Claude-Code-Workflow/.claude/commands/workflow/ui-design/explore-auto.md

name: explore-auto description: Interactive exploratory UI design workflow with style-centric batch generation, creates design variants from prompts/images with parallel execution and user selection argument-hint: "[--input ""] [--targets ""] [--target-type "page|component"] [--session ] [--style-variants ] [--layout-variants ] [--batch-plan]" allowed-tools: SlashCommand(), TodoWrite(), Read(), Bash(), Glob(), Write(), Task(conceptual-planning-agent)

UI Design Auto Workflow Command

Overview & Execution Model

Fully autonomous orchestrator: Executes all design phases sequentially from style extraction to design integration, with optional batch planning.

Unified Target System: Generates style_variants × layout_variants × targets prototypes, where targets can be:

• Pages (full-page layouts): home, dashboard, settings, etc.
• Components (isolated UI elements): navbar, card, hero, form, etc.
• Mixed: Can combine both in a single workflow

Autonomous Flow (⚠️ CONTINUOUS EXECUTION - DO NOT STOP):

1. User triggers: /workflow:ui-design:explore-auto [params]
2. Phase 0c: Target confirmation → User confirms → IMMEDIATELY triggers Phase 1
3. Phase 1 (style-extract) → Attach tasks → Execute → Collapse → Auto-continues to Phase 2.3
4. Phase 2.3 (animation-extract, conditional):
   • IF should_extract_animation: Attach tasks → Execute → Collapse → Auto-continues to Phase 2.5
   • ELSE: Skip (use code import) → Auto-continues to Phase 2.5
5. Phase 2.5 (layout-extract) → Attach tasks → Execute → Collapse → Auto-continues to Phase 3
6. Phase 3 (ui-assembly) → Attach tasks → Execute → Collapse → Auto-continues to Phase 4
7. Phase 4 (design-update) → Attach tasks → Execute → Collapse → Auto-continues to Phase 5 (if --batch-plan)
8. Phase 5 (batch-plan, optional) → Reports completion

Phase Transition Mechanism:

• Phase 0c (User Interaction): User confirms targets → IMMEDIATELY triggers Phase 1
• Phase 1-5 (Autonomous): SlashCommand invocation ATTACHES tasks to current workflow
• Task Execution: Orchestrator EXECUTES these attached tasks itself
• Task Collapse: After tasks complete, collapse them into phase summary
• Phase Transition: Automatically execute next phase after collapsing
• No additional user interaction after Phase 0c confirmation

Auto-Continue Mechanism: TodoWrite tracks phase status with dynamic task attachment/collapse. After executing all attached tasks, you MUST immediately collapse them, restore phase summary, and execute the next phase. No user intervention required. The workflow is NOT complete until reaching Phase 4 (or Phase 5 if --batch-plan).

Task Attachment Model: SlashCommand invocation is NOT delegation - it's task expansion. The orchestrator executes these attached tasks itself, not waiting for external completion.

Target Type Detection: Automatically inferred from prompt/targets, or explicitly set via --target-type.

Core Rules

1. Start Immediately: TodoWrite initialization → Phase 1 execution
2. No Preliminary Validation: Sub-commands handle their own validation
3. Parse & Pass: Extract data from each output for next phase
4. Default to All: When selecting variants/prototypes, use ALL generated items
5. Track Progress: Update TodoWrite dynamically with task attachment/collapse pattern
6. ⚠️ CRITICAL: Task Attachment Model - SlashCommand invocation ATTACHES tasks to current workflow. Orchestrator EXECUTES these attached tasks itself, not waiting for external completion. This is NOT delegation - it's task expansion.
7. ⚠️ CRITICAL: DO NOT STOP - This is a continuous multi-phase workflow. After executing all attached tasks, you MUST immediately collapse them and execute the next phase. Workflow is NOT complete until Phase 4 (or Phase 5 if --batch-plan).

