The Agent Skills Directory

[COMMAND_EXECUTION]: The pdf_to_images.py and validate_presentation.py scripts execute system utilities including pdflatex, pdftoppm (part of poppler-utils), and convert (ImageMagick). These operations are implemented using subprocess.run() with list-based arguments, which is the recommended secure method for calling external binaries as it prevents shell injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied PDF, PowerPoint, and LaTeX files. This constitutes an attack surface for indirect prompt injection via document content or metadata. However, the skill includes specific operational instructions for the AI agent to handle these files safely (e.g., converting PDFs to images to avoid context window issues) and uses standard libraries for file parsing.
Ingestion points: Ingestion occurs when the scripts in the scripts/ directory load .pdf, .pptx, or .tex files provided by the user.
Boundary markers: No explicit boundary markers are used for the content of the processed documents.
Capability inventory: The skill has the capability to write files to the disk (converted images), read document contents, and execute system tools for document processing.
Sanitization: The scripts prevent command injection by passing arguments to subprocess.run as a list rather than a shell string.
[DYNAMIC_EXECUTION]: The validate_presentation.py script performs runtime compilation of LaTeX source code using pdflatex. While LaTeX compilation is technically a form of dynamic execution, it is an intended and documented feature of the skill required to support the Beamer presentation format.

scientific-slides