The Agent Skills Directory

[DYNAMIC_EXECUTION]: The assets/preview.html file implements a live-reload system that dynamically injects and executes code from a local variants.js file every 3 seconds using document.createElement('script') and document.head.appendChild(). This constitutes dynamic loading from a computed path.
[DYNAMIC_EXECUTION]: The preview system in assets/preview.html is susceptible to cross-site scripting (XSS). The renderCard and sizeRamp functions interpolate variant data fields (specifically image paths) directly into <img> tags without adequate sanitization, potentially allowing arbitrary JavaScript execution if malicious filenames are processed.
[COMMAND_EXECUTION]: The references/editing-workflow.md file instructs the agent to use shell commands (open or xdg-open) to launch a web browser and display local files.
[EXTERNAL_DOWNLOADS]: The skill recommends the global installation of the svgo package via npm for SVG processing tasks.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted SVG data and provides instructions to the agent to inspect these files for malicious attributes or scripts, recognizing the attack surface for indirect prompt injection. 1. Ingestion points: User-provided SVG assets referenced in references/svg-basics.md and SKILL.md workflows. 2. Boundary markers: None. 3. Capability inventory: Local file creation/modification and browser execution via CLI. 4. Sanitization: Instructions recommend manual inspection of SVG XML content for scripts, event attributes, and data URLs.

svg-design