skills/yy/claude-scholar/arxiv-prep/Gen Agent Trust Hub

arxiv-prep

Pass

Audited by Gen Agent Trust Hub on May 6, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill uses uvx to download and execute arxiv-latex-cleaner and bib_optimizer from the Python Package Index (PyPI). arxiv-latex-cleaner is a recognized tool maintained by the Google Research organization.
  • [COMMAND_EXECUTION]: The skill executes several shell commands to process files locally. This includes pdflatex for document compilation, tar for creating submission packages, and bibtool for bibliography management. These are standard utilities for LaTeX document preparation.
  • [DATA_EXFILTRATION]: No network exfiltration patterns were detected. All file operations (cleaning, merging, and packaging) are performed on a local copy (_arXiv/ directory) of the user's paper. The skill only provides a URL to the official arXiv submission site for the user to manually upload the final package.
  • [PROMPT_INJECTION]: The skill extracts metadata such as titles and abstracts from user-provided LaTeX files. While this involves ingesting untrusted data, the processing is limited to extracting specific fields for display to the user, presenting a minimal surface for indirect prompt injection.
Audit Metadata
Risk Level
SAFE
Analyzed
May 6, 2026, 02:28 AM
Security Audit — agent-trust-hub — arxiv-prep