# Sketch-First Workflow - Implementation Plan **Feature**: Intelligent sketch-driven project initialization **Agents**: Mimer (Detection), Freya (UX), Saga (Brief), Cascade (Trigger Map) **When**: User uploads sketch(es) to repository **Purpose**: Enable designers to start with sketches, build foundation retroactively **Vision**: From "15 wireframes in a folder" → "Complete WDS project with specs & prototypes" --- ## Overview This plan establishes a **sketch-first initialization flow** where users can: 1. Upload one or multiple sketches 2. Get intelligent analysis and grouping 3. Choose to build full WDS foundation or quick specs 4. End with complete documentation, navigation, and prototypes **Key Innovation**: Start with what's fun (sketching) → Agent guides to proper process --- ## Phase 1: Manual Trigger (Now - v1.0) ### **Core Principle: Human-in-Loop** All processing requires manual trigger and human confirmation at every decision point. ### **Detection (Passive Monitoring)** **What Agent Monitors:** ``` Watch folders: ├── /sketches/ ├── /docs/*/sketches/ ├── /4-scenarios/*/sketches/ └── Any configured sketch directories Detects: ├── New files added ├── Files modified (date/size changed) ├── Files renamed └── Files moved ``` **Agent Behavior:** ``` User: [drops 5 sketches in /sketches/] Mimer: "👋 Hey! I noticed some changes: **New Sketches** (5): ✨ landing-page.jpg ✨ signin.jpg ✨ dashboard.jpg ✨ profile.jpg ✨ settings.jpg Would you like me to analyze them? [Yes, analyze now] [Remind me later]" ``` **User must trigger** - No automatic processing --- ### **Single Sketch Analysis** **Workflow: workshop-a-sketch-to-page.md** ``` User: [uploads landing-page.jpg] Mimer: "I see a sketch! Activating Freya..." Freya: "Hi! Let me analyze this sketch... 🔍 Analysis: - Landing page concept - Hero, features, testimonials, CTA - Looks like SaaS product 📋 Is this part of a larger project? [A] Yes, let's set up full WDS project [B] No, just specify this page [C] Let me think about it" User: "A" Freya: "Perfect! Since this is a new project, let's build the foundation first. I'll hand you to Saga for Product Brief, then we'll come back to spec this sketch with full context. 🎯 Activating Saga..." → Product Brief workflow → Trigger Mapping workflow (if needed) → Return to Freya for page specification ``` **Human confirms:** - ✅ Sketch interpretation (sections, objects) - ✅ Project scope (full WDS vs. quick spec) - ✅ Page naming and purpose - ✅ Content and translations - ✅ Final specification --- ### **Batch Sketch Analysis** **Workflow: workshop-b-batch-sketch-analysis.md** ``` User: [uploads 15 wireframes] Mimer: "Whoa! I see 15 sketches! 🎨 Let me analyze them all... [Analyzing patterns and relationships...] 📊 **What I Found:** **Scenario 1: User Onboarding** (6 pages) 1. landing-page.jpg → Start/Hero 2. signin.jpg → Authentication 3. profile-setup.jpg → Profile creation 4. family-setup.jpg → Family info 5. dog-profile.jpg → Add dog 6. dashboard.jpg → Success/overview **Scenario 2: Daily Features** (7 pages) 7. calendar-view.jpg → Schedule ... **Standalone** (2 pages) 14. about.jpg 15. contact.jpg 🤔 **Questions:** 1. Did I group these correctly? 2. Ready to build full WDS project? 3. Want to adjust grouping first?" User: "Grouping looks good. Let's build the project!" Mimer: "Perfect! Here's the plan: **Phase 1: Foundation** (10 min) → Saga: Product Brief (using sketch 1 as context) → Cascade: Trigger Map (validating against sketches) **Phase 2: Structure** (5 min) → Create scenario folders → Map sketches to pages → Establish navigation flow **Phase 3: Specification** (iterative) → Freya analyzes each sketch → You review and refine **Phase 4: Prototypes** (automatic) → Generate interactive HTML prototypes → Link all pages together → Create clickable flows **Result:** ✅ Complete WDS project ✅ All 15 pages documented ✅ All 15 pages prototyped ✅ Navigation mapped ✅ Interactive demo ✅ Ready for development & testing Let's start! [Begin] [Adjust plan]" User: "Begin" → Full workflow execution with human confirmation at key points ``` **Human confirms:** - ✅ Scenario grouping (can adjust) - ✅ Page naming and sequence - ✅ Each sketch interpretation - ✅ Phase completion before proceeding --- ### **Iterative Page Updates** **Workflow: workshop-c-page-update.md** ``` User: [uploads landing-page-v2.jpg] Freya: "I see we already have specs for Landing Page. Let me compare this to version 1... 