libreoffice-impress
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill extracts text content from presentations, providing a vector for instructions embedded in data to influence agent behavior. 1. Ingestion points: 'scripts/impress/session.py' (via read_text and get_notes). 2. Boundary markers: Absent. 3. Capability inventory: Subprocess calls in 'scripts/uno_bridge.py' and 'scripts/impress/snapshot.py'; file writes in 'scripts/impress/core.py'. 4. Sanitization: Absent.
- [COMMAND_EXECUTION]: The skill executes the 'soffice' binary using the 'subprocess' module to manage presentations. Evidence: 'scripts/uno_bridge.py' launches headless LibreOffice; 'scripts/impress/snapshot.py' executes conversion commands.
- [DYNAMIC_EXECUTION]: The skill dynamically modifies the Python import path to bridge with the system LibreOffice installation. Evidence: 'scripts/uno_bridge.py' inserts detected program directories into 'sys.path' at runtime.
Audit Metadata