Explore app
Let the AI explore your app like a user and report what it finds.
Explore app lets the AI try your app like a user. At each step Astra Scout reads the screen and builds a finite list of safe actions, the AI picks one with a reason, and Astra Scout runs it.
Start an exploration
Choose Device → Explore the App… (Shift+Cmd+E), or click the compass button (Explore) under the device mirror. Exploration runs in its own Explore app window. Then choose:
| Option | Meaning |
|---|---|
| App (appId) | The app to explore |
| Persona | Newcomer, speedrunner, chaos monkey, hurried user, privacy-conscious user or feature explorer |
| Goal (optional) | What to reach, for example "open General → About". Values in quotes are typed into inputs. |
| Max steps | Default 40 |
| Max minutes | Default 10 |
Click Start exploring. A device must be connected. Stop ends it early, and Start a new exploration clears the result.
What the AI can do
The AI picks from a list that Astra Scout builds: tap an element, fill in test data (such as test@example.com), scroll, go back, wait or finish. Astra Scout runs the action itself.
Safety
Purchase buttons and paywalls are offered to the AI labelled "(purchase – sandbox)", so only explore on test devices. Password fields and card fields are never typed into and appear in the Excluded tab with labels such as "excluded (password)".
Exploration stops when it runs out of steps or time, makes no progress (same screen and action three times), the app crashes, or the AI says it is done.
Results
Results are saved with the run's video. The tabs are:
- Map: the screens it found and how often it visited each.
- Journeys: the paths it took.
- Issues: findings grouped by kind, then by severity.
- Suggested flows: a Maestro flow for each journey. Tick the ones you want and click Save N selected flows to write them to the workspace. Nothing is saved unless you choose.
- Log and Excluded.
Each finding has a kind, a severity from 1 (minor) to 5 (blocks the user), the screen, expected and actual behavior, steps and a screenshot.
| Kind | Examples | Fails the run? |
|---|---|---|
| Issue (functional) | Crash, error dialog, a UI defect the AI judges to block the user | Yes |
| Warning (UI or experience) | Cosmetic defects, stuck on a screen, tap not responding, action failed | No |
An exploration has a verdict like a flow run (Passed, Failed, Blocked, Incomplete) with a reason code. UI defects are checked with an assertNoDefectsWithAI style check at most once per screen.
Click Report Bug to Jira… or choose Device → Report Bug to Jira… (Shift+Cmd+B) to send findings. See Jira.
Purchases
Purchase buttons and paywalls are offered to the AI like any other control, labelled "(purchase – sandbox)". Run Explore on test devices only. On a password dialog, or after a purchase tap on a real iPhone, exploration stops in the Tester needed state so you can continue on the device.
From a terminal
The MCP tools explore_app and get_explore (the plugin's explore-app skill) run explorations from an AI CLI. Suggested flows are only returned, and saving needs save_flow after you agree. See The scout plugin.