Your first iPhone task

Navigate a real app, read the evidence and close the session.

Before you start

Complete CLI installation, plugin installation and device setup. Use a new, authenticated local client session with command execution and image reading enabled. A hosted chat without local Mac tools cannot operate the device.

Send the task

Invoke $iphone-use:iphone-use in Codex or /iphone-use:iphone-use in Claude Code, then send:

Open Settings on my iPhone, navigate to General, read the screenshot to verify the page, then disconnect. Do not change any settings.

Keep the phone unlocked. If several phones are connected, state which to use.

What completion looks like

1. The agent checks the CLI and Xcode, discovers the device, connects and keeps the returned session ID.
2. It discovers Settings, opens the app and reads the actual PNG with its image tool.
3. It navigates to General using coordinates from the current observation (or Runner element references). If General is already open, it goes back and enters again.
4. It reads the new screenshot, confirms the General page and disconnects cleanly.

A command success, a screenshot filename or a successful tap alone does not complete this task. No settings need to change. UI labels may differ with the phone’s language.

If the task stops

A background command job ID is different from the device session ID. Collect the original command’s completed output instead of launching it again. If an action returns unknown, observe before deciding what to do next.

Ask the agent to disconnect if possible and report the command, error and unverified result. Continue with troubleshooting.