Parameter Requirements

Recommended Parameter:

• --input "<value>": Unified input source (auto-detects type)
  • Glob pattern (images): "design-refs/*", "screenshots/*.png"
  • File/directory path (code): "./src/components", "/path/to/styles"
  • Text description (prompt): "modern dashboard with 3 styles", "minimalist design"
  • Combination: "design-refs/* modern dashboard" (glob + description)
  • Multiple inputs: Separate with | → "design-refs/*|modern style"

Detection Logic:

• Contains * or matches existing files → glob pattern (images)
• Existing file/directory path → code import
• Pure text without paths → design prompt
• Contains | separator → multiple inputs (glob|prompt or path|prompt)

Legacy Parameters (deprecated, use --input instead):

• --images "<glob>": Reference image paths (shows deprecation warning)
• --prompt "<description>": Design description (shows deprecation warning)

Optional Parameters (all have smart defaults):

• --targets "<list>": Comma-separated targets (pages/components) to generate (inferred from prompt/session if omitted)
• --target-type "page|component|auto": Explicitly set target type (default: auto - intelligent detection)
• --device-type "desktop|mobile|tablet|responsive|auto": Device type for layout optimization (default: auto - intelligent detection)
  • Desktop: 1920×1080px - Mouse-driven, spacious layouts
  • Mobile: 375×812px - Touch-friendly, compact layouts
  • Tablet: 768×1024px - Hybrid touch/mouse layouts
  • Responsive: 1920×1080px base with mobile-first breakpoints
• --session <id>: Workflow session ID (standalone mode if omitted)
• --style-variants <count>: Style variants (default: inferred from prompt or 3, range: 1-5)
• --layout-variants <count>: Layout variants per style (default: inferred or 3, range: 1-5)
• --batch-plan: Auto-generate implementation tasks after design-update

Legacy Target Parameters (maintained for backward compatibility):

• --pages "<list>": Alias for --targets with --target-type page
• --components "<list>": Alias for --targets with --target-type component

Input Rules:

• Must provide: --input OR (legacy: --images/--prompt) OR --targets
• --input can combine multiple input types
• If --targets not provided, intelligently inferred from prompt/session

Supported Target Types:

• Pages (full layouts): home, dashboard, settings, profile, login, etc.
• Components (UI elements):
  • Navigation: navbar, header, menu, breadcrumb, tabs, sidebar
  • Content: hero, card, list, table, grid, timeline
  • Input: form, search, filter, input-group
  • Feedback: modal, alert, toast, badge, progress
  • Media: gallery, carousel, video-player, image-card
  • Other: footer, pagination, dropdown, tooltip, avatar

Intelligent Prompt Parsing: Extracts variant counts from natural language:

• "Generate 3 style variants" → --style-variants 3
• "2 layout options" → --layout-variants 2
• "Create 4 styles with 2 layouts each" → --style-variants 4 --layout-variants 2
• Explicit flags override prompt inference

Execution Modes

Matrix Mode (style-centric):

• Generates style_variants × layout_variants × targets prototypes
• Phase 1: style_variants complete design systems (extract)
• Phase 2: Layout templates extraction (layout-extract)
• Phase 3: Style-centric batch generation (generate)
  • Sub-phase 1: targets × layout_variants target-specific layout plans
  • Sub-phase 2: S style-centric agents (each handles L×T combinations)
  • Sub-phase 3: style_variants × layout_variants × targets final prototypes
  • Performance: Efficient parallel execution with S agents
  • Quality: HTML structure adapts to design_attributes
  • Pages: Full-page layouts with complete structure
  • Components: Isolated elements with minimal wrapper

Integrated vs. Standalone:

• --session flag determines session integration or standalone execution

6-Phase Execution

Phase 0a: Parameter Parsing & Input Detection

# Step 0: Parse and normalize parameters
images_input = null
prompt_text = null

# Handle legacy parameters with deprecation warning
IF --images OR --prompt:
    WARN: "⚠️  DEPRECATION: --images and --prompt are deprecated. Use --input instead."
    WARN: "   Example: --input \"design-refs/*\" or --input \"modern dashboard\""
    images_input = --images
    prompt_text = --prompt

