frontend-slides

Warn

Audited by Gen Agent Trust Hub on Apr 1, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads several third-party dependencies during operation. It installs the python-pptx library for PowerPoint conversion and the vercel and playwright packages from NPM for deployment and PDF export respectively. It also triggers the download of the Chromium browser binary. These downloads target well-known registries and official services.
  • [COMMAND_EXECUTION]: Multiple shell scripts (scripts/deploy.sh, scripts/export-pdf.sh) are used to manage the presentation workflow. These scripts execute commands for environment setup, package management, and service interaction. Notable commands include global package installation (npm install -g vercel) and local server management.
  • [REMOTE_CODE_EXECUTION]: The PDF export functionality in scripts/export-pdf.sh uses a dynamic execution pattern. It writes a Node.js script (export-slides.mjs) to a temporary directory at runtime and then executes it using the node command. This script performs complex operations like starting a local HTTP server and controlling a headless browser to capture screenshots.
  • [DATA_EXFILTRATION]: The deployment feature in scripts/deploy.sh transmits local presentation files and assets to Vercel, an external hosting provider. While this is the intended method for sharing presentations globally, it represents a data export to a third-party service that occurs during the sharing phase.
  • [PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks through its PowerPoint conversion feature. 1. Ingestion points: scripts/extract-pptx.py extracts text and notes from user-provided PPTX files. 2. Boundary markers: No explicit delimiters or instructions are used to isolate extracted content within the generation prompt. 3. Capability inventory: The skill has network access (Vercel deployment), file system write permissions, and dynamic script execution. 4. Sanitization: No sanitization or validation of the extracted PowerPoint content is performed before it is processed by the agent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 1, 2026, 03:40 PM
Security Audit — agent-trust-hub — frontend-slides