Claude Guard

Lightweight request guardian for Claude browser agents. Monitors, blocks, and approves HTTP requests in real-time.

A lightweight Chrome extension that acts as an HTTP request guardian while a Claude browser agent is active. It monitors, blocks, pauses, and gates outgoing requests behind human approval — so risky automated actions are visible and controlled. How It Works When Claude Guard detects an active Claude agent on a page (via the claude-agent-glow-border DOM element), it intercepts outgoing HTTP requests and applies configurable rules: - Auto-mode blocks non-GET/HEAD/OPTIONS requests to same-site URLs or requests carrying Authorization headers unless explicitly allowed. - Blocked requests are routed to a human approval popup where you can allow once, always allow (by hostname), or deny. - Denied requests receive a synthetic 403 response so the agent can handle the rejection gracefully. Intercepted APIs include fetch, XMLHttpRequest, HTMLFormElement.submit/requestSubmit, and navigator.sendBeacon. Features - Auto-blocking of sensitive mutations (POST/PUT/DELETE/PATCH) and authenticated requests - Allow / block lists with glob pattern support - Human approval window for pending requests with headers/body preview - Toolbar popup for toggling auto mode, managing rules, and viewing stats - Visual indicators — tab badge, banner, toasts, and title updates when the guard is active - Request logging with capped history