# Parse unified --input parameter
IF --input:
    # Split by | separator for multiple inputs
    input_parts = split(--input, "|")

    FOR part IN input_parts:
        part = trim(part)

        # Detection logic
        IF contains(part, "*") OR glob_matches_files(part):
            # Glob pattern detected → images
            images_input = part
        ELSE IF file_or_directory_exists(part):
            # File/directory path → will be handled in code detection
            IF NOT prompt_text:
                prompt_text = part
            ELSE:
                prompt_text = prompt_text + " " + part
        ELSE:
            # Pure text → prompt
            IF NOT prompt_text:
                prompt_text = part
            ELSE:
                prompt_text = prompt_text + " " + part

# Step 1: Detect design source from parsed inputs
code_files_detected = false
code_base_path = null
has_visual_input = false

IF prompt_text:
    # Extract potential file paths from prompt
    potential_paths = extract_paths_from_text(prompt_text)
    FOR path IN potential_paths:
        IF file_or_directory_exists(path):
            code_files_detected = true
            code_base_path = path
            BREAK

IF images_input:
    # Check if images parameter points to existing files
    IF glob_matches_files(images_input):
        has_visual_input = true

# Step 2: Determine design source strategy
design_source = "unknown"
IF code_files_detected AND has_visual_input:
    design_source = "hybrid"  # Both code and visual
ELSE IF code_files_detected:
    design_source = "code_only"  # Only code files
ELSE IF has_visual_input OR --prompt:
    design_source = "visual_only"  # Only visual/prompt
ELSE:
    ERROR: "No design source provided (code files, images, or prompt required)"
    EXIT 1

STORE: design_source, code_base_path, has_visual_input

Phase 0a-2: Intelligent Prompt Parsing

# Parse variant counts from prompt or use explicit/default values
IF prompt_text AND (NOT --style-variants OR NOT --layout-variants):
    style_variants = regex_extract(prompt_text, r"(\d+)\s*style") OR --style-variants OR 3
    layout_variants = regex_extract(prompt_text, r"(\d+)\s*layout") OR --layout-variants OR 3
ELSE:
    style_variants = --style-variants OR 3
    layout_variants = --layout-variants OR 3

VALIDATE: 1 <= style_variants <= 5, 1 <= layout_variants <= 5

# Interactive mode (always enabled)
interactive_mode = true  # Always use interactive mode

Phase 0a-2: Device Type Inference

# Device type inference
device_type = "auto"

# Step 1: Explicit parameter (highest priority)
IF --device-type AND --device-type != "auto":
    device_type = --device-type
    device_source = "explicit"
ELSE:
    # Step 2: Prompt analysis
    IF prompt_text:
        device_keywords = {
            "desktop": ["desktop", "web", "laptop", "widescreen", "large screen"],
            "mobile": ["mobile", "phone", "smartphone", "ios", "android"],
            "tablet": ["tablet", "ipad", "medium screen"],
            "responsive": ["responsive", "adaptive", "multi-device", "cross-platform"]
        }
        detected_device = detect_device_from_prompt(prompt_text, device_keywords)
        IF detected_device:
            device_type = detected_device
            device_source = "prompt_inference"

    # Step 3: Target type inference
    IF device_type == "auto":
        # Components are typically desktop-first, pages can vary
        device_type = target_type == "component" ? "desktop" : "responsive"
        device_source = "target_type_inference"

STORE: device_type, device_source

Device Type Presets:

• Desktop: 1920×1080px - Mouse-driven, spacious layouts
• Mobile: 375×812px - Touch-friendly, compact layouts
• Tablet: 768×1024px - Hybrid touch/mouse layouts
• Responsive: 1920×1080px base with mobile-first breakpoints

