The Agent Skills Directory

[COMMAND_EXECUTION]: Runtime compilation of embedded C source code.
The file scripts/office/soffice.py embeds C source code in the _SHIM_SOURCE variable.
The script writes this code to the temporary directory and executes gcc via subprocess.run() to compile it into a shared object library (lo_socket_shim.so).
[COMMAND_EXECUTION]: Process injection via LD_PRELOAD.
The scripts/office/soffice.py script sets the LD_PRELOAD environment variable to the path of the compiled shared library before launching the soffice process.
This technique shims standard socket-related system calls (socket, listen, accept, close, read) to redirect application IPC behavior.
[COMMAND_EXECUTION]: Modification of host application configuration.
The scripts/recalc.py script automatically writes a StarBasic macro (Module1.xba) to the user's LibreOffice configuration directory (~/.config/libreoffice/ or ~/Library/Application Support/LibreOffice/).
This macro is used to force formula recalculation and save the document, which modifies the execution environment of the spreadsheet software.
[COMMAND_EXECUTION]: Execution of system-level binaries.
Multiple scripts use subprocess.run() to invoke external utilities such as git for diffing documents, soffice for recalculation, and gcc for compilation tasks.
[PROMPT_INJECTION]: Attack surface for indirect prompt injection.
Ingestion points: scripts/office/unpack.py and scripts/recalc.py process untrusted external spreadsheet files (.xlsx, .csv) and XML content.
Capability inventory: The skill has extensive file-write and command-execution capabilities, including the ability to compile and inject binaries.
Sanitization: While the skill uses defusedxml to mitigate XML-based attacks, the lack of boundary markers or instructions to ignore embedded content in processed data creates a potential surface for indirect injection attacks.

xlsx