tikz-figures
Pass
Audited by Gen Agent Trust Hub on Sep 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes several Python scripts that use
subprocess.runto execute standard technical utilities such asxelatex(for LaTeX compilation),dot(Graphviz, for layout calculation), andpdftoppm(for PDF-to-image conversion). These executions use list-based arguments without a shell, minimizing injection risks, and are essential for the primary purpose of the skill. - [EXTERNAL_DOWNLOADS]: The workflow documentation (
SKILL.md) instructs the use ofuv runto manage Python dependencies. This tool fetches well-known packages likepdfplumber,pymupdf(fitz), andopencv-python-headlessfrom the official Python Package Index (PyPI) registry. These downloads are standard for environment setup and are performed from a well-known service. - [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied figure descriptions to generate TikZ code and refined layouts. It features a complex "visual feedback loop" where output from geometric analysis scripts (in JSON format) is ingested back into the agent's context. While this data path represents an attack surface, the skill emphasizes human-in-the-loop oversight and systematic triaging of detected issues to prevent accidental obedience to any instructions embedded within processed data.
Audit Metadata