Detection Keywords:

• Prompt contains "mobile", "phone", "smartphone" → mobile
• Prompt contains "tablet", "ipad" → tablet
• Prompt contains "desktop", "web", "laptop" → desktop
• Prompt contains "responsive", "adaptive" → responsive
• Otherwise: Inferred from target type (components→desktop, pages→responsive)

Phase 0b: Run Initialization & Directory Setup

design_id = "design-run-$(date +%Y%m%d)-$RANDOM"
relative_base_path = --session ? ".workflow/WFS-{session}/${design_id}" : ".workflow/.design/${design_id}"

# Create directory and convert to absolute path
Bash(mkdir -p "${relative_base_path}/style-extraction")
Bash(mkdir -p "${relative_base_path}/prototypes")
base_path=$(cd "${relative_base_path}" && pwd)

Write({base_path}/.run-metadata.json): {
  "design_id": "${design_id}", "session_id": "${session_id}", "timestamp": "...",
  "workflow": "ui-design:auto",
  "architecture": "style-centric-batch-generation",
  "parameters": { "style_variants": ${style_variants}, "layout_variants": ${layout_variants},
                  "targets": "${inferred_target_list}", "target_type": "${target_type}",
                  "prompt": "${prompt_text}", "images": "${images_input}",
                  "input": "${--input}",
                  "device_type": "${device_type}", "device_source": "${device_source}" },
  "status": "in_progress",
  "performance_mode": "optimized"
}

# Initialize default flags for animation extraction logic
animation_complete = false  # Default: always extract animations unless code import proves complete
needs_visual_supplement = false  # Will be set to true in hybrid mode
skip_animation_extraction = false  # User preference for code import scenario

Phase 0c: Unified Target Inference with Intelligent Type Detection

# Priority: --pages/--components (legacy) → --targets → --prompt analysis → synthesis → default
target_list = []; target_type = "auto"; target_source = "none"

# Step 1-2: Explicit parameters (legacy or unified)
IF --pages: target_list = split(--pages); target_type = "page"; target_source = "explicit_legacy"
ELSE IF --components: target_list = split(--components); target_type = "component"; target_source = "explicit_legacy"
ELSE IF --targets:
    target_list = split(--targets); target_source = "explicit"
    target_type = --target-type != "auto" ? --target-type : detect_target_type(target_list)

# Step 3: Prompt analysis (Claude internal analysis)
ELSE IF prompt_text:
    analysis_result = analyze_prompt(prompt_text)  # Extract targets, types, purpose
    target_list = analysis_result.targets
    target_type = analysis_result.primary_type OR detect_target_type(target_list)
    target_source = "prompt_analysis"

# Step 4: Session synthesis
ELSE IF --session AND exists(role analysis documents):
    target_list = extract_targets_from_synthesis(); target_type = "page"; target_source = "synthesis"

# Step 5: Fallback
IF NOT target_list: target_list = ["home"]; target_type = "page"; target_source = "default"

# Validate and clean
validated_targets = [normalize(t) for t in target_list if is_valid(t)]
IF NOT validated_targets: validated_targets = ["home"]; target_type = "page"
IF --target-type != "auto": target_type = --target-type

# Interactive confirmation
DISPLAY_CONFIRMATION(target_type, target_source, validated_targets, device_type, device_source):
  "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
  "{emoji} {LABEL} CONFIRMATION (Style-Centric)"
  "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
  "Type: {target_type} | Source: {target_source}"
  "Targets ({count}): {', '.join(validated_targets)}"
  "Device: {device_type} | Source: {device_source}"
  "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
  "Performance: {style_variants} agent calls"
  "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
  "Modification Options:"
  "  • 'continue/yes/ok' - Proceed with current configuration"
  "  • 'targets: a,b,c' - Replace target list"
  "  • 'skip: x,y' - Remove specific targets"
  "  • 'add: z' - Add new targets"
  "  • 'type: page|component' - Change target type"
  "  • 'device: desktop|mobile|tablet|responsive' - Change device type"
  "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"

user_input = WAIT_FOR_USER_INPUT()

