The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py implements a sophisticated runtime code generation and process injection mechanism.
It contains a hardcoded C source code string (_SHIM_SOURCE) designed to shim system socket calls (socket, listen, accept).
It writes this code to a temporary file and executes gcc to compile it into a shared object library (lo_socket_shim.so).
It then executes the soffice (LibreOffice) binary with the LD_PRELOAD environment variable set to point to this compiled library, effectively injecting custom code into the process to bypass sandbox-imposed network socket restrictions.
[COMMAND_EXECUTION]: Multiple components of the skill execute shell commands to interact with the host system.
scripts/recalc.py and scripts/office/soffice.py execute soffice and utility commands like timeout or gtimeout to manage document processing.
scripts/office/validators/redlining.py executes git diff to perform comparison tasks during validation.
scripts/recalc.py modifies the user's application environment by writing a StarBasic macro (RecalculateAndSave) to the LibreOffice configuration directories (~/.config/libreoffice or ~/Library/Application Support/LibreOffice).
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a notable attack surface for indirect prompt injection based on its intended use and capabilities.
Ingestion points: The skill is designed to read and process untrusted Excel and Word files using openpyxl and lxml (scripts/recalc.py, scripts/office/unpack.py).
Boundary markers: Absent. There are no delimiters or instructions to treat embedded file content as data rather than instructions.
Capability inventory: The skill possesses high-risk capabilities including arbitrary command execution, runtime compilation, and process injection (scripts/office/soffice.py).
Sanitization: There is no evident sanitization of content extracted from processed documents before it influences the agent's logic or environment.

xlsx