🔍 **Changes detected:** ✅ Hero section - unchanged ✅ Feature Cards - unchanged ➕ Feature Card 3 - NOW COMPLETE (was TBD) ✏️ Header - NEW: Search bar with filters ➕ Footer - NOW COMPLETE (was blank) Should I: [A] Update all changed sections [B] Pick specific sections to update [C] Show me detailed comparison first" User: "A" Freya: "Great! I'll update: 1. Add search bar specs to Header 2. Complete Feature Card 3 3. Generate Footer specifications Preserving existing: ✅ Hero (no changes) ✅ Feature Cards 1-2 (no changes) Ready to proceed? [Yes/Adjust]" User: "Yes" → Updates only changed sections, preserves rest ``` **Human confirms:** - ✅ Change detection accuracy - ✅ Which sections to update - ✅ New specifications - ✅ Preserved sections remain unchanged --- ### **Configuration (v1.0)** ```yaml # wds-workflow-status.yaml sketch_monitoring: enabled: true watch_folders: - "sketches/" - "4-scenarios/**/sketches/" detection_mode: "passive" # Detects but doesn't process notification_mode: "proactive" # Alerts user immediately auto_process: false # Requires manual trigger human_confirmation: "all" # Every decision needs confirmation batch_threshold: 5 # Group if 5+ sketches detected ``` --- ## Phase 2: Semi-Auto (Future - v1.5-2.0) ### **Enhanced Detection** **Smart Categorization:** ``` Agent behavior: ├── 👀 Monitor: Continuous detection ├── 🧠 Analyze: Auto-categorize when idle ├── 📋 Queue: Store results for review ├── 🔔 Notify: "Ready for your review" └── 🤝 Review: Human approves/adjusts New sketch detected: payment-flow.jpg ├─ Check: Matching page spec exists? │ ├─ YES → "Updated sketch for existing page" │ └─ NO → "New sketch, needs page creation" │ ├─ Check: Name suggests scenario? │ ├─ "payment-flow" → Could be new scenario │ └─ "landing-v2" → Update to existing │ ├─ Check: Where was it added? │ ├─ /4-scenarios/2-checkout/ → Part of existing scenario │ └─ /sketches/ → Needs placement │ └─ Confidence Level: ├─ High → Can suggest auto-processing └─ Low → Requires manual review ``` **Smart Batching:** ``` Agent groups by: ├── Related scenarios ├── Update vs. new ├── Priority (core pages first) └── Dependencies (flows must be sequential) Example: Batch 1: "Core Flow Updates" (3 sketches - HIGH priority) → Existing pages, quick updates → Estimated: 10 minutes → [Auto-process] [Review first] [Skip] Batch 2: "New Booking Feature" (8 sketches - MEDIUM priority) → New scenario, full specification → Estimated: 30 minutes → [Auto-process] [Review first] [Later] ``` **Auto-Analysis During Idle:** ``` Agent: "While you were away, I analyzed 5 sketches. Here's what I found. Please review: Sketch 1: landing-page.jpg ├── Sections: Hero, Features, CTA ├── Status: Ready for spec ├── Confidence: High ✅ └── [Approve] [Adjust] [Reject] Sketch 2: dashboard.jpg ├── Sections: Complex layout detected ├── Status: Needs clarification ├── Confidence: Medium ⚠️ └── [Manual Review Required] [Approve All High-Confidence] [Review Each]" ``` **Why Later:** - ⚠️ Requires mature AI with proven accuracy - ⚠️ Risk of incorrect auto-interpretation - ✅ Saves time on obvious cases - ✅ Human still reviews all decisions --- ### **Configuration (v1.5-2.0)** ```yaml sketch_monitoring: enabled: true detection_mode: "active" # Processes when idle notification_mode: "batch_summary" # Summarizes results auto_analysis: true # Analyzes during idle time auto_process: false # Still requires approval human_confirmation: "changes" # Only for modifications confidence_threshold: 0.85 # High confidence threshold ``` --- ## Phase 3: Full Auto (Far Future - v3.0+) ### **Autonomous Processing** **Only when:** - ⚠️ Extremely high AI accuracy (>95%) - ⚠️ User explicitly opts in - ⚠️ Well-established patterns in project - ✅ Full audit trail available - ✅ Easy rollback mechanism ``` Agent behavior: ├── 👀 Monitor: Continuous ├── 🤖 Process: Fully automatic ├── ✅ Execute: Generate specs & prototypes ├── 📧 Notify: "5 pages completed" └── 🔄 Review: "Review when ready" Example: Agent: "I processed 5 sketches automatically: ✅ 5 specifications generated ✅ 5 prototypes created ✅ Navigation updated ✅ Tests passed 📋 Review PR #47 when ready. All changes are in a branch for your review." ``` **Why Far Future:** - ⚠️ Requires extremely high accuracy - ⚠️ Less learning for user - ⚠️ Could miss important context - ⚠️ Only for mature, well-established projects --- ## Workflows to Create ### **Immediate (v1.0)** 1. **workshop-a-sketch-to-page.md** - Single sketch → page specification - Context detection (new vs. existing project) - Routes to Product Brief if needed - Full human-in-loop 2. **workshop-b-batch-sketch-analysis.md** - Multiple sketches → scenario grouping - Pattern recognition & relationship detection - Smart grouping with user confirmation - Complete project initialization 3. **workshop-c-page-update.md** - Updated sketch → incremental