# Process user modifications
MATCH user_input:
  "continue|yes|ok" → proceed
  "targets: ..." → validated_targets = parse_new_list()
  "skip: ..." → validated_targets = remove_items()
  "add: ..." → validated_targets = add_items()
  "type: ..." → target_type = extract_type()
  "device: ..." → device_type = extract_device()
  default → proceed with current list

STORE: inferred_target_list, target_type, target_inference_source

# ⚠️ CRITICAL: User confirmation complete, IMMEDIATELY initialize TodoWrite and execute Phase 1
# This is the only user interaction point in the workflow
# After this point, all subsequent phases execute automatically without user intervention

Helper Function: detect_target_type()

detect_target_type(target_list):
    page_keywords = ["home", "dashboard", "settings", "profile", "login", "signup", "auth", ...]
    component_keywords = ["navbar", "header", "footer", "hero", "card", "button", "form", ...]

    page_matches = count_matches(target_list, page_keywords + ["page", "screen", "view"])
    component_matches = count_matches(target_list, component_keywords + ["component", "widget"])

    RETURN "component" IF component_matches > page_matches ELSE "page"

Phase 0d: Code Import & Completeness Assessment (Conditional)

IF design_source IN ["code_only", "hybrid"]:
    REPORT: "🔍 Phase 0d: Code Import ({design_source})"
    command = "/workflow:ui-design:import-from-code --design-id \"{design_id}\" --source \"{code_base_path}\""

    TRY:
        # SlashCommand invocation ATTACHES import-from-code's tasks to current workflow
        # Orchestrator will EXECUTE these attached tasks itself:
        #   - Phase 0: Discover and categorize code files
        #   - Phase 1.1-1.3: Style/Animation/Layout Agent extraction
        SlashCommand(command)
    CATCH error:
        WARN: "⚠️ Code import failed: {error}"
        WARN: "Cleaning up incomplete import directories"
        Bash(rm -rf "{base_path}/style-extraction" "{base_path}/animation-extraction" "{base_path}/layout-extraction" 2>/dev/null)

        IF design_source == "code_only":
            REPORT: "Cannot proceed with code-only mode after import failure"
            EXIT 1
        ELSE:  # hybrid mode
            WARN: "Continuing with visual-only mode"
            design_source = "visual_only"

    # Check file existence and assess completeness
    style_exists = exists("{base_path}/style-extraction/style-1/design-tokens.json")
    animation_exists = exists("{base_path}/animation-extraction/animation-tokens.json")
    layout_count = bash(ls {base_path}/layout-extraction/layout-*.json 2>/dev/null | wc -l)
    layout_exists = (layout_count > 0)

    style_complete = false
    animation_complete = false
    layout_complete = false
    missing_categories = []

    # Style completeness check
    IF style_exists:
        tokens = Read("{base_path}/style-extraction/style-1/design-tokens.json")
        style_complete = (
            tokens.colors?.brand && tokens.colors?.surface &&
            tokens.typography?.font_family && tokens.spacing &&
            Object.keys(tokens.colors.brand || {}).length >= 3 &&
            Object.keys(tokens.spacing || {}).length >= 8
        )
        IF NOT style_complete AND tokens._metadata?.completeness?.missing_categories:
            missing_categories.extend(tokens._metadata.completeness.missing_categories)
    ELSE:
        missing_categories.push("style tokens")

    # Animation completeness check
    IF animation_exists:
        anim = Read("{base_path}/animation-extraction/animation-tokens.json")
        animation_complete = (
            anim.duration && anim.easing &&
            Object.keys(anim.duration || {}).length >= 3 &&
            Object.keys(anim.easing || {}).length >= 3
        )
        IF NOT animation_complete AND anim._metadata?.completeness?.missing_items:
            missing_categories.extend(anim._metadata.completeness.missing_items)
    ELSE:
        missing_categories.push("animation tokens")

