The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use sudo for environment setup, which involves executing system commands with administrative privileges. Evidence: SKILL.md contains sudo apt-get install commands for pandoc, libreoffice, and poppler-utils.\n- [COMMAND_EXECUTION]: The skill executes external command-line utilities via subprocess.run to perform core tasks like document extraction, validation, and diffing. Evidence: SKILL.md (pandoc), ooxml/scripts/pack.py (soffice), and ooxml/scripts/validation/redlining.py (git diff).\n- [EXTERNAL_DOWNLOADS]: Runtime software installation from public package registries is required for skill functionality. Evidence: pip install defusedxml and npm install -g docx in SKILL.md.\n- [PROMPT_INJECTION]: The skill maintains an attack surface for indirect prompt injection as it processes untrusted content from Word documents while having access to system-level tools.\n
Ingestion points: ooxml/scripts/unpack.py and ooxml/scripts/validation/docx.py (reading XML from unzipped Word documents).\n
Boundary markers: Absent; extracted document text is not encapsulated with instructions to ignore embedded prompts.\n
Capability inventory: High-risk command execution and software installation capabilities.\n
Sanitization: While defusedxml is used for major XML operations, ooxml/scripts/validation/base.py uses standard lxml.etree.parse which may be vulnerable to XML-based attacks if malicious documents are provided.

docx