latex-translation-guide
Pass
Audited by Gen Agent Trust Hub on Apr 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Recommends the installation of the
latextransandmathtranslatepackages via pip to enable automated translation functionality. - [EXTERNAL_DOWNLOADS]: References the
arxiv-latex-cleanerutility and documentation from Google Research's official public repository. - [DATA_EXFILTRATION]: The skill documents the use of external translation backends (Google Translate, DeepL, OpenAI, Claude), which involves sending document text to third-party services for processing.
- [PROMPT_INJECTION]: Detects a vulnerability surface for indirect prompt injection where malicious instructions inside a processed LaTeX document could influence the behavior of the translation engine.
- Ingestion points: Reads content from local
.texfiles (paper.tex) and downloaded arXiv documents (translate_arxiv) as shown in SKILL.md. - Boundary markers: Absent; while math environments are protected by placeholders (e.g.,
__MATH_0__), there are no delimiters or instructions to the translation LLM to ignore embedded commands in the text. - Capability inventory: Includes file system access (read/write), shell command execution (
latextrans,translate_arxiv), and network operations to translation APIs. - Sanitization: Absent; the provided scripts do not perform validation or sanitization of the text content to prevent the execution of embedded instructions by the LLM backend.
- [COMMAND_EXECUTION]: Provides instructions for executing shell commands to install software and perform document translation tasks.
Audit Metadata