latex-document

Warn

Audited by Gen Agent Trust Hub on Jun 13, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill relies on several local shell and Python scripts (e.g., compile_latex.sh, generate_chart.py, mermaid_to_image.sh) to perform core document processing tasks.
  • [COMMAND_EXECUTION]: The compilation workflow is designed to auto-install system packages such as texlive and poppler-utils if they are not present, a process that typically involves administrative (sudo) access.
  • [REMOTE_CODE_EXECUTION]: The inclusion of Jinja2 templating for mail merges and support for LuaLaTeX introduces dynamic code execution capabilities during the rendering process.
  • [EXTERNAL_DOWNLOADS]: The skill fetches citation data from the well-known doi.org service and downloads system utilities from official package repositories during setup.
  • [PROMPT_INJECTION]: Processing untrusted external data from PDFs, CSVs, and JSON files for interpolation into LaTeX templates presents a surface for indirect prompt injection.
  • Ingestion points: PDF documents, CSV/JSON data sources, and handwritten notes processed via OCR.
  • Boundary markers: No specific delimiters or "ignore" instructions are documented for the data ingestion workflows.
  • Capability inventory: The skill possesses extensive local execution capabilities, file system write access, and network connectivity (DOI fetch).
  • Sanitization: There is no explicit evidence of sanitization or validation of the extracted external content before it is processed by the templates.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 13, 2026, 11:31 AM
Security Audit — agent-trust-hub — latex-document