The Problem: When Color Coding Was the Only Priority System
On a Windows device, Google Sheets. Dozens of request items accumulated daily. Yellow meant 「urgent reply needed」, blue meant 「on hold」, green meant 「completed」. These were color rules created by hand.
The trouble was simple: some people remembered the rules, some forgot them. Some rows were colored correctly, others weren't. And most importantly, reading those colors consumed dozens of minutes every single day.
「The color coding is so clear. Why can't the bot read it?」
This small question was the beginning.
The Discovery: Claude Code Can Extract Color Data
Combine Google Sheets API with Claude Code, and you can read cell background colors as RGB values. In other words, converting visual signals into data became possible.
The code was simple.
// Read Google Sheets color values (RGB extraction)
const range = sheet.getRange('A1:D50');
const backgrounds = range.getBackgrounds();
// Map color RGB to priority
const priorityMap = {
'#ffff00': 'URGENT', // Yellow
'#0000ff': 'WAITING', // Blue
'#00ff00': 'DONE' // Green
};
When the bot read this code, the sorting task that once depended on human eyes was completed in 0.3 seconds.
Execution: From Auto-Sorting to Auto-Alerts
The first week was experimental. The bot read colors and auto-sorted them by priority in a separate sheet tab. Two days later, we improved it. When the bot detected a 「yellow」 row, it automatically sent a Slack message.
Three weeks later, it became more sophisticated.
• Bot detects when colors change
• Change history logs automatically
• If a 「yellow」 item sits untouched for 12+ hours, a reminder fires
• Every Monday at 9 AM, a summary report of all 「yellow」 items for the week
The most surprising part: not a single line of SQL was needed to build all of this. Claude Code implemented the entire logic in Google Apps Script.
Unexpected Effect: The Team Started Coloring More Carefully
Something interesting happened. When the team learned that the bot reads colors, they began following the color rules more rigorously. Because incorrect coloring meant the bot's alerts would also be wrong.
Paradoxically, automation improved human accuracy.
Next Experiment: Multi-Color Pattern Recognition
Right now we only recognize single colors. The next phase is to assign meaning to combined color patterns like 「yellow with gray stripes」. That way, color coding alone could represent much richer data.
Small discovery, but it reminds us how thin the line between 「data」 and 「visual」 really is.