The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from arbitrary external URLs provided by the user, which presents a surface for indirect prompt injection.
Ingestion points: External website content is ingested via playwright-cli during the extraction of CSS styles, SVG logos, and font files.
Boundary markers: None explicitly implemented for the browser-side extraction logic, though the resulting data is mapped into a fixed JSON schema.
Capability inventory: The skill writes files to the local assets/ and docs/ directories and executes Node.js scripts.
Sanitization: The skill uses a dedicated decoding script (decode-binary-assets.cjs) that provides extensive sanitization for file paths and content, preventing malicious input from affecting the host filesystem.
[EXTERNAL_DOWNLOADS]: The skill documentation guides the user through installing @playwright/cli from the official NPM registry. This is a well-known package maintained by Microsoft and is considered a safe dependency.
[DYNAMIC_EXECUTION]: The skill executes browser-side JavaScript via playwright-cli eval and a local Node.js script (decode-binary-assets.cjs) to handle asset processing.
The decode-binary-assets.cjs script includes significant security hardening: it validates that output paths do not escape the target directory, checks for path traversal characters, and verifies that neither the target file nor its parent directories are symbolic links (O_NOFOLLOW). It also includes a limit on the number of files written to prevent resource exhaustion.
[COMMAND_EXECUTION]: The skill uses playwright-cli for browser automation and standard system commands like mkdir or New-Item to manage output directories. These are standard operations for the skill's purpose.

create-design-system