The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The documentation encourages downloading and installing executable binaries (excelcli.exe, mcp-excel.exe) and a .NET tool (Sbroenne.ExcelMcp.CLI) from a third-party GitHub repository (github.com/sbroenne/mcp-server-excel) and NuGet, which are not verified sources.
[REMOTE_CODE_EXECUTION]: The skill provides tools specifically designed to import and execute VBA macros (vba import-module, vba run-macro), which allows for the execution of arbitrary code within the Microsoft Excel process.
[REMOTE_CODE_EXECUTION]: The skill enables the creation and execution of Power Query M code and DAX measures. These languages have the capability to perform network requests (e.g., Web.Contents) and access local file systems, presenting a vector for code execution or data access.
[COMMAND_EXECUTION]: The skill operates primarily through a CLI tool (excelcli). If the AI assistant processes untrusted user input to construct these CLI commands, it could lead to argument injection or unintended command execution.
[DATA_EXFILTRATION]: The tool includes functionality to capture screenshots of ranges or worksheets as PNG files. If the agent is compromised via prompt injection, this capability could be used to exfiltrate sensitive visual data.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Data is ingested from Excel workbooks, CSV files, and external database connections (OLEDB/ODBC) via the range get-values and Power Query tools.
Boundary markers: The instructions do not define boundary markers or provide guidance on isolating data from instructions when the AI processes worksheet content.
Capability inventory: The skill has extensive capabilities including file system writes (save, export-all), VBA execution, Power Query execution, and screenshot capture.
Sanitization: There is no evidence of sanitization or validation of data retrieved from Excel before it is processed by the AI.

excel-mcp-automation