mikepenz/agent-belay ? reverse-engineered prompt

Build me a desktop app called Agent Belay that sits between me and my AI coding tools and makes me approve risky actions before they happen.

I want it to catch requests like file edits, shell commands, and web fetches from tools such as Claude Code and GitHub Copilot, then show them in a clean review screen where I can approve, deny, or let certain patterns always through. It should block obviously dangerous stuff automatically, like destructive commands, credential grabbing, or sneaky command bypasses, and keep a searchable history of what happened. If possible, add optional risk scoring so safer requests can be auto approved and dangerous ones can be auto denied.

Make it feel like a real background app with a system tray icon, notifications, settings for protections and capabilities, and a usage view that shows token counts, cost, and latency from local session files. Cross platform would be great, but at least make it run well on macOS. If you need current docs for the agent integrations, look them up online.