The Agent Skills Directory

[COMMAND_EXECUTION]: Runtime Compilation and Process Injection
File: scripts/office/soffice.py contains a hardcoded C source string (_SHIM_SOURCE) that is written to a temporary file and compiled using gcc at runtime to create a shared library (lo_socket_shim.so).
The resulting library is then injected into the LibreOffice (soffice) process via the LD_PRELOAD environment variable. This technique is used to bypass environment restrictions on Unix sockets but is a high-risk capability that can be repurposed for malicious process interference.
[COMMAND_EXECUTION]: Persistent Macro Installation
File: scripts/recalc.py creates and writes a StarBasic macro to the user's permanent LibreOffice configuration directory (~/.config/libreoffice/ on Linux or ~/Library/Application Support/LibreOffice/ on macOS). This code persists outside the AI agent session and will be executed by LibreOffice during recalculation tasks.
[COMMAND_EXECUTION]: System Tool Invocation
The skill utilizes subprocess.run to execute various system tools including soffice, gcc, and git across multiple script files (scripts/recalc.py, scripts/office/soffice.py, scripts/office/validators/redlining.py).
[PROMPT_INJECTION]: Indirect Prompt Injection Surface
Ingestion points: Spreadsheet data enters the agent context through pandas.read_excel and openpyxl.load_workbook in SKILL.md examples and user scripts.
Boundary markers: None present. The skill does not provide instructions to ignore or delimit embedded natural language instructions within the spreadsheet data.
Capability inventory: The skill has extensive capabilities including file system access, network potential through subprocess-invoked tools, and dynamic code generation.
Sanitization: None. Data from files is processed as-is, which could allow a malicious spreadsheet to influence subsequent agent actions.

xlsx