The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill's setup instructions in SKILL.md include sudo apt-get install commands for system dependencies. While these are common for environmental setup, encouraging the use of administrative privileges increases the risk of unauthorized system modifications if the agent executes these commands without human oversight.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to read and process external spreadsheet files which may contain untrusted content.\n
Ingestion points: Data is ingested through openpyxl.load_workbook and pandas functions, particularly in references/examples/openpyxl/read_existing_spreadsheet.py.\n
Boundary markers: The instructions do not define clear boundaries or 'ignore' tags for data processed from these spreadsheets.\n
Capability inventory: The skill has permission to write files and execute shell commands for rendering tasks.\n
Sanitization: No sanitization or validation logic is applied to the content of the spreadsheet cells before they are processed or displayed to the agent.\n- [COMMAND_EXECUTION]: In SKILL.md, the skill uses shell-based tools (soffice, pdftoppm) for rendering. These commands use environment variables like $INPUT_XLSX and $OUTDIR. If these variables contain unsanitized filenames or paths provided by an external source, it could lead to command injection.\n- [EXTERNAL_DOWNLOADS]: The skill lists standard Python packages (openpyxl, pandas, matplotlib) and official system repositories as sources for necessary dependencies. These are legitimate and consistent with the skill's purpose.

openai-spreadsheet