The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The file ooxml/scripts/pack.py contains a validate_document function that uses subprocess.run to execute the soffice (LibreOffice) command. While this is used for validating the integrity of generated Office files, spawning subprocesses to handle untrusted data paths represents a potential security risk if the environment is not properly isolated.
EXTERNAL_DOWNLOADS (LOW): The skill depends on several external Python libraries including lxml, defusedxml, python-pptx, and six. While these are standard packages, they must be verified in the target environment.
SAFE (SAFE): The implementation demonstrates security awareness by using defusedxml.minidom in ooxml/scripts/pack.py and ooxml/scripts/unpack.py instead of the standard xml.dom.minidom, which effectively mitigates XML External Entity (XXE) vulnerabilities.
[Indirect Prompt Injection] (LOW): The skill processes untrusted Office documents which can serve as a vector for indirect prompt injection if the agent interprets text content from these files as instructions.
Ingestion points: ooxml/scripts/unpack.py (unpacks arbitrary ZIP/Office files) and scripts/rearrange.py (reads PPTX templates).
Boundary markers: Absent. The scripts do not implement specific delimiters or warnings for the agent when processing document text.
Capability inventory: File system write access (zipfile.write), subprocess execution (soffice conversion).
Sanitization: The skill uses defusedxml to sanitize the XML structure against parser-level attacks, but it does not sanitize the natural language content for embedded instructions.

pptx