The Agent Skills Directory

[DYNAMIC_EXECUTION]: Runtime creation and compilation of system components.
The script scripts/office/soffice.py generates a C source file and compiles it using gcc into a shared library (lo_socket_shim.so).
This library is loaded via the LD_PRELOAD environment variable to shim socket system calls, allowing LibreOffice to function in restricted execution environments where AF_UNIX sockets are blocked.
[COMMAND_EXECUTION]: Systematic use of subprocesses for core spreadsheet tasks.
The skill invokes several external utilities including soffice (LibreOffice) for formula recalculation, gcc for library compilation, git for comparing document versions, and timeout for process control.
[PERSISTENCE_MECHANISMS]: Persistent modification of application configuration state.
The script scripts/recalc.py writes a StarBasic macro to the user's LibreOffice configuration directory (e.g., ~/.config/libreoffice/4/user/basic/Standard/Module1.xba). This is intended to automate spreadsheet recalculation but involves modifying application state across sessions.
[INDIRECT_PROMPT_INJECTION]: Processing of untrusted spreadsheet data.
The skill parses and manipulates content from user-provided .xlsx, .csv, and other spreadsheet files using pandas and openpyxl. This creates an entry point for adversarial data or malformed formulas to influence the agent's behavior.
[DATA_EXPOSURE_AND_EXFILTRATION]: Potentially unsafe XML parsing during validation.
The validation components in scripts/office/validators/ utilize lxml.etree to parse XML components of Office documents. Without explicit entity resolution restrictions, this could potentially expose the environment to XML External Entity (XXE) vulnerabilities when processing malicious files.

xlsx