The Agent Skills Directory

[PROMPT_INJECTION]: The documentation in SKILL.md, docx-js.md, and ooxml.md contains steering instructions (e.g., "MANDATORY
READ ENTIRE FILE", "NEVER set any range limits") that attempt to dictate the agent's operational behavior and tool usage parameters to ensure documentation is processed in its entirety.
[PROMPT_INJECTION]: The skill processes untrusted user-provided office documents, creating a potential surface for indirect prompt injection attacks where malicious content in a document could influence agent behavior during analysis.
Ingestion points: XML content is unpacked and parsed in ooxml/scripts/unpack.py and scripts/document.py.
Boundary markers: No specific delimiters or "ignore instructions" warnings are utilized when presenting extracted markdown to the agent.
Capability inventory: The skill can execute system commands via subprocess.run (e.g., soffice, git, pandoc) and perform file system operations.
Sanitization: Uses the defusedxml library to mitigate XML-based attacks like XXE.
[COMMAND_EXECUTION]: Documentation provides setup instructions that include running commands with sudo for installing system dependencies like libreoffice and pandoc.
[EXTERNAL_DOWNLOADS]: The skill requires external packages from standard registries, including the docx npm package and various Python libraries.

docx