    # Layout completeness check
    IF layout_exists:
        # Read first layout file to verify structure
        first_layout = bash(ls {base_path}/layout-extraction/layout-*.json 2>/dev/null | head -1)
        layout_data = Read(first_layout)
        layout_complete = (
            layout_count >= 1 &&
            layout_data.template?.dom_structure &&
            layout_data.template?.css_layout_rules
        )
        IF NOT layout_complete:
            missing_categories.push("complete layout structure")
    ELSE:
        missing_categories.push("layout templates")

    needs_visual_supplement = false

    IF design_source == "code_only" AND NOT (style_complete AND layout_complete):
        REPORT: "⚠️  Missing: {', '.join(missing_categories)}"
        REPORT: "Options: 'continue' | 'supplement: <images>' | 'cancel'"
        user_response = WAIT_FOR_USER_INPUT()
        MATCH user_response:
            "continue" → needs_visual_supplement = false
            "supplement: ..." → needs_visual_supplement = true; --images = extract_path(user_response)
            "cancel" → EXIT 0
            default → needs_visual_supplement = false
    ELSE IF design_source == "hybrid":
        needs_visual_supplement = true

    # Animation reuse confirmation (code import with complete animations)
    IF design_source == "code_only" AND animation_complete:
        REPORT: "✅ 检测到完整的动画系统（来自代码导入）"
        REPORT: "   Duration scales: {duration_count} | Easing functions: {easing_count}"
        REPORT: ""
        REPORT: "Options:"
        REPORT: "  • 'reuse' (默认) - 复用已有动画系统"
        REPORT: "  • 'regenerate' - 重新生成动画系统（交互式）"
        REPORT: "  • 'cancel' - 取消工作流"
        user_response = WAIT_FOR_USER_INPUT()
        MATCH user_response:
            "reuse" → skip_animation_extraction = true
            "regenerate" → skip_animation_extraction = false
            "cancel" → EXIT 0
            default → skip_animation_extraction = true  # Default: reuse

    STORE: needs_visual_supplement, style_complete, animation_complete, layout_complete, skip_animation_extraction

Phase 1: Style Extraction

IF design_source == "visual_only" OR needs_visual_supplement:
    REPORT: "🎨 Phase 1: Style Extraction (variants: {style_variants})"
    command = "/workflow:ui-design:style-extract --design-id \"{design_id}\" " +
              (images_input ? "--images \"{images_input}\" " : "") +
              (prompt_text ? "--prompt \"{prompt_text}\" " : "") +
              "--variants {style_variants} --interactive"

    # SlashCommand invocation ATTACHES style-extract's tasks to current workflow
    # Orchestrator will EXECUTE these attached tasks itself
    SlashCommand(command)

    # After executing all attached tasks, collapse them into phase summary
ELSE:
    REPORT: "✅ Phase 1: Style (Using Code Import)"

Phase 2.3: Animation Extraction

# Determine if animation extraction is needed
should_extract_animation = false

IF (design_source == "visual_only" OR needs_visual_supplement):
    # Pure visual input or hybrid mode requiring visual supplement
    should_extract_animation = true
ELSE IF NOT animation_complete:
    # Code import but animations are incomplete
    should_extract_animation = true
ELSE IF design_source == "code_only" AND animation_complete AND NOT skip_animation_extraction:
    # Code import with complete animations, but user chose to regenerate
    should_extract_animation = true

IF should_extract_animation:
    REPORT: "🚀 Phase 2.3: Animation Extraction"

    # Build command with available inputs
    command_parts = [f"/workflow:ui-design:animation-extract --design-id \"{design_id}\""]

    IF images_input:
        command_parts.append(f"--images \"{images_input}\"")

    IF prompt_text:
        command_parts.append(f"--prompt \"{prompt_text}\"")

    command_parts.append("--interactive")

    command = " ".join(command_parts)

