The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/install_deps.sh (sourced by setup.sh and various document automation tools) uses sudo to acquire administrative privileges for installing system packages across multiple platforms (apt, brew, dnf, apk, pacman). This allows the skill to execute commands with root permissions to install heavy dependencies like texlive, poppler, and imagemagick.
[EXTERNAL_DOWNLOADS]: Multiple Python scripts (csv_to_latex.py, generate_chart.py, pdf_fill_form.py, etc.) contain a helper function _ensure_package that automatically downloads and installs missing Python libraries from PyPI using pip at runtime. These installations are unversioned and occur without explicit user confirmation during the document processing flow. Additionally, scripts/plantuml_to_pdf.sh fetches a JAR file from GitHub if not found locally, and scripts/fetch_bibtex.sh performs network requests to doi.org and arxiv.org.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its PDF-to-LaTeX conversion pipeline.
Ingestion points: Page images extracted from user-provided PDFs via pdf_to_images.sh are processed by a vision-capable AI agent as described in references/pdf-conversion.md.
Boundary markers: Absent; there are no specific delimiters or instructions for the agent to ignore commands within the source data in the provided prompts.
Capability inventory: The skill can execute arbitrary shell commands via the Bash scripts, write to the local filesystem, and perform network operations via curl and wget.
Sanitization: Absent; while LaTeX special characters are escaped for formatting, there is no filtering or validation of the semantic content extracted from source documents.
[COMMAND_EXECUTION]: The skill uses dynamic execution patterns, including __import__ calls for runtime module loading in several Python scripts and a monkey-patch of the pypdf library in scripts/pdf_fill_form.py to modify its internal behavior at runtime.

latex-document