VibeView gives you ready-made prompts to paste into your coding agent (Claude Code, Codex, Cursor or another agent that can run a shell command or use MCP). The agent then does the job in your repository with the VibeView CLI, and the page you started from shows the result when it arrives, so you don’t need to refresh.
Where you find them
- Apps: Add your first app (or Add your own app above the sample apps) and, once you have an app, the hint under the table: build your app for a simulator and upload it.
- Builds: before your first cloud build, and the hint under the build list, which sets up only the platforms that aren’t set up for cloud builds yet.
- Tests: before your first test, and the CI hint under your tests. When the agent has created your tests, Continue opens them under Not in a group on the Tests page. Create test has the same prompt at its bottom: Write tests with your coding agent.
- Get started: the app step’s Build it with your agent, and the build, agent and develop-live paths. See Getting started.
On Apps, Get the prompt, in Add your first app or in the hint under the table, opens the upload dialog on its Let your agent do it tab. The dialog has two tabs, Upload a file and Let your agent do it, and the same tab is in Upload build on an app’s page. When the build arrives, the dialog closes and shows the usual upload message instead of offering Continue. See Let your coding agent build and upload it.
On the other pages, Get the prompt opens a panel with three tabs:
- Coding agent: pick your agent and click Copy prompt for your agent, then paste it into the agent running in your app’s repository. View prompt shows the full text first. Your choice of agent is remembered in this browser.
- Terminal: the same job as commands to run yourself.
- Dashboard (where there is one): do it in the dashboard instead.
The panel watches for the result (a new build, a new cloud build, new tests) and offers Continue when it arrives. If nothing arrives after a few minutes, check your agent’s output for an error. The CI prompt’s result is a workflow file in your repository, so its panel has nothing to watch.
The prompts on Apps, Builds and Tests are shown to Developers and Admins, the roles that can upload apps, start builds and write tests.
The prompts
| Prompt | What the agent does |
|---|---|
| Build my app and upload it | Builds a simulator .app (iOS, Apple TV) or an .apk (Android, Android TV), or uses your Roku channel .zip, and uploads it with vibeview upload-app. A React Native or Expo build has to run without Metro (a release build, or a debug build with its JavaScript bundle embedded). Reports the app and build IDs. |
| Set up cloud builds | Adds a build block to vibeview.json for the platforms you picked (or the ones it finds in the project, or only the ones not set up yet), checks that secret files are in .gitignore, starts one cloud build per platform without waiting on it, then follows each build’s log until it finishes. Reports the build IDs and every file it changed. |
| Check my change on a live device | Runs your app as the code is now: through vibeview dev for a React Native app, otherwise by building it and uploading the new build first. Uses the screens your latest changes touch (or the main screen) the way a user would, reports what it saw with a screenshot, and stops the session. |
| Run my React Native app live | Starts vibeview dev for your React Native app and tells you when it is running on the device. |
| Write AI tests for my main flows | Finds three important flows in your code, creates an AI test (beta) for each with vibeview create-test (steps from a file), and runs each once. It asks you for test credentials rather than inventing any, and leaves out sample apps. Each run spends usage credit, so the prompt tells the agent not to retry a failed test more than once without asking you. |
| Run my tests in CI | Asks which groups to run, then adds a GitHub Actions workflow that runs them on every pull request, following CI testing. It shows you the workflow before committing. |
Signing in
Prompts never contain a token. The agent runs npx vibeview login, which prints a sign-in link and opens your browser; you finish signing in there within 2 minutes. If your agent can’t open a browser, it shows you the link. The CI prompt asks you to add the VIBEVIEW_API_TOKEN repository secret yourself; the agent never writes a token into a file.
Claude Code prompts tell the agent to run npx vibeview agent-setup first if the VibeView skill isn’t installed yet. Other agents use the CLI through npx vibeview, or the MCP server (npx vibeview mcp). For how an agent drives a device once connected, see AI agent control.
Tests over MCP
Over MCP an agent can also create and run tests itself: create_test, run_test, run_group and get_run. A run with no session_id starts its own new device and stops it afterwards; pass session_id to run on the device the agent is driving. See AI agent control.