The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The utility script ooxml/scripts/unpack.py uses zipfile.extractall() to decompress Office documents. This function does not natively prevent path traversal; if an archive contains members with absolute paths or parent directory sequences (e.g., ../../), it can write files to locations outside the intended output directory. This Zip Slip vulnerability could allow a malicious presentation to overwrite user scripts or configuration files in the agent's workspace.
[COMMAND_EXECUTION]: The skill uses subprocess.run to call system binaries including soffice, pdftoppm, and git. These calls utilize list-based arguments which mitigate common shell injection risks. These commands are necessary for document conversion, thumbnail generation, and change-tracking validation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted input in the form of PowerPoint and HTML files. While core components use defusedxml to prevent XML-based attacks, the validation routines in ooxml/scripts/validation/ utilize lxml and standard ElementTree, which may be vulnerable to XML External Entity (XXE) attacks if not properly configured. This represents an attack surface where malicious presentation content could influence agent execution or attempt local data access.

pptx