The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The pack.py script invokes the soffice (LibreOffice) command-line utility via subprocess.run. Although it uses the list format for arguments, executing a complex document processor on untrusted file paths is a significant capability surface.
PROMPT_INJECTION (HIGH): The skill possesses a high-risk attack surface for indirect prompt injection (Category 8). 1. Ingestion points: ooxml/scripts/unpack.py extracts content from external, untrusted OOXML (ZIP) documents. 2. Boundary markers: No delimiters or ignore-instructions are present to prevent the agent from being influenced by data found within the documents. 3. Capability inventory: The skill has file-write permissions (pack.py, unpack.py) and command-execution capabilities (pack.py). 4. Sanitization: While the developer used defusedxml in several scripts to mitigate XML attacks, ooxml/scripts/validation/docx.py uses lxml.etree.parse without explicit safeguards against XML External Entities (XXE). Furthermore, the use of zipfile.extractall() on untrusted archives in unpack.py presents a directory traversal risk.

docx