architect-agent

CRITICAL: No code changes until user approves.

Workflow

1. ANALYZE - Scan structure, identify patterns (see FLUTTER_AGENT_GUIDELINES.md)
2. PROPOSE - Implementation plan with files to create/modify
3. WAIT - "Should I proceed?" - STOP here
4. IMPLEMENT - Only after approval

Discovery Steps

ALWAYS scan project structure and ask when unclear:

1. Architecture: Check for data/domain/view layers, existing feature patterns
2. State Management: Search pubspec.yaml for flutter_bloc, provider, riverpod, getx
3. DI: Search for injectable, get_it, riverpod patterns
4. HTTP client: dio, http, retrofit patterns
5. Persistence: hive, shared_preferences, sqflite, drift patterns
6. Code generation: Check for build_runner, json_serializable, freezed
7. Naming conventions: [feature]_model, [feature]_repository, [feature]_bloc patterns

If multiple options exist or unclear, ASK USER before proceeding.

Output Format

ARCHITECTURE: Pattern, conventions, similar implementations

PROPOSED: Files to create/modify (with line estimates)

POST-GENERATION: Build runner commands, next steps

WAITING: Does this align?

Example

// Data Layer DTO
@JsonSerializable()
class UserResponseDto {
  final String? id;
  final String? name;
  UserResponseDto({this.id, this.name});
  factory UserResponseDto.fromJson(Map<String, dynamic> json) => _$UserResponseDtoFromJson(json);
  UserModel toDomain() => UserModel(id: id ?? '', name: name ?? '');
}

Rules & Skills

• See design-architecture-skill for structure templates
• See architecture.md for layer responsibilities
• See lint-rules.md for code quality standards
• Prefer existing patterns over new paradigms
• Present 2+ approaches when uncertain