frontend-slides
Pass
Audited by Gen Agent Trust Hub on Jul 11, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses shell commands to manage software installation and deployment.
- Evidence: Uses "pip install python-pptx -q" to install Python libraries.
- Evidence: Uses "vercel deploy" to publish generated presentations to the Vercel platform.
- [EXTERNAL_DOWNLOADS]: The skill downloads software and dependencies from public registries and official sources.
- Evidence: Fetches the "python-pptx" package from the Python Package Index (PyPI).
- Evidence: Downloads the "chromium" browser and "playwright" automation library via the npm registry.
- [REMOTE_CODE_EXECUTION]: The skill runs dynamic code snippets provided as string arguments to system interpreters.
- Evidence: Uses "python -c" with an inline script to parse ".pptx" files and convert content to JSON.
- Evidence: Uses "node -e" with an inline script to control a browser instance for PDF export tasks.
- [PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection via external file processing.
- Ingestion points: The "Phase 4" workflow reads content from user-provided ".pptx" file paths using a Python extractor.
- Boundary markers: No delimiters or protective instructions are specified to isolate the content extracted from the PowerPoint file from the agent's core instructional logic.
- Capability inventory: The agent can execute shell commands (pip, python, node), perform network operations (vercel), and write files to the local system.
- Sanitization: Content extracted from the PowerPoint file is processed and used to generate executable HTML/JS presentations without explicit validation or escaping of the ingested text.
Audit Metadata