Like many web developers I’m integrating agentic AI (just Claude Code for now) into my workflows. So far, for security, I’ve landed on an approach where I:

• Run Claude in auto mode in devcontainers (and in Claude Code Web). These devcontainers are either running locally on my computer, or on a remote server. These usually don’t need permission or extra input, but they may finish a task.
• On my own computer, I don’t use auto; but keep to manual or edit-automatically; leading to more permission prompts.

I have not yet found a good and simple way where I can keep tabs on all my agents on all my hosts in an unobtrusive yet permanently visible way.

Instead of a 230$ hardware device, I first thought of a macOS status bar widget to show the status of my agents, which in the world of vibecoding already exists, of course.

I (well, Claude) found these related status bar projects that looked nice:

• https://github.com/m1ckc3s/claude-status-bar
• https://github.com/gmr/claude-status

… but that only support hooking into a locally-running Claude.

Hence, I’d like to fork one of these (probably the first one), and add support for remote or containerized agents using a daemon in the container that forwards Claude hooks toa relay server, which forwards them on to my computer.