The Agent Skills Directory

[Persistence Mechanisms] (MEDIUM): The script modifies the user's application configuration by creating a persistent LibreOffice macro file (Module1.xba) in ~/.config/libreoffice or ~/Library/Application Support/LibreOffice. This configuration remains on the system after the skill completes execution.
[Dynamic Execution] (MEDIUM): The skill dynamically generates LibreOffice Basic code (RecalculateAndSave macro) and executes it via the soffice binary. This pattern of script generation and execution is a common security concern, although the macro content here is hardcoded.
[Indirect Prompt Injection] (LOW): The skill processes external Excel files and returns cell contents (error locations and types) to the agent. This creates a surface where malicious data in a spreadsheet could influence agent behavior.
Ingestion points: recalc.py (lines 103, 119) loads and iterates through all cells in the provided Excel file.
Boundary markers: None. The script does not use delimiters or warnings when processing cell content.
Capability inventory: subprocess.run (lines 39, 79) executes external commands including soffice and timeout utilities.
Sanitization: None. Cell values are checked for specific error strings and the resulting locations are returned as raw strings.
[Command Execution] (LOW): The skill invokes several external binaries (soffice, timeout, gtimeout) using subprocess.run. While it uses list-based arguments to prevent shell injection, it relies on these tools being present and executable in the environment.

xlsx