Files
buzz/mobile/test/features/agents/create_agent_page_test.dart
Wes 1a30e51689 mobile: start new agents on the server-recommended model (#9151)
## Summary

Builderlab's model catalog can now recommend a model and thinking effort
for new agents (`recommended_selection` on
`/v3/beekeeper/get-model-catalog`). The mobile create-agent screen now
starts on that recommendation, so the default can change on the server
without an app release.

New agents start on the first of these that applies:

1. The model an agent link suggests, when the catalog offers it
(unchanged).
2. The server's recommended model and effort.
3. The first Sonnet model at medium effort, as added in #9034. Used
while a deployment sends no recommendation.
4. Blank.

A setup resumed after a failure still keeps its saved choice. A
recommendation that isn't in the returned catalog, or that is malformed,
is ignored. Existing agents' saved models are never changed.

The server only sends a recommendation when the deployment's catalog
config sets `recommended_selection`. Until then, the app keeps falling
back to Sonnet.

## Testing

- Unit tests cover the recommendation beating the Sonnet fallback, a
draft model beating the recommendation, and recommendations that should
be ignored (unsupported effort, unknown model, missing effort,
malformed, absent).
- A new widget test checks that the form opens on the recommended model
and effort instead of Sonnet.
- `just mobile-check` and the full mobile suite (`flutter test
--dart-define=BUZZ_PUSH_GATEWAY_URL=https://push.example`) pass locally.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Signed-off-by: Wes <wesbillman@users.noreply.github.com>
Co-authored-by: Carl <32a2e2c9d428ee08902cab75d956da2c1d235a22d4766b0dd4138bf6e2e5db1d@buzz.block.builderlab.xyz>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-05 18:00:58 -07:00

