Archive / Stop babysitting your coding agent and let a light tell you what it's doing

Stop babysitting your coding agent and let a light tell you what it's doing

Use the hooks built into Claude Code and Codex to turn a smart bulb into a status light, so you know from across the room whether the agent is working, waiting for you or done.

For: Solo founders, Small business owners

Topics: Agent management, Productivity

Why I'm sharing it

Anyone who gives Claude Code or Codex a real job knows the hover. You hand over the task, then keep glancing back at the terminal. Is it still going? Is it stuck on a question? Did it finish twenty minutes ago while you were in another tab?

A builder posted about solving this with light. Their bulbs turn one color while the agent works, pulse another when it needs permission and change again when it's done. They built an app for it called AC Bridge. What I liked is that the idea doesn't need their app. Both tools already have the switch built in, and a cheap Wi-Fi bulb can do the rest.

What makes it work

The tools tell you when something happens. Claude Code and Codex both run hooks, small commands that fire at set moments in a session. You don't have to ask the agent to remember to signal you. The tool does it.

Three moments cover most of the hovering. In Claude Code, PermissionRequest fires when a tool call needs your approval, Stop fires when Claude finishes responding, and a Notification hook with the idle_prompt matcher fires when Claude has been waiting on you for about a minute. Codex has PermissionRequest and Stop too.

A light stays on until you deal with it. A phone ping is gone once you swipe it. A red lamp is still red when you walk back in, so you know at a glance whether you missed something. It's also silent and doesn't pull you into your phone.

Rules can show progress, not just stops. You can add a few lines to the instruction file the agent reads every session, CLAUDE.md for Claude Code or AGENTS.md for Codex, asking it to signal when it finishes a stage or when tests fail. Treat those as requests the agent follows, not guarantees. The hooks are the part that always fires.

You can answer without going back to the desk. Claude Code's Remote Control, started with /remote-control or its short form /rc, lets you continue the session from the Claude app on your phone or from claude.ai/code. The docs say it's available on Pro, Max, Team and Enterprise plans.

The prompt

Run this inside Claude Code in the project you want to watch. Swap in your own bulb, colors and moments.

Prompt

I want a smart bulb to show me what you're doing in this project, so I can leave the room during long jobs. My bulb: [brand and model, for example a Wi-Fi color bulb, or a light already in Home Assistant]. How I control it today: [the brand's app, Home Assistant, or nothing yet]. My computer: [Mac, Windows or Linux]. 1. Work out how to change this bulb's color from a command on my home network. Prefer local control over a cloud account. If it needs a key or token, tell me where to create it and where to store it safely, outside this repo. Don't ask me to paste it into this chat. 2. Write one small script that takes a state name and sets the light: working = [color], needs permission = [color, pulsing], waiting for me = [color], done = [color]. Test each state while I watch, and stop if the bulb doesn't respond. 3. Show me the hooks you'd add before you add them: PermissionRequest for needs permission, Notification with the idle_prompt matcher for waiting for me, Stop for done, and UserPromptSubmit to switch back to working when I reply. Say which settings file they go in and why. 4. Only after I say yes, add the hooks and run one short test task so I can see every color change. 5. Suggest up to three lines for CLAUDE.md that ask you to flash the light when you finish a stage of a plan or when tests fail. Don't add them until I approve.

How to use it

Start with one bulb you can see from where you usually sit. A bedside lamp with a color Wi-Fi bulb in it is enough.

Put hooks you want everywhere in your user settings file, ~/.claude/settings.json, and hooks for one project in that project's .claude/settings.json. Hooks run commands on your machine without asking, so read every one before you save it. Codex makes you review and trust hooks before they run, through its /hooks command.

Keep the colors few and obvious. Four states is plenty. If you can't tell them apart from across the room in daylight, change them.

If you use Codex, ask it the same prompt. Its hooks live in a hooks.json file or in config.toml, and it has no separate waiting-for-you event, so lean on PermissionRequest and Stop.

If you'd rather not build anything, AC Bridge is one ready-made app that sets this up for you. Its maker says it works with many bulb brands and with Home Assistant. I haven't tried it myself.

Based on a post by the maker of AC Bridge about using smart lights to follow Claude Code and Codex. The author wasn't named in the copy I saved, so if it's yours, tell me and I'll credit you by name. The prompt and the hook details here are my own, checked against the Claude Code and Codex docs on November 19, 2026.

Use it for: Walking away from a long Claude Code or Codex job without checking the screen every few minutes to see if it needs you.

Back to the archive