The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external documents (PDF, DOCX, XLSX, PPTX) which can contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Data is extracted using libraries like markitdown, pdfplumber, pandas, and pypdf across multiple scripts, including SKILL.md and scripts/pdf_extract_fields.py.
Boundary markers: No explicit delimiters or delimiters are defined to separate untrusted content from the agent's instructions.
Capability inventory: The skill can execute shell commands via subprocess.run (LibreOffice) and perform file-write operations, providing an exploit path if the agent follows instructions embedded in data.
Sanitization: The use of defusedxml in scripts/pack.py and scripts/unpack.py provides protection against XML-based attacks like XXE.
[COMMAND_EXECUTION]: The skill executes external system commands using subprocess.run to interact with LibreOffice (soffice) and other utilities for file conversion and office file operations in scripts/soffice.py, scripts/accept_changes.py, and scripts/recalc.py.
[DYNAMIC_EXECUTION]: The skill dynamically generates LibreOffice Basic macros and writes them to the user's configuration directory or a temporary profile before execution to perform specific document tasks in scripts/accept_changes.py and scripts/recalc.py.

document-generator