The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted user-provided source material (such as PDFs or Markdown files) to generate HTML artifacts, which creates a potential surface for indirect prompt injection.
Ingestion points: Phase 1 utilizes the Read tool to ingest arbitrary source content from user-specified files.
Boundary markers: The instructions do not specify the use of delimiters or directives to ignore instructions that might be embedded within the ingested data.
Capability inventory: The skill uses the Bash tool for image processing, file validation, and PDF export, as well as the Write tool to save the generated HTML file.
Sanitization: Ingested content is distilled into the HTML output without explicit sanitization or filtering of potential instructions.
[COMMAND_EXECUTION]: The skill executes several shell commands to facilitate its workflow, including image optimization via cwebp, encoding via base64, and file validation via ls, python3 (utilizing the standard html.parser module), and grep. It also includes a local utility script, scripts/export-pdf.sh, which launches a headless browser (Chromium/Chrome) to generate PDF exports of the created artifacts.
[EXTERNAL_DOWNLOADS]: The generated HTML pages are configured to load fonts and libraries from well-known content delivery networks, including cdn.jsdelivr.net, unpkg.com, cdnjs.cloudflare.com, and Google Fonts domains. These references are used for intended functionality and target established service providers.

rich-page