xlsx
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION] (HIGH): The skill ingests untrusted Excel data which can influence agent behavior. Ingestion points: recalc.py via openpyxl.load_workbook. Boundary markers: None. Capability inventory: Subprocess execution (soffice) and filesystem writes. Sanitization: None. Malicious spreadsheets could trigger unintended actions.
- [COMMAND_EXECUTION] (MEDIUM): The script executes system binaries (soffice, timeout) via subprocess.run. While it uses argument lists to mitigate shell injection, it performs low-level system interactions using user-provided paths.
- [DYNAMIC_EXECUTION] (MEDIUM): The script dynamically creates and installs a LibreOffice Basic macro on the host system to perform recalculations, which modifies the user's software environment and executes generated code.
Recommendations
- AI detected serious security threats
Audit Metadata