Build, test and publish an app #
On this page
Prompt:
Build me a notes app with sign-in. Let me test it before you put it live.
Follow this sequence to take an app from code to a checked production version. Each phase links to the procedure your AI tool follows.
What your AI tool does #
Your AI tool writes or adapts the app, tests its behavior, and deploys it to a development environment. It promotes the tested version when you ask, then checks production.
What you need #
- An AI coding tool connected and signed in through Connect your AI assistant.
- A description of what the app should do, or access to the existing project's files.
- For local runs and tests, the software in What your computer needs. Its Windows PowerShell section gives the Windows commands.
Without a shell, use Deploy from a chat for the file workflow. Your AI tool can check the hosted app, but it reports local tests as not run.
Before your AI starts #
For the AI tool: confirm the intended behavior, the project folder, and whether this is a new or existing application. Ask the person to choose an application's size where needed; Size and usage explains the choices.
For this workflow, confirm that the person wants development before the first deploy. With nobody to ask, turn it on only if their request already asks for development. Otherwise stop and report before deploying. An application starts with production alone, so a deploy without development puts the version live.
For an unattended run or a computer using a token, follow Deploy with nobody at the keyboard. Keep the token's value out of chat.
Steps #
For the AI tool: follow these phases in order. Read each linked procedure before doing its work.
1. Start from a template or existing code #
For a new app, follow Starting a new service to choose a template and create the project. The notes-app template provides sign-in and private database notes.
For existing code, read skill:move-existing-app with read_context and use Bring an existing application to prepare the code. Keep its project layout; do not scaffold over it. Before adding a capability, read Use the library to check for a feature package and its integration instructions.
In this phase, prepare the project only. Follow phases 2 and 3 below before any linked procedure reaches a deploy.
2. Create the application and turn development on #
Read the deployment skill with read_context, id skill:deploy. Follow Create the application, reusing the application's identifier if it already exists.
Read skill:develop-then-promote and skill:release with read_context. Then follow Turn the development environment on, before deploying. Record the development hostname the response returns. Take production's hostname from a later platform response, rather than composing it. Keep each check on its intended environment.
3. Declare the services and prepare local settings #
Read skill:manifest-author with read_context, then follow Author the manifest to submit the whole declaration. For sign-in, also read skill:add-sign-in and follow Add sign-in to your app.
Before deploying, follow Store a secret to supply each secret a manifest setting references, at a scope the target environment reads. Check development and production separately; a value stored only for one does not supply the other.
For a local run, follow Run your app on your computer. Its provision command writes the development credentials into the environment file without passing their values through chat. Keep that file out of Git and follow the guide for settings the command does not write.
4. Test locally and deploy to development #
Follow Test your application locally to check the app's behavior. Start and check the real local server using Run your app on your computer, then stop the processes you started.
Fix failing checks before deployment. Follow Deploy to development and its linked upload procedure. Wait until the deploy ends successfully, following the returned status call where needed.
Without a shell, use the linked chat workflow instead. Report which local checks could not run, then perform the hosted checks in phase 5.
5. Check the task in development #
Follow Check development before promoting. Check the behavior the person requested, its saved data where applicable, and the logs. A successful health check alone does not prove the app works.
For sign-in, follow Test sign-in with test end users, explicitly choosing development. For the notes app, check that a signed-in user can save and read their notes, and another user cannot read them.
Fix a failed check and deploy the corrected version to development before repeating it. Report the checked version and results when asking the person whether to publish it.
6. Promote and check production #
Promote only when the person has asked for that promote. Otherwise ask and wait. Follow Promote to production, including its database migration cautions, and wait for the outcome.
Follow Check the release against production. That section distinguishes read-only checks, first-deploy test-user checks, and other production writes needing the owner's confirmation on an app with real users. Apply those production-write limits when checking signed-in behavior too.
Report the production address, the version serving there, the checks that passed, and anything still unchecked. If the promote fails, read its recorded outcome before choosing the recovery in the linked guide.
Expected result #
Development serves the version whose requested behavior you checked. After an authorized promote, production's status shows that version as deployed, and the production checks report what worked and what remains unchecked.
Your AI tool gives you the production address and check results. In a run without a shell, it identifies local checks as not run.
Refusals #
This table covers common refusals only. The linked procedures and the refusals reference give the other causes and remedies.
| Refusal | Cause | Remedy |
|---|---|---|
manifest_missing |
The application has no manifest on file. | Submit it using phase 3 before deploying. |
setting_secret_missing |
The target environment lacks a secret a setting needs. | Use Store a secret, then repeat the refused operation. |
deploy_in_flight |
Another deploy is still running. | Read its status and wait for its outcome before starting another. |
nothing_to_promote |
Development serves no version and the promote names none. | Complete the development deploy in phase 4 first. |
environment_not_created |
The application has production alone. | Confirm the intended workflow and turn development on using phase 2. |
For a local test failure, use Test your application locally: Refusals and the affected package's test-double instructions. If a prerequisite remains unavailable, stop and report what is needed before continuing.
Related #
- Find the guide for a task covers other application tasks.
- Deploy from your computer gives the full deploy procedure.
- Test in development, then promote explains the environment changes and production switch.
- Manage versions and environments covers version history and rollback.