The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation provides instructions for installing several well-known system utilities and libraries necessary for document manipulation, including 'pandoc', 'libreoffice', 'poppler-utils', the 'docx' npm package, and the 'defusedxml' Python library. All referenced sources are established industry standard repositories.
[COMMAND_EXECUTION]: The skill documentation includes examples of environment setup using 'sudo apt-get' to install the required system dependencies. Additionally, the skill's Python scripts utilize subprocess calls to perform legitimate local tasks, such as document validation via 'soffice' and revision comparisons using 'git diff'. These operations are tied directly to the skill's primary purpose.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it processes untrusted data from external .docx files.
Ingestion points: Untrusted data enters the agent context when documents are converted to Markdown (e.g., current.md) or when raw XML files are read after unpacking.
Boundary markers: The instructions do not explicitly mandate the use of delimiters or 'ignore embedded instructions' prompts when the agent reads the extracted document content.
Capability inventory: The skill provides robust local file system access, zip extraction, and subprocess execution capabilities (scripts/document.py, pack.py).
Sanitization: There are no documented steps for sanitizing or filtering instructions that might be embedded in the text of documents provided by users.

docx