pptx
Pass
Audited by Gen Agent Trust Hub on Feb 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
subprocessmodule to execute system utilities includingsoffice(LibreOffice) for document validation and conversion, andpdftoppm(Poppler) for generating slide thumbnails. - [COMMAND_EXECUTION]: Utilizes the Playwright library to launch a headless Chromium browser instance (
chromium.launch) for rendering HTML slides during the presentation creation workflow. - [COMMAND_EXECUTION]: Invokes the
gitcommand viasubprocess.runto perform word-level comparisons during the validation of tracked changes in documents. - [PROMPT_INJECTION]: Indirect prompt injection surface detected. (1) Ingestion points: Processes untrusted content from PowerPoint (.pptx) files and HTML files via scripts such as
inventory.py,unpack.py, andhtml2pptx.js. (2) Boundary markers: Absent in the processing pipeline. (3) Capability inventory: Executes arbitrary system commands and launches a headless browser based on document processing logic. (4) Sanitization: Implements secure XML parsing viadefusedxml, but does not provide specific sanitization for natural language instructions that might be embedded in the slides to influence the agent's behavior.
Audit Metadata