The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied content to generate HTML documents, which are subsequently rendered using a headless browser (Playwright) to create PNG previews. This workflow creates an attack surface where maliciously crafted input could attempt to influence the rendering environment.
Ingestion points: User-provided text is interpolated into HTML templates in assets/card-template.html and references/html-skeleton.md.
Boundary markers: The instructions include guidelines for Mermaid syntax (e.g., quoting labels with slashes) but do not define security boundaries or instructions to ignore embedded commands within the processed data.
Capability inventory: The skill possesses capabilities for filesystem writes (mktemp), local script execution (node scripts/render_report.mjs), and headless browser operations via Playwright.
Sanitization: The instructions do not specify HTML sanitization or output encoding for user-supplied data before it is rendered by the browser.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute local Node.js scripts (lint_mermaid.js and render_report.mjs) to validate and render the generated artifacts.
[EXTERNAL_DOWNLOADS]: The skill fetches resources from well-known services to provide styling, typography, and diagramming capabilities. Evidence: The HTML templates reference cdn.tailwindcss.com, fonts.googleapis.com, and cdn.jsdelivr.net (for Mermaid.js). The render_report.mjs script utilizes the Playwright library, which requires the installation of browser binaries.
[DYNAMIC_EXECUTION]: The skill generates HTML files from user-supplied templates and executes dynamic logic in a browser environment via Playwright's page.evaluate and page.waitForFunction to ensure correct diagram rendering.

hai-visual-explainer