The Agent Skills Directory

Persistence Mechanisms (MEDIUM): The script recalc.py writes a LibreOffice Basic macro to the user's permanent configuration directory (~/.config/libreoffice/4/user/basic/Standard/Module1.xba on Linux or the equivalent on macOS). This modification persists across sessions and affects the global behavior of the LibreOffice application for that user.
Dynamic Execution (MEDIUM): The skill generates executable macro code as a string and writes it to disk before invoking it. While the code is hardcoded in the script, this pattern of 'generate-write-execute' is a common vector for more serious vulnerabilities if any part of the string becomes attacker-controlled.
Command Execution (LOW): The script uses subprocess.run to call system binaries including soffice, timeout, and gtimeout. This is required for its primary purpose but represents a significant capability for an AI agent.
Indirect Prompt Injection (LOW): The skill processes untrusted Excel files provided by users.
Ingestion points: Files are opened via openpyxl.load_workbook in recalc.py.
Boundary markers: None; the script iterates through all sheets and cells.
Capability inventory: The script triggers ThisComponent.calculateAll() in LibreOffice. A malicious Excel file could contain formulas like WEBSERVICE or LINK that trigger network requests or external data access when recalculated.
Sanitization: No validation is performed on the content of the formulas before requesting LibreOffice to execute them.

xlsx