The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains a dangerous pattern in scripts/setup.sh and routes/latex.md where a remote shell script is downloaded via curl and piped directly into sh (curl -fsSL https://drop-sh.fullyjustified.net | sh). This allows for arbitrary code execution from a non-whitelisted external domain.
[EXTERNAL_DOWNLOADS]: The skill performs several unpinned or external downloads at runtime:
The scripts/compile_latex.py script attempts to automatically install the pypdf library using pip install if it is not found.
The scripts/pdf.sh script attempts to install playwright (Node.js) and its Chromium browser binaries, as well as pikepdf and pdfplumber (Python) during its fix command.
The Tectonic LaTeX engine is downloaded from an external site as part of the setup process.
[COMMAND_EXECUTION]: The skill makes extensive use of shell command execution to perform its primary functions:
scripts/compile_latex.py uses subprocess.run to execute the tectonic compiler.
scripts/cmd_convert.py uses subprocess.run to call LibreOffice (soffice) for document conversion.
scripts/html_to_pdf.js uses Playwright to automate browser instances for PDF rendering.
[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to process external PDF and HTML files. Scripts like scripts/cmd_extract.py extract text, tables, and images from potentially untrusted user-provided PDFs and feed this content back to the AI agent. There are no explicit sanitization steps or boundary markers described to prevent the agent from obeying instructions embedded within these processed files.

pdf