The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to call system tools soffice (from LibreOffice) and git. These invocations are implemented securely using a list of arguments without shell interpretation, minimizing command injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and modify external .docx files, which constitutes a typical attack surface. 1. Ingestion points: XML data from document files is unpacked and read into the agent's context using ooxml/scripts/unpack.py and scripts/document.py. 2. Boundary markers: The skill relies on XML tags as delimiters but does not include specific system instructions to prevent the model from accidentally obeying natural language commands embedded in the document text. 3. Capability inventory: The skill has the ability to write files and execute system commands for packing and validation purposes. 4. Sanitization: The skill employs the defusedxml library for parsing, which protects against common XML vulnerabilities like XXE, although it does not sanitize document text for prompt injection patterns.

docx