Show HN: Greenlight – Manage your AI coding agents from your phone

My main agent is Claude Code but I've been dabbling with Copilot CLI, Cursor CLI, and Codex CLI too. The biggest productivity killer across all of them is the same: agents blocking on permission prompts while I'm away from my desk.

So I made Greenlight: an iOS app that forwards agent permission requests to your phone as push notifications. And it worked! And then Anthropic released "Remote Control". That was a tough day. But I realized something: there's no agent-agnostic remote control. So I've extended Greenlight to work with Copilot, Cursor and Codex too.

The companion CLI (`greenlight connect`) wraps your agent session. The agent runs in full auto mode while Greenlight intercepts actions before they execute. This inverts control: instead of the agent deciding what permissions to ask you for, you decide what actions the agent is permitted to take. Did you know that Claude Code runs `ssh ... git@github.com` every time you start a session?

Other stuff I like about it: compound shell commands are split into segments and color-coded by risk level so it's easy to judge the risk of a command. Over time your rules are tuned to the project and you only receive requests for novel or destructive commands. My favorite feature is "pull the plug": you can sigkill any agent that is rabbitholing and ignoring you.

The server mostly a router - it never inspects or saves your transcript. Would love feedback, especially from anyone juggling multiple agents.