325 lines
11 KiB
Dart

import 'dart:io';
import 'package:buzz/shared/widgets/frosted_app_bar.dart';
import 'package:buzz/shared/widgets/preset_agent_avatar.dart';
import 'package:buzz/shared/agents/agent_avatar_catalog.dart';
import 'package:buzz/features/agents/agent_avatar_editor_page.dart';
import 'dart:async';
import 'package:buzz/shared/agents/remote_agents.dart';
import 'package:buzz/features/agents/agents_page.dart';
import 'package:buzz/features/agents/create_agent_page.dart';
import 'package:flutter/material.dart';
import 'package:flutter_test/flutter_test.dart';
import 'package:nostr/nostr.dart' as nostr;
import 'agents_harness.dart';
void main() {
final existing = existingDesktopKeys.public;
testWidgets(
'creation overlays the retained list, then failure and resume both land',
(tester) async {
final h = await AgentsHarness.create();
final controller = h.controller;
final added = nostr.Keys('0' * 63 + '2').public;
await h.open(tester, directory: [existing]);
final list = tester.element(find.byType(AgentsPage));
if (!supportsRemoteAgents(h.community)) {
expect(find.byTooltip('Add agent'), findsNothing);
expect(find.text('Sign in to create an agent'), findsNothing);
expect(find.byKey(ValueKey(existing)), findsOneWidget);
return;
}
await tester.tap(find.byTooltip('Add agent'));
await tester.pumpAndSettle();
expect(find.byType(CreateAgentPage), findsOneWidget);
final createContext = tester.element(find.byType(CreateAgentPage));
expect(Navigator.of(list), same(Navigator.of(createContext)));
expect(find.text('Thinking'), findsOneWidget);
expect(find.byIcon(Icons.unfold_more), findsNWidgets(2));
expect(
tester.getTopLeft(find.text('Name')).dy,
lessThan(tester.getTopLeft(find.text('Model')).dy),
);
expect(
tester.getTopLeft(find.text('Model')).dy,
lessThan(tester.getTopLeft(find.text('Instructions')).dy),
);
await tester.ensureVisible(find.text('Model'));
await tester.tap(find.text('Model'));
await tester.pumpAndSettle();
await tester.tap(find.text('Test model').last);
await tester.pumpAndSettle();
await tester.tap(find.text('Thinking'));
await tester.pumpAndSettle();
await tester.tap(find.text('high').last);
await tester.pumpAndSettle();
// Switching models must clear an effort unsupported by the new choice.
await tester.tap(find.text('Test model'));
await tester.pumpAndSettle();
await tester.tap(find.text('Other model').last);
await tester.pumpAndSettle();
expect(
tester
.widget<IconButton>(find.widgetWithIcon(IconButton, Icons.check))
.onPressed,
isNull,
);
await tester.tap(find.text('Other model'));
await tester.pumpAndSettle();
await tester.tap(find.text('Test model').last);
await tester.pumpAndSettle();
await tester.tap(find.text('Thinking'));
await tester.pumpAndSettle();
await tester.tap(find.text('high').last);
await tester.pumpAndSettle();
controller.gate = Completer<void>();
await tester.enterText(
find.descendant(
of: find.byWidgetPredicate(
(widget) =>
widget is Semantics && widget.properties.label == 'Name',
),
matching: find.byType(TextField),
),
'Scout',
);
await tester.enterText(
find.descendant(
of: find.byWidgetPredicate(
(widget) =>
widget is Semantics &&
widget.properties.label == 'Instructions',
),
matching: find.byType(TextField),
),
'Reply concisely.',
);
await tester.pump();
final originalAvatar = tester
.widget<PresetAgentAvatar>(find.byType(PresetAgentAvatar))
.imageUrl;
final originalPack = AgentAvatarPack.values.firstWhere(
(pack) => pack.indices.any(
(i) => agentAvatarCatalog[i].url == originalAvatar,
),
);
final targetIndex = originalPack.indices.firstWhere(
(i) => agentAvatarCatalog[i].url != originalAvatar,
);
await tester.ensureVisible(find.text('Choose avatar'));
// Repeated callbacks before the next rebuild must still open one editor.
final openAvatar = tester
.widget<FilledButton>(
find.widgetWithText(FilledButton, 'Choose avatar'),
)
.onPressed!;
// A stalled CDN must not prevent entry to the editor.
final stalledClient = _StalledPosterClient();
debugNetworkImageHttpClientProvider = () => stalledClient;
addTearDown(() => debugNetworkImageHttpClientProvider = null);
await NetworkImage(originalAvatar).evict();
openAvatar();
openAvatar();
await tester.pump();
await tester.pump(const Duration(milliseconds: 301));
await tester.pumpAndSettle();
expect(stalledClient.requests, greaterThan(0));
debugNetworkImageHttpClientProvider = null;
expect(find.byType(AgentAvatarEditorPage), findsOneWidget);
expect(find.text('Instructions'), findsNothing);
await tester.tap(find.byKey(ValueKey('agent-avatar-$targetIndex')));
await tester.pumpAndSettle();
// This harness disables motion, so the return transition completes
// immediately. Check suppression while editing, then restoration on exit.
final chooseButton = find.widgetWithText(
FilledButton,
'Choose avatar',
skipOffstage: false,
);
expect(tester.widget<FilledButton>(chooseButton).onPressed, isNull);
