Skip to content

Security: liutingqiu/raw-browser

Security

SECURITY.md

Security policy

Supported versions

Security fixes are applied to the latest release.

Reporting a vulnerability

Use GitHub's private vulnerability reporting for this repository: https://github.com/liutingqiu/raw-browser/security/advisories/new.

Include the affected version, operating system, reproduction steps, impact, and any suggested fix. Do not include secrets or account data. Please avoid filing a public issue until a fix is available.

Trust model

raw intentionally executes model-generated browser scripts with the current user's operating-system permissions. The worker used for timeouts isolates CPU-bound scripts; it is not a security sandbox. The extension bridge runs with the browser's full extension authority (<all_urls>, debugger, cookies, proxy, downloads, history, management), and plain CDP mode controls a dedicated browser profile. The calling agent's prompt and permission mode are the authority for which actions run.

The bridge is defended at the transport boundary: listeners bind only to 127.0.0.1, every HTTP and WebSocket request requires a per-browser random token, native messaging is restricted to the pinned extension id, and a host only attaches to a debug port it owns. Release host binaries are published with SHA256SUMS.txt; scripts/install-edge-bridge.mjs and scripts/install-edge-bridge.ps1 verify the checksum when they download one. Cookies, tokens and page contents are secrets; do not log or forward them to external services.

Configuration, downloads, bridge tokens and the dedicated browser profile are stored locally under ~/.raw/. Remote page content is untrusted input to the model and can influence subsequent tool calls; use the calling agent's permission mode when reviewing content that should not trigger state changes automatically.

There aren't any published security advisories