The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a publishing utility at scripts/publish.ts designed to be executed via the Bun runtime. This script performs file system operations, including reading a user-provided HTML file and writing a generated React component to a temporary workspace path.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface through its processing of untrusted HTML data. The scripts/publish.ts script extracts HTML, CSS, and JavaScript from a source file and injects it into a new React component using dangerouslySetInnerHTML. This mechanism does not perform sanitization or validation of the extracted content, potentially allowing malicious scripts within a presentation file to be deployed and executed on the user's hosted domain.
Ingestion points: Reads local HTML presentation files via the --file argument in scripts/publish.ts.
Boundary markers: None identified.
Capability inventory: File system read and write access, and integration with a web deployment tool (update_space_route).
Sanitization: Employs basic escaping for template literal syntax but lacks sanitization for HTML tags, attributes, or embedded JavaScript code.
[REMOTE_CODE_EXECUTION]: The presentation boilerplate and the publishing script generate code that dynamically loads external libraries—including Reveal.js, Chart.js, and various optional animation or visualization packages—from public CDNs such as JSDelivr and UNPKG. While these are well-known services, the skill's reliance on remote code execution in the browser for its core features is a documented behavior.

revealjs-presentation