The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection by processing external spreadsheet and CSV data which may contain embedded instructions meant to influence the agent.
Ingestion points: Spreadsheets are loaded via the ExcelJS library and CSV files via csv-parse in scripts/spreadsheet_cli.mjs.
Boundary markers: The skill does not employ specific boundary markers or instructions for the agent to distinguish between cell data and potential instructions during runtime processing.
Capability inventory: The skill can execute external tools (LibreOffice, PDF renderers) via execFile and perform file system operations, including writing files outside the internal workspace via the deliver command.
Sanitization: The skill includes robust validation logic to reject unsafe ZIP package entries (preventing path traversal) and malicious relationship targets in XLSX packages.
[EXTERNAL_DOWNLOADS]: The skill fetches its runtime environment dependencies from the public NPM registry during its initialization phase.
Evidence: scripts/spreadsheet.sh executes npm ci to install pinned dependencies defined in runtime/package-lock.json.
[COMMAND_EXECUTION]: The skill relies on external command-line utilities to provide functionality for spreadsheet recalculation and rendering.
Evidence: scripts/spreadsheet_cli.mjs executes soffice (LibreOffice), pdftoppm, mutool, and magick using the execFile method.

spreadsheets