openxiangshan/xiangshanlab/english-word-card

English Word Card

Generate a kid-friendly illustrated vocabulary card with text explanation.

Workflow (optimized ~65s)

1. Reuse existing Gemini tab or navigate to https://gemini.google.com/app:

• If a Gemini tab is already open, use browser action=navigate on that tab
• Only browser action=open if no tab exists
• Skip unnecessary snapshot/click - type directly using the input ref (typically e14 on fresh page, e18/e19 on existing conversation)

1. Type prompt and submit (one step, no separate click):

`` browser action=act, kind=type, ref=<input_ref>, text="Generate an image for a child to learn the English word '[WORD]'. Draw a cute cartoon illustration that shows the meaning of '[WORD]'. Use bright colors, kid-friendly style. Write the word '[WORD]' in big friendly colorful letters at the top.", submit=true ``

1. Wait ~15s for image generation (not 18s - 15s is enough)

1. Snapshot to find "Copy image" button ref

1. Click "Copy image" button

1. Extract PNG from macOS clipboard (sleep 1s then extract):

`` sleep 1 && python3 -c " import subprocess p = subprocess.run(['osascript', '-e', 'the clipboard as «class PNGf»'], capture_output=True) text = p.stdout.decode('utf-8', errors='replace') hex_str = text.split('«data PNGf')[1].split('»')[0] binary = bytes.fromhex(hex_str) with open('/Users/christopherxu/.openclaw/workspace/word-card.png', 'wb') as f: f.write(binary) print(f'Saved {len(binary)} bytes') " `` This gives the original 1024x1024 PNG (~1.3-1.6MB)

1. Send image + text via message action=send with media + caption

Text Format

🐴 **WORD** /pronunciation/
 
**Meaning:** English definition in simple terms
 
**Example sentences:**
• Sentence 1
• Sentence 2
 
💡 **Extra:** Any extended/colloquial usage
 
---
 
**中文解释：** Chinese explanation
**例句翻译：** Chinese translation of examples

Optimization Tips

• Reuse tabs - don't open new tabs each time, navigate existing ones
• Skip redundant snapshots - after page load, type directly if you know the ref
• Combine exec steps - sleep + python extraction in one exec call
• 15s generation wait is sufficient (tested), not 18-20s
• Each tool call has ~1-2s overhead, minimize total number of calls

Fallback

If "Copy image" + clipboard fails, fall back to element screenshot:

• browser action=screenshot with selector="img[alt*='AI generated']"
• Then crop with sips to remove UI chrome

Notes

• Keep language simple - target audience is a young child (Emily 11yr, Catherine 8yr)
• Illustrations should be cute, colorful, and clearly depict the word's meaning
• If Gemini web UI is unavailable, fall back to nano-banana-pro skill (API)
• Always send image and text together in one message
• Download button does NOT work (Blob URL issue in managed Chrome profile)