The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Automated security scanners have confirmed critical detections for resources included with the skill. The file library/sites.json is flagged for malware (FileRepMalware), and the reference URLs https://thedeepend.space and https://extension.textmode.art are blacklisted by reputation services as malicious. The skill also programmatically fetches content from public APIs including Openverse and Wikimedia Commons.
[REMOTE_CODE_EXECUTION]: The assets/scenes/stage.js script dynamically loads and executes the GSAP, ScrollTrigger, and Lenis JavaScript libraries from the JSDelivr content delivery network.
[COMMAND_EXECUTION]: The visual inspection framework in scripts/lib/cdp.mjs identifies and spawns local browser executables (Chrome, Edge, or Chromium) to perform headless rendering and analysis of web documents.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection. * Ingestion points: Retreives media metadata from public search APIs in scripts/assets.mjs and analyzes the content of user-provided HTML files in scripts/inspect.mjs. * Boundary markers: The skill's instructions do not mandate the use of delimiters when processing external tool output. * Capability inventory: Spawns local processes and executes arbitrary JavaScript in the browser context via scripts/lib/cdp.mjs. * Sanitization: Metadata from external sources is processed with insufficient filtering to prevent instruction override.
[DYNAMIC_EXECUTION]: The design auditing logic uses the Chrome DevTools Protocol to evaluate arbitrary JavaScript strings within a headless browser instance during the visual measurement pass.

design-taste-frontend