Please report vulnerabilities by opening an issue in the GitHub repository or contacting the maintainer directly. We aim to address all security vulnerabilities within 48 hours.

Prism acts as a multiplexing browser for AI services. Security is paramount to ensure user data isolation and application integrity.

Each AI service (ChatGPT, Gemini, Claude, Grok) runs in a separate Electron webview with its own partition:

• persist:chatgpt
• persist:gemini
• persist:claude
• persist:grok

This strict partitioning ensures that cookies, local storage, and session data are never shared between services.

The application adheres to dangerous-function mitigation best practices:

• Sandbox: Enabled. Renderer processes are sandboxed.
• Context Isolation: Enabled (contextIsolation: true). Preload scripts run in a separate context from the webpage.
• Node Integration: Disabled (nodeIntegration: false). Web content cannot access Node.js primitives directly.

We enforce a strict CSP in the main renderer window:

• default-src 'self': Only load local resources by default.
• script-src 'self': No external scripts except trusted CDNs (Tailwind).
• style-src 'self': No external styles except trusted CDNs (Google Fonts).

Communication between the renderer (UI) and webviews (AI Services) uses ipc-message and sendToHost.

• Input Validation: All messages (broadcasts, scroll sync) are validated for type and length before processing.
• No Remote Module: The dangerous remote module is not used.

• All dependencies are pinned to exact versions in package.json to prevent supply chain attacks via auto-updates.
• Regular npm audit checks are performed.