The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data (Markdown, HTML, and report data) and renders it using a headless browser (Puppeteer) or PDF libraries. This configuration presents an indirect prompt injection surface where malicious instructions embedded in the source data could potentially influence agent behavior or attempt SSRF (Server-Side Request Forgery) via the browser interface. * Ingestion points: Processes files from local paths (mdPath, inputPath) and structured objects (data parameter). * Boundary markers: No explicit delimiters or boundary markers are defined to isolate untrusted data from the processing logic or the browser context. * Capability inventory: Includes file system read/write access (fs) and the ability to launch a headless browser (puppeteer) which can access network resources. * Sanitization: The implementation does not include sanitization, validation, or escaping of the input content before it is parsed and rendered.
[COMMAND_EXECUTION]: The skill utilizes the puppeteer library which launches a headless Chromium instance to render PDF documents from HTML content. While this is a standard industry practice for PDF generation, it involves spawning external browser processes which increases the system's attack surface.
[EXTERNAL_DOWNLOADS]: The skill's documentation and dependencies section recommend installing external Node.js packages (puppeteer, pdfkit, pdf-lib, marked) from the npm registry. These are well-known and widely used libraries for the described purposes.

PDF Generator