The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script scripts/gz-fonts.mjs fetches font files from the official Google Fonts repository on GitHub (raw.githubusercontent.com/google/fonts/).
[COMMAND_EXECUTION]: The scripts/render.mjs script utilizes execFileSync to run the local scripts/gz-theme.mjs utility for processing brand-specific color themes.
[REMOTE_CODE_EXECUTION]: The scripts/export.mjs script launches a headless browser instance using Puppeteer with the --no-sandbox and --allow-file-access-from-files flags. These settings bypass standard browser security boundaries, which can be exploited to read local files if the browser is directed to load malicious or untrusted content.
[INDIRECT_PROMPT_INJECTION]: The rendering logic in scripts/render.mjs processes annotations (intended to be generated by an AI agent) and injects raw SVG and HTML content into the final document without sanitization. This creates a Cross-Site Scripting (XSS) vulnerability surface.
Ingestion points: scripts/render.mjs reads the input Markdown file and an annotations.json file.
Boundary markers: Absent for the embed component's svg and html fields in render.mjs.
Capability inventory: File system writes via writeFileSync, subprocess execution via execFileSync, and browser-based rendering/export via puppeteer-core.
Sanitization: Partial; while text labels are escaped using escHtml, the svg and html fields in annotations.json are interpolated directly into the output.

guizang-page-skill