tool.
Evidence: The documentation directs users to run scripts such as scripts/classify-ad-source.mjs, scripts/harvest-ecommerce-assets.mjs, and scripts/validate-creative.mjs to automate the production workflow.

remotion-ad-video