The Agent Skills Directory

[DATA_EXFILTRATION]: The scripts yt-thumb.mjs, yt-boxes.mjs, and rasterize-logo.mjs utilize a headless Playwright browser to compose images. They use the file:// protocol to load local assets provided via command-line arguments (e.g., --face, --scene, --icon). This creates a vulnerability where an attacker could influence the agent to pass sensitive local file paths (such as ~/.ssh/id_rsa or .env files), causing the browser to render their contents into the resulting screenshot, effectively exfiltrating the data as an image.- [SAFE]: The skill relies on legitimate, well-known libraries including playwright for rendering, rembg for background removal, and pillow for image manipulation.- [SAFE]: The skill uses standard package managers (pip and npm) for installation and does not contain unauthorized remote code execution patterns.- [SAFE]: The Node.js scripts contain hardcoded absolute paths pointing to a specific user directory (/Users/jasonqwwen/Library/Caches/ms-playwright). While not directly malicious, this leaks developer metadata and may cause the scripts to fail or behave unexpectedly on other systems.

youtube-thumbnail