- Replace content-aware extractCleanTemplate with structure-only extractStructureSignature for template comparison
- Add extractTemplateForStorage to preserve actual content for meaningful template display
- Update generateTemplateSignature to use purely structural comparison ignoring text content
- Remove redundant extractClassSignature function (functionality moved to extractStructureSignature)
- Resolves issue where identical DOM structures created multiple templates due to content differences
- Knowledge cards and other collections now correctly deduplicate to single templates while preserving content for previews
This commit addresses multiple collection management issues to improve user experience:
## Template Selection Modal Improvements
- Replace inline styles with CSS classes for reliable visual feedback
- Fix default template selection conflicts that showed multiple templates as selected
- Add styled template previews that show actual CSS styling differences
- Improve modal responsiveness and visual hierarchy
## Collection Item Creation Fixes
- Fix empty collection items with no content/height that were unclickable
- Preserve template placeholder content during item creation instead of clearing it
- Implement proper positioning system using GetMaxPosition to place new items at collection end
- Add position calculation logic to prevent new items from jumping to beginning
## Backend Positioning System
- Add GetMaxPosition method to all repository implementations (SQLite, PostgreSQL, HTTPClient)
- Update CreateCollectionItemFromTemplate to calculate correct position (maxPos + 1)
- Maintain reconstruction ordering by position ASC for consistent item placement
## Frontend Template Selection
- CSS class-based selection states replace problematic inline style manipulation
- Template previews now render actual HTML with real page styling
- Improved hover states and selection visual feedback
- Fixed auto-selection interference with user interaction
These changes ensure collection items appear in expected order and template selection
provides clear visual feedback with actual styling previews.
Restore missing content hydration logic in reconstructCollectionItems method that was accidentally removed during the engine file split (b46f643). Collection items were appearing empty instead of displaying original developer content. This fix restores the database-first behavior where content is properly extracted, stored, and injected back into .insertr elements within collection items.
Split monolithic engine.go (776 lines) into specialized files:
- engine.go: Core orchestration (142 lines, 82% reduction)
- collection.go: Collection processing and management (445 lines)
- content.go: Content injection and extraction (152 lines)
- discovery.go: Element discovery and DOM traversal (85 lines)
Benefits:
- Single responsibility principle applied to each file
- Better code organization and navigation
- Improved testability of individual components
- Easier team development and code reviews
- Maintained full API compatibility with no breaking changes
