Archive / See what your AI agents are doing at a glance, then build your own view

See what your AI agents are doing at a glance, then build your own view

A tiny pixel-art office that acts out what Claude Code is doing gave me the idea. Here's what a good agent view shows, and a prompt to design your own from Claude Code's hooks.

For: Solo founders, Small business owners

Topics: Agent management, Productivity

Why I'm sharing it

I'm a visual person. A scrolling wall of tool calls tells me everything and shows me nothing. What I want is to glance over and see which agent is planning, which one is working and which one is stuck waiting for me.

This week I found Board Room, a small open-source Claude Code plugin by John Schettini. It docks a pixel-art office next to the chat. A little crew acts out what Claude is doing: planning happens in a boardroom, the work happens in an open office, finished work goes to a town hall, and an alert shows when Claude needs you. There's also a screen with numbers for each turn and the session.

It's a fun idea, and I don't want to copy it. It made me want to design a view that fits how I work. That's what this post is about.

What makes it work

It shows states, not text. Planning, working, waiting for you and done are four pictures you can read in a second.

The data is already there. Claude Code runs hooks at set moments, like when a tool is used, when it needs permission and when it finishes. A view only has to listen to those moments.

It watches without touching. Board Room's page says its hooks only observe and never change or delay what Claude does. That's the bar for any view you build.

It makes "needs you" loud. Of all the states, that's the one that costs you time when you miss it.

It's honest about limits. Board Room works in Claude Code on your own computer, not in cloud sessions, and its sound only plays on a Mac. It's also brand new, so expect changes.

The prompt

Run this in Claude Code. It plans first and builds nothing until you say yes.

Prompt

I want a simple visual view of what you're doing in my Claude Code sessions, so I can see it at a glance instead of reading the chat. Plan it with me before you build anything. 1. Ask me which states I care about. Suggest a starting set: planning, working, using a tool, waiting for my permission, waiting for my reply, done. Ask how I'd like each one to look: [colors, icons, characters or a simple board]. 2. Propose how the view gets its data from Claude Code hooks. For each state, name the hook event that signals it and what the hook should record. Record only the event name, the time, the session, and the tool name. Never record my prompts, file contents, tool inputs or any keys. 3. Propose the simplest way to show it: [a local web page, a small menu bar app, or a terminal panel]. Keep everything on my computer, with no outside service. 4. Show me the full plan, the hook settings and which settings file they go in. Wait for my yes. 5. After my yes, build it, then run a short test task so I can watch every state appear.

How to use it

Start small. Four states on a plain page is a real win. You can add characters and rooms later.

Read a plugin's hooks before you install it, Board Room included. Hooks run commands on your computer by themselves, so check what each one runs and what it records.

Keep the log boring on purpose. Times and state names are enough to draw a view. Your prompts and files don't need to leave the chat.

If you'd rather not build a screen at all, my post on turning a smart bulb into a status light is the low-tech version of the same idea.

Inspired by Board Room, an open-source Claude Code plugin by John Schettini, MIT license, at github.com/johnschettini-ship-it/Board-Room. The description here is from its project page as of November 19, 2026. The prompt is my own.

Use it for: Designing a simple screen that shows what your coding agents are doing right now, without reading the whole chat.

Back to the archive