The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script setup_fonts.py fetches Open Font License (OFL) typography assets from Google's official GitHub repository. This is a standard and safe operation to provide the fonts required for the design replication process.
[COMMAND_EXECUTION]: The shared utility module _common.py invokes a sibling script (asset-generator) using subprocess.run(). The arguments are passed as a list rather than a concatenated string, which is a security best practice that prevents shell command injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied PDF documents, which introduces a surface for indirect prompt injection where text inside a PDF could attempt to influence the agent's behavior.
Ingestion points: User-provided PDF files are parsed in analyze_pdf.py.
Boundary markers: The vision model prompts include specific constraints to return only structured JSON data.
Capability inventory: The skill maintains capabilities for local script execution, file system writes (for assets and the final PDF), and network requests via Gemini and Playwright.
Sanitization: The build_pdf.py script applies html.escape to all text content extracted from the reference document before it is interpolated into the HTML template for final PDF generation.

pdf-remix