    # SlashCommand invocation ATTACHES animation-extract's tasks to current workflow
    # Orchestrator will EXECUTE these attached tasks itself
    SlashCommand(command)

    # After executing all attached tasks, collapse them into phase summary
ELSE:
    REPORT: "✅ Phase 2.3: Animation (Using Code Import)"

# Output: animation-tokens.json + animation-guide.md
# When phase finishes, IMMEDIATELY execute Phase 2.5 (auto-continue)

Phase 2.5: Layout Extraction

targets_string = ",".join(inferred_target_list)

IF (design_source == "visual_only" OR needs_visual_supplement) OR (NOT layout_complete):
    REPORT: "🚀 Phase 2.5: Layout Extraction ({targets_string}, variants: {layout_variants}, device: {device_type})"
    command = "/workflow:ui-design:layout-extract --design-id \"{design_id}\" " +
              (images_input ? "--images \"{images_input}\" " : "") +
              (prompt_text ? "--prompt \"{prompt_text}\" " : "") +
              "--targets \"{targets_string}\" --variants {layout_variants} --device-type \"{device_type}\" --interactive"

    # SlashCommand invocation ATTACHES layout-extract's tasks to current workflow
    # Orchestrator will EXECUTE these attached tasks itself
    SlashCommand(command)

    # After executing all attached tasks, collapse them into phase summary
ELSE:
    REPORT: "✅ Phase 2.5: Layout (Using Code Import)"

Phase 3: UI Assembly

command = "/workflow:ui-design:generate --design-id \"{design_id}\"" + (--session ? " --session {session_id}" : "")

total = style_variants × layout_variants × len(inferred_target_list)

REPORT: "🚀 Phase 3: UI Assembly | Matrix: {s}×{l}×{n} = {total} prototypes"
REPORT: "   → Pure assembly: Combining layout templates + design tokens"
REPORT: "   → Device: {device_type} (from layout templates)"
REPORT: "   → Assembly tasks: {total} combinations"

# SlashCommand invocation ATTACHES generate's tasks to current workflow
# Orchestrator will EXECUTE these attached tasks itself
SlashCommand(command)

# After executing all attached tasks, collapse them into phase summary
# When phase finishes, IMMEDIATELY execute Phase 4 (auto-continue)
# Output:
# - {target}-style-{s}-layout-{l}.html (assembled prototypes)
# - {target}-style-{s}-layout-{l}.css
# - compare.html (interactive matrix view)
# - PREVIEW.md (usage instructions)

Phase 4: Design System Integration

command = "/workflow:ui-design:update" + (--session ? " --session {session_id}" : "")

# SlashCommand invocation ATTACHES update's tasks to current workflow
# Orchestrator will EXECUTE these attached tasks itself
SlashCommand(command)

# After executing all attached tasks, collapse them into phase summary
# When phase finishes:
#   - If --batch-plan flag present: IMMEDIATELY execute Phase 5 (auto-continue)
#   - If no --batch-plan: Workflow complete, display final report

Phase 5: Batch Task Generation (Optional)

IF --batch-plan:
    FOR target IN inferred_target_list:
        task_desc = "Implement {target} {target_type} based on design system"

        # SlashCommand invocation ATTACHES plan's tasks to current workflow
        # Orchestrator will EXECUTE these attached tasks itself
        SlashCommand("/workflow:plan --agent \"{task_desc}\"")

TodoWrite Pattern

// Initialize IMMEDIATELY after Phase 0c user confirmation to track multi-phase execution (5 orchestrator-level tasks)
TodoWrite({todos: [
  {"content": "Execute style extraction", "status": "in_progress", "activeForm": "Executing style extraction"},
  {"content": "Execute animation extraction", "status": "pending", "activeForm": "Executing animation extraction"},
  {"content": "Execute layout extraction", "status": "pending", "activeForm": "Executing layout extraction"},
  {"content": "Execute UI assembly", "status": "pending", "activeForm": "Executing UI assembly"},
  {"content": "Execute design integration", "status": "pending", "activeForm": "Executing design integration"}
]})