final playback = find.byKey(
const ValueKey('create-agent-avatar-playback'),
skipOffstage: false,
);
expect(tester.widget<TickerMode>(playback).enabled, isFalse);
final sourceBar = find.descendant(
of: find.byType(CreateAgentPage, skipOffstage: false),
matching: find.byType(FrostedAppBar, skipOffstage: false),
skipOffstage: false,
);
expect(
tester.widget<FrostedAppBar>(sourceBar).nativeViewSuppressed?.value,
isTrue,
);
await tester.tap(find.byTooltip('Cancel'));
await tester.pumpAndSettle();
expect(tester.widget<FilledButton>(chooseButton).onPressed, isNotNull);
expect(tester.widget<TickerMode>(playback).enabled, isTrue);
expect(
tester.widget<FrostedAppBar>(sourceBar).nativeViewSuppressed?.value,
isFalse,
);
expect(
tester
.widget<PresetAgentAvatar>(find.byType(PresetAgentAvatar))
.imageUrl,
originalAvatar,
);
await tester.tap(find.text('Choose avatar'));
await tester.pumpAndSettle();
await tester.tap(find.byKey(ValueKey('agent-avatar-$targetIndex')));
await tester.pumpAndSettle();
await tester.tap(find.byTooltip('Done'));
await tester.pumpAndSettle();
expect(
tester
.widget<PresetAgentAvatar>(find.byType(PresetAgentAvatar))
.imageUrl,
isNot(originalAvatar),
);
expect(find.text('Scout'), findsOneWidget);
expect(find.text('Reply concisely.'), findsOneWidget);
await tester.tap(find.byTooltip('Create'));
await tester.pump(const Duration(milliseconds: 100));
expect(controller.busy, isTrue);
expect(
find.descendant(
of: find.byType(CreateAgentPage),
matching: find.byType(LinearProgressIndicator),
),
findsOneWidget,
);
expect(
tester
.widget<IconButton>(
find.ancestor(
of: find.byTooltip('Create'),
matching: find.byType(IconButton),
),
)
.onPressed,
isNull,
);
int catalogLoads() =>
h.registry.calls.where((call) => call == 'get-model-catalog').length;
final loadsBeforeFailure = catalogLoads();
controller.gate!.complete();
await tester.pumpAndSettle();
expect(find.byType(CreateAgentPage), findsOneWidget);
// A rejected choice may be stale, so the catalog reloads.
expect(catalogLoads(), loadsBeforeFailure + 1);
expect(find.byType(CreateAgentPage), findsOneWidget);
expect(find.text('Creation failed. Retry.'), findsOneWidget);
expect(controller.submittedName, 'Scout');
expect(controller.submittedInstructions, 'Reply concisely.');
// The real registry flow then succeeds and returns to the same list.
controller.gate = null;
await tester.tap(find.byTooltip('Create'));
await tester.pumpAndSettle();
expect(find.byType(BottomSheet), findsNothing);
expect(tester.element(find.byType(AgentsPage)), same(list));
expect(find.byKey(ValueKey(added)), findsOneWidget);
expect(h.registry.calls, contains('publish-policy'));
},
);
testWidgets('a new agent starts on the affordable default model', (
tester,
) async {
final h = await AgentsHarness.create();
h.registry.models = [
{
'id': 'premium-model',
'display_name': 'Premium model',
'supported_reasoning_efforts': ['low', 'medium', 'high'],
},
{
'id': 'kgoose-claude-sonnet-4-6',
'display_name': 'Sonnet',
'supported_reasoning_efforts': ['low', 'medium', 'high'],
},
];
await h.open(tester, directory: [existing]);
if (!supportsRemoteAgents(h.community)) return;
await tester.tap(find.byTooltip('Add agent'));
await tester.pumpAndSettle();
expect(find.text('Sonnet'), findsOneWidget);
expect(find.text('Premium model'), findsNothing);
expect(find.text('medium'), findsOneWidget);
// Naming it is all that's left before Create.
await tester.enterText(
find.descendant(
of: find.byWidgetPredicate(
(widget) => widget is Semantics && widget.properties.label == 'Name',
),
matching: find.byType(TextField),
),
'Scout',
);
await tester.pump();
expect(
tester
.widget<IconButton>(
find.ancestor(
of: find.byTooltip('Create'),
matching: find.byType(IconButton),
),
)
.onPressed,
isNotNull,
);
});
testWidgets('a new agent starts on the service-recommended model', (
tester,
) async {
final h = await AgentsHarness.create();
h.registry.models = [
{
'id': 'premium-model',
'display_name': 'Premium model',
'supported_reasoning_efforts': ['low', 'medium', 'high'],
},
{
'id': 'kgoose-claude-sonnet-4-6',
'display_name': 'Sonnet',
'supported_reasoning_efforts': ['low', 'medium', 'high'],
},
];
h.registry.recommendedSelection = {
'model_id': 'premium-model',
'reasoning_effort': 'high',
};
await h.open(tester, directory: [existing]);
if (!supportsRemoteAgents(h.community)) return;
await tester.tap(find.byTooltip('Add agent'));
await tester.pumpAndSettle();
expect(find.text('Premium model'), findsOneWidget);
expect(find.text('Sonnet'), findsNothing);
expect(find.text('high'), findsOneWidget);
});
}
class _StalledPosterClient implements HttpClient {
int requests = 0;
@override
Future<HttpClientRequest> getUrl(Uri url) {
requests++;
return Completer<HttpClientRequest>().future;
}
@override
dynamic noSuchMethod(Invocation invocation) => null;
}