Skip to content

Design

The Design feature provides a pipeline for managing design assets, extracting design tokens, and defining interaction specifications. Upload screenshots, style guides, prototypes, and animation snippets — then let AI analyze them and cascade design context into task execution.

AI agents produce better code when they understand the design intent. The design context pipeline ensures that token values, interaction patterns, and visual references are available to agents during sandbox execution, reducing back-and-forth between design and implementation.

  • Asset types: screenshot, tokens, styleGuide, prototype, interactionSpec, animationSnippet
  • Asset status: uploaded, analyzing, analyzed, error
  • Upload zone: DesignUploadZone handles drag-and-drop upload of design assets
  • Upload queue: useDesignUploadQueue manages concurrent uploads with progress tracking
  • Gallery and hierarchy views: Switch between gallery (visual grid) and hierarchy (structured tree) views
  • AI analysis panel: DesignAnalysisPanel shows AI-generated analysis of design assets with extracted tokens and patterns
  • Token editor: DesignTokenEditor provides a visual editor for design token values (colors, spacing, typography)
  • Interaction spec table: InteractionSpecTable defines interaction patterns with trigger, behavior, and duration
  • Workstream and requirement scoping: Design assets can be filtered by workstream and linked to specific requirements
  • Token cascading: Design tokens flow from program-level defaults through workstream overrides to task-specific values
  1. Navigate to [Program] > Design.
  2. Upload design assets by dragging files into the upload zone.
  3. AI analysis runs automatically — results appear in the analysis panel.
  4. Switch between gallery and hierarchy views to browse assets.
  5. Open the Token Editor to review and adjust extracted token values.
  6. Define interaction specs in the Interaction Spec Table.
  7. Scope assets to workstreams or requirements for targeted injection into sandboxes.

This feature uses the following tables: