The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The script ooxml/scripts/pack.py uses subprocess.run to execute the soffice (LibreOffice) binary for document validation. This is a legitimate operation for the skill's primary function of processing OOXML files.
Evidence: subprocess.run(["soffice", "--headless", ...]) in validate_document function.
Security Note: The command is constructed using a list (not a shell string) with a specific, local binary, minimizing risks associated with shell injection.
[PROMPT_INJECTION] (LOW): The skill presents an indirect prompt injection surface as it processes external, potentially untrusted Office documents provided by the user.
Ingestion points: ooxml/scripts/unpack.py extracts contents from user-provided Office files.
Boundary markers: Absent. The scripts process raw XML files from the document archives.
Capability inventory: File system access (zipfile, shutil), command execution (soffice), and XML parsing.
Sanitization: The skill uses defusedxml in its main processing scripts to mitigate XML-based attacks. However, ooxml/scripts/validation/docx.py uses standard lxml.etree.parse, which is a best-practice violation regarding XXE (XML External Entity) protection.
[SAFE] (SAFE): The skill is developed by a trusted author (Anthropic) and performs its stated tasks using industry-standard libraries. No indicators of data exfiltration, obfuscation, or persistence were found.

docx