FIELD GUIDE / 04
Create a play with an AI agent
Start with a clear idea, let an AI coding agent develop and build it, then install the play on a real AI Passport.
STEP / 01
Start with a problem worth solving
The small display, physical buttons, and wearable form can support learning tools, mini games, timers, identity displays, and more. Pick a real context and describe in one sentence what the user should accomplish after picking up the device.

- Context: when and by whom will it be used?
- Goal: what action means the experience succeeded?
- Scope: keep only one core experience in the first version.
STEP / 02
Try an official play and confirm the connection
Connect AI Passport with a USB data cable, choose a published release from the official play library, and install it once. This confirms that the cable, browser serial permission, and device connection work before custom development begins.

This does not replace development. It gives the custom play workflow a known-good connection baseline.
STEP / 03
Define the smallest version of your play
Reduce the idea to a first version that can be tested on real hardware. Specify what appears on the display, what each of the three buttons does, and whether the experience needs sound, Bluetooth, networking, or local storage.

- Interface: list the main screens, text, and states
- Input: define short- and long-press behavior for Up, Down, and Confirm
- Feedback: describe success, failure, loading, and exit behavior
- Constraints: include display size, offline needs, asset limits, and target hardware
STEP / 04
Choose an AI coding agent
Use TRAE, Codex, Claude Code, or another AI coding agent that can read a project, edit code, and run build commands. Open the play project directory and have the agent inspect its instructions, examples, and build workflow before making changes.
Please download and use this project as a reference:
https://github.com/folotoy/ai-passport
Help me build a wooden-fish tapping app.
Do not work from an isolated code snippet. Giving the agent the complete project lets it reuse the correct dependencies, hardware configuration, and build process.
STEP / 05
Describe the requirements clearly
Give the agent the context, interface, button behavior, and constraints you prepared, then ask it to restate the design and implementation plan. Specific input makes the first build much closer to a usable experience.

- What I am making and who it is for
- The first screen and the complete interaction flow
- How each button behaves in each state
- Existing code, assets, or interfaces that must be reused
- Definition of done: a successful build and an installable play file
STEP / 06
Let the agent develop, build, and check its work
Have the agent implement the plan, run the project's build commands, and continue fixing errors. At the end, ask for a summary of changes, test results, the play file path, and any remaining limitations.

Generated source code is not a finished result. Confirm that the build succeeds and that the project produced an actual play file.
STEP / 07
Install the play and test real hardware
Open the local play installer, select the play built by the agent, and connect AI Passport. Check the file and device before installation. Restart when it finishes, then test the display, buttons, sound, and connectivity one by one.

- Do not unplug the cable or close the page during installation
- Use a development device first and keep a recoverable official play
- Record unexpected screens, button behavior, and logs, then give them back to the agent
STEP / 08
Use it in context, then iterate
Try the play away from the computer. Check whether text is readable, one-handed controls feel natural, feedback is timely, and battery and connectivity fit the context. Prioritize the issues that most affect use, then ask the agent for another focused build.

- Fix problems that block the core flow first
- Change only a small set of testable details per iteration
- Keep a version number, source commit, and play backup for every usable build