The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes system commands via subprocess.run to manage document processing utilities like pandoc and pdftoppm.
[COMMAND_EXECUTION]: In scripts/office/soffice.py, the skill generates C source code at runtime, compiles it using gcc, and uses LD_PRELOAD to inject the resulting library into the LibreOffice process. This mechanism is used to bypass socket restrictions in sandboxed environments.
[PROMPT_INJECTION]: The skill processes content from user-supplied documents, creating an attack surface for indirect prompt injection.
Ingestion points: Files are read into the agent's context using scripts/office/unpack.py and scripts/office/soffice.py.
Boundary markers: Absent. The skill does not use specific delimiters or instructions to isolate document content from system prompts.
Capability inventory: Across various scripts, the skill can execute arbitrary system commands, perform runtime compilation, and read/write filesystem content.
Sanitization: The skill utilizes defusedxml for XML parsing which helps mitigate XXE vulnerabilities, although it does not filter natural language instructions within documents.

docx