The Agent Skills Directory

Privilege Escalation (HIGH): The 'SKILL.md' file instructs users to use sudo apt-get install for system dependencies (libreoffice, poppler-utils), which grants root-level access during setup.- Indirect Prompt Injection (HIGH): The skill is designed to process external, untrusted data which presents a high risk for indirect injection. * Ingestion points: Reads .xlsx, .csv, and .tsv files via openpyxl.load_workbook (seen in read_existing_spreadsheet.py) and pandas. * Boundary markers: None identified; instructions do not specify delimiters or warnings to ignore embedded content within cells. * Capability inventory: The skill can create directories (mkdir), write files (wb.save), and execute system commands (soffice, pdftoppm) as seen in SKILL.md and example scripts. * Sanitization: None identified; cell content is processed and potentially used in further logic or rendered without filtering.
External Downloads (LOW): The skill requires multiple external dependencies (pandas, openpyxl, matplotlib, libreoffice, poppler). While these are from trusted repositories, the installation happens at runtime or setup. [TRUST-SCOPE-RULE] Applied: Severity downgraded from MEDIUM to LOW because the author 'openai' is a recognized trusted organization.
Command Execution (MEDIUM): The workflow in 'SKILL.md' utilizes shell commands (soffice --headless, pdftoppm -png) to convert and render documents. This creates a risk if filenames or sheet content are not properly sanitized before being passed to the shell.

spreadsheet