xlsx

Warn

Audited by Gen Agent Trust Hub on Aug 26, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill performs runtime compilation and process injection.
  • File: scripts/office/soffice.py contains logic in the _ensure_shim function that writes a string of C source code to a temporary file (lo_socket_shim.c), compiles it into a shared library (lo_socket_shim.so) using a gcc subprocess, and then injects this library into the LibreOffice environment using the LD_PRELOAD environment variable.
  • Risk: While documented as a workaround for socket restrictions in restricted environments, runtime compilation and library injection are high-risk patterns that could be used for process hijacking.
  • [COMMAND_EXECUTION]: The skill executes multiple external system commands via shell subprocesses.
  • File: recalc.py and scripts/recalc.py execute soffice (LibreOffice) and system utilities like timeout or gtimeout.
  • File: scripts/office/validators/redlining.py executes git diff to perform document comparisons.
  • File: scripts/office/soffice.py executes gcc to compile runtime shims.
  • [PERSISTENCE_MECHANISMS]: The skill modifies local application configuration files to install persistent logic.
  • File: recalc.py and scripts/recalc.py configure the local LibreOffice installation by creating or overwriting a macro file (Module1.xba) in the user's basic scripts directory to facilitate automatic recalculation.
  • [INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for processing untrusted external data.
  • Ingestion points: The skill processes user-supplied spreadsheet files (XLSX, CSV, TSV) and document components (DOCX, PPTX).
  • Boundary markers: Absent. The skill extracts and interpolates data from XML components without explicit delimiters or warnings to ignore embedded instructions.
  • Capability inventory: The skill possesses high-privilege capabilities including arbitrary command execution, file system modification, and runtime code compilation.
  • Sanitization: Incomplete. While some components use defusedxml to mitigate XML risks, other parts of the validator suite use lxml or xml.etree.ElementTree without visible protection against XML External Entity (XXE) or other data-driven attacks.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 26, 2026, 04:32 PM