The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains logic to compile C source code at runtime using gcc. The resulting shared object is then injected into the LibreOffice (soffice) process environment via the LD_PRELOAD environment variable. This is a high-risk pattern that allows for the execution of arbitrary compiled code and process-level manipulation (File: scripts/office/soffice.py).
[COMMAND_EXECUTION]: The skill uses subprocess.run to execute multiple system commands, including invoking the C compiler (gcc) and launching the LibreOffice suite with specialized arguments (Files: scripts/office/soffice.py, scripts/recalc.py).
[COMMAND_EXECUTION]: The recalculation script automatically writes a StarBasic macro to the user's application configuration directory (e.g., ~/.config/libreoffice/...). This constitutes unauthorized modification of application configuration and persistence of custom executable logic within the productivity suite (File: scripts/recalc.py).
[PROMPT_INJECTION]: The skill represents a significant surface for indirect prompt injection (Category 8):
Ingestion points: Reads external spreadsheet data using pandas.read_excel and openpyxl.load_workbook (File: SKILL.md).
Boundary markers: No delimiters or instructions are provided to the agent to ignore potentially malicious content or formulas within the processed files.
Capability inventory: The skill has extensive capabilities including file system write access, arbitrary command execution via subprocess, and runtime code compilation (File: scripts/office/soffice.py).
Sanitization: Content is parsed for XML well-formedness (using defusedxml), but there is no sanitization or filtering of data/formulas before they are processed by the agent or the recalculation engine.

xlsx