The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes content from untrusted .docx and .doc files which could contain hidden instructions for the agent.\n
Ingestion points: Content is extracted from documents via pandoc and XML unpacking in SKILL.md.\n
Boundary markers: There are no instructions to use delimiters or ignore embedded commands when the agent reads the extracted text.\n
Capability inventory: The skill has the ability to execute system commands (soffice, pandoc, pdftoppm), run local scripts, and install NPM packages.\n
Sanitization: No sanitization or validation of the extracted document text is performed before it reaches the agent context.\n- [COMMAND_EXECUTION]: The skill relies on various system binaries and local scripts to perform its functions.\n
Executes system utilities including pandoc, LibreOffice (soffice), and Poppler (pdftoppm) for document analysis and conversion.\n
Invokes several local Python scripts (unpack.py, pack.py, validate.py, comment.py) to manage Word XML structures.\n
Uses npm to install the docx library globally as part of the setup process.\n- [EXTERNAL_DOWNLOADS]: The skill requires downloading external software and libraries to operate.\n
Instructions command the global installation of the docx package from the NPM registry.\n
References external system dependencies like LibreOffice and Poppler for specific workflow stages.

docx