update - Change detection & comparison - Selective section updates - Preserve unchanged specifications 4. **page-init-lightweight.md** - Quick page setup: name, purpose, navigation - Create folder structure - Generate placeholder with navigation - Ready for sketch analysis ### **Future (v1.5+)** 5. **smart-batch-processor.md** - Auto-categorization during idle - Confidence-based suggestions - Queue for review 6. **change-detection-engine.md** - Advanced visual comparison - Semantic diff understanding - Smart merge strategies --- ## Success Metrics ### **v1.0 (Manual Trigger)** - ✅ Detection accuracy: 100% (all sketches found) - ✅ User triggers: 100% required - ✅ Confirmation points: All decisions - ✅ Time saved: 50% vs. manual documentation ### **v1.5 (Semi-Auto)** - ✅ Analysis accuracy: 85%+ - ✅ Auto-categorization: 80%+ - ✅ Human review: All results - ✅ Time saved: 70% vs. manual ### **v2.0+ (Full Auto)** - ✅ Processing accuracy: 95%+ - ✅ Auto-approval: High confidence only - ✅ Human review: Available - ✅ Time saved: 90% vs. manual --- ## Benefits ### **For Designers** - ✅ Start with fun part (sketching) - ✅ Agent builds foundation retroactively - ✅ Complete documentation from wireframes - ✅ Interactive prototypes automatically ### **For Teams** - ✅ Consistent documentation - ✅ Navigable specifications - ✅ Development-ready handoffs - ✅ Testable prototypes ### **For Process** - ✅ Lower barrier to entry - ✅ Encourages proper documentation - ✅ Validates designs against strategy - ✅ Creates institutional knowledge --- ## Implementation Priority ### **Now (Week 1-2)** 1. Passive sketch detection (Mimer) 2. Single sketch analysis workflow (Freya) 3. Page init lightweight (navigation setup) 4. Basic change detection ### **Soon (Week 3-4)** 5. Batch sketch analysis 6. Smart scenario grouping 7. Complete project initialization 8. Prototype generation integration ### **Later (Month 2-3)** 9. Auto-analysis during idle 10. Confidence-based suggestions 11. Smart batching 12. Advanced change detection --- ## Technical Requirements ### **Detection System** - File system watcher - Pattern matching (new/modified/moved) - Folder configuration (watch list) - Event throttling (batch rapid changes) ### **Analysis Engine** - Image analysis (if AI vision available) - Pattern recognition (grouping similar pages) - Relationship detection (flow connections) - Confidence scoring ### **Human-in-Loop Interface** - Clear confirmation points - Visual diffs (old vs. new) - Batch approval options - Undo/rollback capability ### **Integration Points** - Existing sketch analysis workflow (4b-sketch-analysis.md) - Product Brief workflow (Saga) - Trigger Map workflow (Cascade) - Prototype generation --- ## Configuration Example ```yaml # wds-workflow-status.yaml # Phase 1: Manual Trigger (v1.0) sketch_workflow: version: "1.0" mode: "manual-trigger" monitoring: enabled: true watch_folders: - "sketches/" - "4-scenarios/**/sketches/" detect_changes: true notify_user: true processing: auto_analyze: false # Requires manual trigger auto_process: false # Requires approval human_confirmation: "all" # Every decision point batching: enabled: true threshold: 5 # Group if 5+ sketches suggest_grouping: true # Show suggested groups require_approval: true # User must approve workflows: single_sketch: "workshop-a-sketch-to-page.md" batch_sketches: "workshop-b-batch-sketch-analysis.md" page_update: "workshop-c-page-update.md" page_init: "page-init-lightweight.md" ``` --- ## Guardrails ### **Always Required** - ✅ User triggers all processing - ✅ Human confirms all interpretations - ✅ Clear undo/rollback mechanism - ✅ Audit trail of all changes - ✅ Preserve user's work ### **Never Allowed** - ❌ Auto-process without trigger - ❌ Auto-approve without review - ❌ Overwrite without confirmation - ❌ Delete user's specifications - ❌ Skip confirmation points --- **Created**: December 28, 2025 **Feature Owner**: Freya (UX), Mimer (Detection) **Status**: Planning Complete → Ready for Implementation **Next Step**: Create v1.0 workflows (manual trigger, human-in-loop) --- ## Summary: From Sketches to Specs ``` Before: ├── 15 random wireframes in a folder ├── No documentation ├── No structure └── "Now what?" After (v1.0): ├── Complete WDS project ├── All pages documented with specs ├── Proper navigation between pages ├── Interactive prototypes ├── Strategic foundation (Brief + Trigger Map) └── Ready for development & testing Time: 1-2 hours vs. 2-3 days manual Quality: Consistent, complete, navigable Experience: Start with fun, end with foundation ``` **The Magic**: Agent detects → User triggers → Human confirms → Complete project! 🎨→📋→💻