ml-paper-writing

Warn

Audited by Gen Agent Trust Hub on Feb 18, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • EXTERNAL_DOWNLOADS (MEDIUM): The file templates/neurips2025/Makefile includes an upgrade target that downloads a ZIP archive from media.neurips.cc using curl. This domain is not included in the list of trusted external sources, making the integrity of the downloaded content unverifiable through static analysis.
  • COMMAND_EXECUTION (LOW): The Makefile in templates/neurips2025/ is designed to execute shell commands (curl, unzip, mv, rm) to update local LaTeX style files. While standard for build systems, these operations should be monitored when triggered by an agent.
  • PROMPT_INJECTION (LOW): The skill documentation (arxiv-search-guide.md) and data structures (kaiming_he_injection_record.json) describe automated workflows for searching, extracting, and analyzing research papers from arXiv. This ingestion of external, untrusted content presents an indirect prompt injection surface.
  • Ingestion points: arXiv abstract pages, full paper content, and paper metadata extracted via Chrome MCP.
  • Boundary markers: None identified in the provided templates or guides.
  • Capability inventory: Automated pattern mining, text extraction, and knowledge base updating (writing to structure.md, writing-techniques.md).
  • Sanitization: No explicit sanitization or validation steps are documented for the incoming research data.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 18, 2026, 02:52 PM