icml-review
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes external PDF utilities (pdfinfo, pdffonts, pdftotext) via subprocess.run in scripts/check_submission.py. These calls use argument lists rather than shell strings, which is a secure practice to prevent shell injection while performing file analysis.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted research papers (LaTeX and PDF source files) which could potentially contain hidden instructions. It proactively mitigates this risk by including specialized detection logic in scripts/check_submission.py to identify and flag hidden text (such as white-on-white text or near-zero font sizes) and common prompt injection keywords designed to manipulate LLM-based reviewers.
- [DYNAMIC_EXECUTION]: The script scripts/check_submission.py performs runtime execution of command-line tools for PDF parsing and text extraction. This is a functional requirement for auditing compiled documents and relies on standard system utilities (poppler-utils).
Audit Metadata