docx

Fail

Audited by Gen Agent Trust Hub on Aug 19, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The script scripts/office/soffice.py dynamically creates C source code, compiles it into a shared library using gcc, and then uses LD_PRELOAD to inject that library into the LibreOffice process to intercept standard library system calls.
  • [COMMAND_EXECUTION]: Multiple Python scripts perform arbitrary command execution using the subprocess module, invoking system tools like gcc, soffice, git, pandoc, and pdftoppm with arguments influenced by document content.
  • [REMOTE_CODE_EXECUTION]: scripts/accept_changes.py dynamically generates LibreOffice StarBasic macros and installs them into a temporary profile directory to automate document processing.
  • [PROMPT_INJECTION]: The skill ingests untrusted XML content from Word documents and processes it using sensitive utilities without robust sanitization, adequate boundary markers, or hardened XML parsers, creating a surface for indirect prompt injection and XXE-related attacks.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Aug 19, 2026, 07:46 PM
Security Audit — agent-trust-hub — docx