xlsx
Warn
Audited by Gen Agent Trust Hub on Mar 9, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The script 'scripts/office/soffice.py' contains an embedded C source string that is written to a temporary directory and compiled at runtime using 'gcc'. The resulting shared library is loaded via the 'LD_PRELOAD' environment variable when invoking LibreOffice. This mechanism serves as a technical workaround for communication in restricted sandbox environments.
- [COMMAND_EXECUTION]: Subprocess calls are used to execute 'soffice' for spreadsheet operations, 'gcc' for compiling the compatibility shim, and 'git' for generating text diffs during document validation.
- [PROMPT_INJECTION]: The skill processes untrusted user spreadsheet files, which represents a surface for indirect prompt injection. 1. Ingestion points: Spreadsheet data enters the agent context through pandas.read_excel and openpyxl.load_workbook. 2. Boundary markers: No explicit delimiters or instructions are used to wrap untrusted data. 3. Capability inventory: The skill can execute system commands and perform file system operations. 4. Sanitization: The library 'defusedxml' is used to mitigate XML External Entity (XXE) vulnerabilities during file parsing.
Audit Metadata