base

Pass

Audited by Gen Agent Trust Hub on Jun 30, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [SAFE]: Analysis of the instructions and provided scripts indicates the skill is designed for legitimate database automation and management tasks.
  • [EXTERNAL_DOWNLOADS]: The skill suggests the installation of well-known, widely used Python packages (pyodbc, sqlalchemy) from standard package registries.
  • [COMMAND_EXECUTION]: Uses standard command-line tools (soffice, killall) to control the LibreOffice application and its headless background service.
  • [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection through data ingestion.
  • Ingestion points: Reads data from external sources including CSV files, spreadsheets, and database queries mentioned in SKILL.md.
  • Boundary markers: None identified in the provided instructions to delimit untrusted data from the agent's context.
  • Capability inventory: Includes file writing via doc.storeToURL, shell command execution via soffice, and network database connectivity via UNO/SDBC.
  • Sanitization: Recommends the use of parameterized queries in the Best Practices section to mitigate injection risks.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 30, 2026, 01:47 PM
Security Audit — agent-trust-hub — base