The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill performs dynamic execution by generating C source code at runtime and compiling it using gcc in scripts/office/soffice.py. The resulting shared library is injected into the LibreOffice process via the LD_PRELOAD environment variable.\n- [COMMAND_EXECUTION]: High-risk command execution is present through the use of system binaries.\n * Evidence: The skill calls subprocess.run to execute gcc, soffice, git, and pdftoppm in soffice.py, accept_changes.py, and redlining.py.\n- [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection through document data ingestion.\n * Ingestion points: Word document XML content processed from word/document.xml by the unpack.py script.\n * Boundary markers: Absent.\n * Capability inventory: The skill can execute arbitrary system commands (soffice, gcc) and write to the local filesystem.\n * Sanitization: The skill employs defusedxml to mitigate XML external entity (XXE) attacks during parsing.

docx