// ⚠️ CRITICAL: Dynamic TodoWrite task attachment strategy:
//
// **Key Concept**: SlashCommand invocation ATTACHES tasks to current workflow.
// Orchestrator EXECUTES these attached tasks itself, not waiting for external completion.
//
// Phase 1-5 SlashCommand Invocation Pattern:
// 1. SlashCommand invocation ATTACHES sub-command tasks to TodoWrite
// 2. TodoWrite expands to include attached tasks
// 3. Orchestrator EXECUTES attached tasks sequentially
// 4. After all attached tasks complete, COLLAPSE them into phase summary
// 5. Update next phase to in_progress
// 6. IMMEDIATELY execute next phase SlashCommand (auto-continue)
//
// Benefits:
// ✓ Real-time visibility into sub-command task progress
// ✓ Clean orchestrator-level summary after each phase
// ✓ Clear mental model: SlashCommand = attach tasks, not delegate work
// ✓ Dynamic attachment/collapse maintains clarity

Completion Output

✅ UI Design Explore-Auto Workflow Complete!

Architecture: Style-Centric Batch Generation
Run ID: {run_id} | Session: {session_id or "standalone"}
Type: {icon} {target_type} | Device: {device_type} | Matrix: {s}×{l}×{n} = {total} prototypes

Phase 1: {s} complete design systems (style-extract with multi-select)
Phase 2: {n×l} layout templates (layout-extract with multi-select)
  - Device: {device_type} layouts
  - {n} targets × {l} layout variants = {n×l} structural templates
  - User-selected concepts generated in parallel
Phase 3: UI Assembly (generate)
  - Pure assembly: layout templates + design tokens
  - {s}×{l}×{n} = {total} final prototypes
Phase 4: Brainstorming artifacts updated
[Phase 5: {n} implementation tasks created]  # if --batch-plan

Assembly Process:
✅ Separation of Concerns: Layout (structure) + Style (tokens) kept separate
✅ Layout Extraction: {n×l} reusable structural templates
✅ Multi-Selection Workflow: User selects multiple variants from generated options
✅ Pure Assembly: No design decisions in generate phase
✅ Device-Optimized: Layouts designed for {device_type}

Design Quality:
✅ Token-Driven Styling: 100% var() usage
✅ Structural Variety: {l} distinct layouts per target (user-selected)
✅ Style Variety: {s} independent design systems (user-selected)
✅ Device-Optimized: Layouts designed for {device_type}

📂 {base_path}/
  ├── .intermediates/          (Intermediate analysis files)
  │   ├── style-analysis/      (analysis-options.json with embedded user_selection, computed-styles.json if URL mode)
  │   ├── animation-analysis/  (analysis-options.json with embedded user_selection, animations-*.json if URL mode)
  │   └── layout-analysis/     (analysis-options.json with embedded user_selection, dom-structure-*.json if URL mode)
  ├── style-extraction/        ({s} complete design systems)
  ├── animation-extraction/    (animation-tokens.json, animation-guide.md)
  ├── layout-extraction/       ({n×l} layout template files: layout-{target}-{variant}.json)
  ├── prototypes/              ({total} assembled prototypes)
  └── .run-metadata.json       (includes device type)

🌐 Preview: {base_path}/prototypes/compare.html
  - Interactive {s}×{l} matrix view
  - Side-by-side comparison
  - Target-specific layouts with style-aware structure
  - Toggle between {n} targets

{icon} Targets: {', '.join(targets)} (type: {target_type})
  - Each target has {l} custom-designed layouts
  - Each style × target × layout has unique HTML structure (not just CSS!)
  - Layout plans stored as structured JSON
  - Optimized for {device_type} viewing

Next: [/workflow:execute] OR [Open compare.html → Select → /workflow:plan]