The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/soffice.py contains a embedded C source code string (_SHIM_SOURCE) that is written to a temporary file and compiled at runtime using gcc. The resulting shared object is then dynamically loaded into the LibreOffice process using the LD_PRELOAD environment variable to intercept and redirect system calls related to Unix domain sockets.
[PERSISTENCE]: The script scripts/recalc.py persistently modifies the user's LibreOffice configuration by writing a StarBasic macro to the application's local macro storage directory (Module1.xba). This modification persists across application restarts and can affect the behavior of future LibreOffice sessions on the system.
[COMMAND_EXECUTION]: The skill executes multiple external system commands via the subprocess module, including gcc for library compilation, soffice for spreadsheet processing, and git for comparing document content. While the arguments are generally handled as lists to prevent shell injection, the reliance on these system tools expands the attack surface.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted spreadsheet files which could contain malicious data designed to influence the behavior of the AI agent when the content is read into context.
Ingestion points: Data enters the context via pd.read_excel and load_workbook as documented in SKILL.md.
Boundary markers: None identified in the provided instructions to separate spreadsheet data from instructions.
Capability inventory: The skill has extensive capabilities including file system access, command execution, and dynamic code execution.
Sanitization: While the skill uses defusedxml to prevent XML-based attacks during parsing, it lacks explicit sanitization of cell content before the agent processes it.

xlsx