The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py dynamically generates C source code and compiles it at runtime into a shared library (.so) using gcc. This library is then injected into the soffice process using the LD_PRELOAD environment variable to intercept and shim low-level system calls such as socket, listen, accept, and close.
[COMMAND_EXECUTION]: Several scripts, including scripts/office/soffice.py, scripts/convert_hwp.py, and scripts/convert_to_pdf.py, use the subprocess module to execute system commands such as gcc and soffice with user-supplied arguments and environment modifications.
[EXTERNAL_DOWNLOADS]: The documentation in SKILL.md instructs the user to install the external Node.js package @ssabrojs/hwpxjs and mentions dependencies on external tools like LibreOffice and Poppler (pdftoppm).
[INDIRECT_PROMPT_INJECTION]: The skill provides tools for reading and extracting content from HWPX documents, creating a potential surface for indirect prompt injection if the processed documents contain malicious instructions.
Ingestion points: Document content is read from external HWPX files via scripts/extract_text.py and scripts/extract_tables.py.
Boundary markers: The skill does not implement delimiters or safety instructions to distinguish ingested document data from the agent's core instructions.
Capability inventory: The skill has significant capabilities, including file system access (pack.py, unpack.py), command execution, and runtime compilation of native code.
Sanitization: Extracted XML and text data are processed and displayed to the agent without any visible sanitization or filtering.

hwpx