The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting content from a project's index.html file and embedding it into a generated overview.html page.
Ingestion points: The generation process in SKILL.md and template.html extracts :root CSS variables, scene-specific CSS styles, and scene HTML structure from the user-provided index.html file to populate the {{SCENE_VARS}}, {{SCENE_STYLES}}, and {{SLIDES_HTML}} placeholders.
Boundary markers: The template.html lacks isolation boundaries (such as sandboxed iframes) or specific delimiters to prevent potentially malicious code within the injected blocks from interacting with the viewer's JavaScript logic.
Capability inventory: The skill includes a shell script (serve.sh) for local command execution and JavaScript in the viewer that performs local network requests (fetch('/save')) and interacts with the system clipboard.
Sanitization: While the instructions mandate the removal of specific timing attributes and animation classes, they do not prescribe comprehensive sanitization (e.g., removing <script> tags) of the HTML and CSS content being transferred.
[COMMAND_EXECUTION]: The serve.sh script executes a shell script located in a sibling skill directory (../hyperframes-overview-edit/serve-live.sh). This facilitates local execution of logic defined in the associated edit module.
[EXTERNAL_DOWNLOADS]: The template.html file references external typography assets from Google Fonts (fonts.googleapis.com and fonts.gstatic.com), which are established and trusted services.

hyperframes-overview