The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess.run and subprocess.Popen to manage external binaries for its core functionality. These include pandoc for document conversion, google-chrome for high-fidelity PDF rendering, and the poppler-utils suite (pdftoppm, pdftotext, pdfinfo) for generating visual previews and performing typography linting. The commands are constructed using list-based arguments, which is a secure practice that prevents shell injection vulnerabilities.
[DYNAMIC_EXECUTION]: On macOS systems, the core script dynamically modifies the DYLD_LIBRARY_PATH environment variable to ensure that the weasyprint library can locate its required dependencies in Homebrew directories. Furthermore, internal test scripts demonstrate dynamic code execution patterns, such as modifying sys.path or assembling Python commands for execution via subprocess to verify rendering logic.
[INDIRECT_PROMPT_INJECTION]: The skill processes external Markdown content and renders it through an HTML-to-PDF pipeline. This workflow creates an inherent attack surface for indirect prompt injection if malicious HTML or JavaScript is embedded in the source Markdown files and processed by the rendering backends without strict sandboxing.
Ingestion points: Markdown files provided by the user are read and processed in scripts/md_to_pdf.py and scripts/batch_convert.py.
Boundary markers: The skill does not employ specific delimiters or "ignore instructions" markers to isolate processed data from the agent's internal logic.
Capability inventory: Across its scripts, the skill has the capability to perform shell command execution, file system operations, and potentially network requests if the Chrome backend is triggered to load remote assets.
Sanitization: The skill relies on the standard parsing behavior of pandoc and does not apply additional sanitization or filtering to the HTML content before it is passed to the rendering engines.

pdf-creator