ibara
Install ibara

FIRST TASK

Your first task.

A small task on a page built for it: nothing to log in to, nothing saved, nothing to break. It shows the whole loop, from asking to checking the result. Then give your agent the same kind of job on an app you’re building.

Updated

1 · Ask your agent#

If your agent isn’t connected yet, choose Connect an Agent in the console’s toolbar, then Copy Prompt, and paste it to your agent, or run ibara prompt and paste what it prints.

Then paste this into your agent, with your spare’s name:

On spare-mini, open https://ibara.app/practice and check the Settings menu with the mouse and the keyboard. Take a screenshot and tell me what worked. Don’t change system settings.

The practice page has a small Settings menu that works with the mouse, Tab, Enter, Space, the arrow keys and Escape.

2 · Watch it work#

Right-click the ibara mark in your bar to open the console, then choose the spare. You’ll see the agent’s cursor open the browser, go to the page and work through the menu. Take Control at any point if you want to try it yourself, then Hand Back.

3 · Check the result#

Success looks like this:

  • The agent reports each check as passed or failed.
  • It brings back at least one screenshot showing the menu open.
  • The spare is left as it was, with the task finished and control released.

Check an app you’re building#

The useful next step is the check you’d otherwise do by hand after every change. Your agent built the page; now let it open the app, use it and see what happens. Checked, not just claimed.

On spare-mini, open http://my-laptop:3000/settings and check the notification settings. Turn on weekly summaries with the mouse, save, reload and confirm it stuck. Then turn it off using only the keyboard, save and reload again. Make each of these a check and take a screenshot after each reload. Write a short report to report.md in the task’s workspace and publish it. Don’t delete anything or send any email.

Swap in your app’s address. The spare has to reach it: use a staging URL, or your computer’s name on your tailnet with the dev server listening on your network, not only on localhost.

What comes back

  • Checks, passed or failed. The agent sets its checks when it starts the task, and each one comes back passed or failed when it finishes. For example: “Saved with the mouse: passed. Saved with the keyboard: failed, Tab never reaches the Save button.”
  • Screenshots of the page after each reload, returned to your agent with its report.
  • Files in the task’s workspace. Each task gets its own folder on the spare, and report.md is written there. A published file appears under Task results on the spare’s Activity tab, where Collect File saves a copy to your Downloads folder.

When the task finishes, the agent closes the windows it opened and releases control.

What waits for you

Checking an app shouldn’t need anything risky. By default, steps ibara recognizes as sending, spending, deleting or changing access wait for your approval. The request appears in the bottom-right corner of the console in plain words, such as “claude@studio wants to click the “Send” button in Chromium on staging.example/invite, which sends something”, with Approve, Deny, Always Allow and Details. A desktop notification offers Approve, Deny and Always Allow too. An approval covers that one step. Approvals explains what they cover, and how to turn them off for your own computers.

Watch, or step in

The work happens on the spare, so you can keep working on your own computer and look in when you like. Open the spare in the console to watch its screen. Take Control to try something yourself: the agent’s input stops before yours starts. Hand Back when you’re done. The spare is free again, and the agent’s old session doesn’t restart, so ask your agent to carry on if there’s more to check.

Next#

Point your agent at the rest of your app: a sign-up flow, a dashboard, a page you’ve just changed. Keep tasks specific, say what not to touch, and ask for evidence. Once one spare is busy, connect another and give each agent a computer of its own. A task, step by step shows the tools behind each part.