The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (MEDIUM): Potential 'Zip Slip' (path traversal) vulnerability detected in ooxml/scripts/unpack.py and ooxml/scripts/validation/docx.py.
Evidence: The code calls zipfile.ZipFile(...).extractall(output_path) on user-provided Office files.
Risk: If an attacker provides a specially crafted ZIP/DOCX archive containing filenames with path traversal sequences (e.g., ../../target), it may allow overwriting arbitrary files on the local system outside the intended directory.
[COMMAND_EXECUTION] (LOW): Execution of external binaries via subprocess in ooxml/scripts/pack.py.
Evidence: The function validate_document executes soffice (LibreOffice) to perform headless conversion for validation.
Risk: While it uses a list-based argument structure to prevent shell injection, processing untrusted file paths through complex external software increases the attack surface for the host environment.
[DATA_EXFILTRATION] (LOW): Potential XML External Entity (XXE) surface in ooxml/scripts/validation/docx.py.
Evidence: The script uses lxml.etree.parse(str(xml_file)) to process document components.
Risk: Unlike other modules in the skill that correctly use defusedxml, this module uses the standard lxml parser without explicitly disabling network or entity resolution, which could be exploited to read local files if a malicious XML is processed.

pptx