Icon Generate

---
allowed-tools: Bash, Read, Write, Edit, Glob, Grep, Task, WebSearch, WebFetch, mcp__fetch__fetch
description: App icon design pipeline. Runs research → design → validation → user feedback loop.
argument-hint: [app name] [--url URL] [--description "app description"]
---

# /_icon-generate - App Icon Design Pipeline

Designs app icons. Iterates with user feedback until satisfaction.
Packaging is run separately with `/_icon-package`.

## Execution Flow

### [1] Environment Setup

1. Check Python dependencies:
```bash
cd D:/AI/AppIcon2
pip install -r requirements.txt --quiet

2. Create output folder (only candidates/, no source/):

python -c "
from src.config import create_output_dir
output_dir = create_output_dir('$ARGUMENTS_app_name')
print(f'OUTPUT_DIR={output_dir}')
"

Use the generated OUTPUT_DIR in all subsequent steps.

[2] Research (icon-researcher agent)

Invoke the icon-researcher agent.

Information to pass:

• App name: extracted from $ARGUMENTS
• App description: --description value, or ask the user
• URL: --url value (if provided)
• Output path: {OUTPUT_DIR}

The agent writes {OUTPUT_DIR}/research_brief.md.

Checkpoint: Show the design brief to the user and have them select 3 concept directions.

• Default: researcher's top 3 auto-recommended
• User can choose a different combination (e.g., "concepts 1, 3, 5" or "color from 2 + shape from 4")
• 12 variations are distributed across 3 concepts: A(0104), B(0508), C(09~12)

[3] Design (icon-designer agent)

Invoke the icon-designer agent.

Information to pass:

• research_brief.md path
• User-selected 3 concept directions (A, B, C)
• Output path: {OUTPUT_DIR}

The agent generates:

• {OUTPUT_DIR}/candidates/01~12.svg (3 concepts × 4 styles)
• {OUTPUT_DIR}/preview.html (browser preview)

No PNG conversion — SVGs are verified directly in the browser.

[4] Light Validation (icon-reviewer agent)

Invoke the icon-reviewer agent with review_mode=light.

Information to pass:

• review_mode: "light"
• {target_dir}: {OUTPUT_DIR}/candidates/

Light review: SVG validity check only (no PNG validation).

Failed candidates are auto-regenerated (up to 3 times). Only passing candidates are kept. After regeneration, preview.html is also refreshed.

[5] Present Candidates to User

Provide the preview.html path to the user:

Check the candidates in your browser:
{OUTPUT_DIR}/preview.html

Use AskUserQuestion to request user selection:

• Which icon number would you like to select?
• Any modifications needed? (color change, shape adjustment, style change, etc.)

[6] Feedback Loop (repeat until user is satisfied)

Branch based on the user's response:

A. "Confirm as-is" → Create output/ + Full validation

1. Create output/ folder and copy SVG:

mkdir -p '{OUTPUT_DIR}/output'
cp '{OUTPUT_DIR}/candidates/{selected_number}.svg' '{OUTPUT_DIR}/output/icon.svg'

2. Convert selected SVG to 1024x1024 PNG:

cd D:/AI/AppIcon2
python -c "
from src.converter import svg_to_png
from src.config import FULL_SIZE
svg_to_png('{OUTPUT_DIR}/output/icon.svg', '{OUTPUT_DIR}/output/icon_1024.png', width=FULL_SIZE, height=FULL_SIZE)
print('1024x1024 conversion complete')
"

3. Invoke the icon-reviewer agent with review_mode=full:

   • review_mode: "full"
   • {target_dir}: {OUTPUT_DIR}/output/
   • {OUTPUT_DIR}/research_brief.md

4. Handle full review results:

   • PASS → Confirmation complete
   • FAIL → Report failed items to user, request choice to fix or accept

5. On confirmation complete, output:

======================================================
  Icon Design Complete
======================================================
  App: {app name}
  Output: {OUTPUT_DIR}/output/

  Confirmed files:
    - icon.svg      (SVG original)
    - icon_1024.png  (1024x1024 PNG)

  For packaging:
    /_icon-package {OUTPUT_DIR}/output/icon_1024.png
======================================================

Pipeline ends.

B. Modification request → Revise design and re-display

1. Re-invoke the icon-designer agent. Information to pass:

   • Selected candidate SVG path
   • User's modification request (color change, shape adjustment, etc.)
   • Output path: {OUTPUT_DIR}/candidates/ (overwrite existing number or new number)

2. Regenerate preview.html:

cd D:/AI/AppIcon2
python -c "
from src.preview import generate_preview
generate_preview('{OUTPUT_DIR}', '{app_name}')
print('preview.html refreshed')
"

3. Re-validate modified results with icon-reviewer at review_mode=light

4. Present preview.html to user again

5. Return to [6] — repeat until user is satisfied

Usage Examples

# Basic usage
/_icon-generate MyApp --description "A to-do management app"

# With URL
/_icon-generate MyApp --url https://myapp.com --description "A to-do management app"

Error Handling

• Dependency installation failure: cairosvg/pycairo requires Cairo runtime. Display guidance message
• SVG conversion failure: SVG code error. Request redesign from the design team
• 3 validation failures: Exclude the candidate and display remaining candidates only