The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The script ooxml/scripts/pack.py executes the soffice (LibreOffice) binary using subprocess.run to validate document conversion. While it uses a list for arguments to mitigate shell injection, invoking complex third-party parsers on generated files is a known attack surface.
DATA_EXFILTRATION (MEDIUM): In ooxml/scripts/validation/docx.py, the skill uses lxml.etree.parse() to process XML components of Office documents. Unlike the packing/unpacking scripts which utilize defusedxml, the validation module lacks explicit protections against XML External Entity (XXE) attacks. If an attacker provides a crafted document, this could lead to local file exposure.
PROMPT_INJECTION (LOW): The skill exhibits surface area for indirect prompt injection (Category 8) as it parses and processes content from external Office documents.
Ingestion points: Untrusted XML data is read in ooxml/scripts/unpack.py and ooxml/scripts/validation/docx.py.
Boundary markers: No specific delimiters or "ignore instructions" warnings were found when interpolating document data.
Capability inventory: The skill possesses the ability to execute external commands (soffice) and perform file system read/write operations.
Sanitization: While defusedxml is used in pack.py and unpack.py, the lxml parser in the validation logic lacks security hardening against malicious entities.

docx