What they did
Subhendra Basu, who lives in Noida, wanted more than a single AQI number during smog season. In one night of pairing with Claude Code he went from asking which satellite data is free to telling it to build the NCR tracker. Claude Code wrote a single web page that pulls CAMS pollution data for 12 locations through the keyless Open-Meteo API in one request, computes India's CPCB AQI from 24-hour means, shows a 4-day forecast and Delhi's wind direction, and a standard-library Python script that pulls NASA FIRMS fire detections over Punjab and Haryana. It tested both in a headless browser against mocked API responses. Afterwards he had it write a CLAUDE.md and turn the workflow into a Claude skill.
How it went
The tracker is open source and live on GitHub Pages, with a GitHub Action refreshing fire data every three hours and no server or bill. Two map problems came up along the way: OpenStreetMap tiles were blocked when the page was opened as a local file, and CARTO now watermarks keyless tiles; Claude explained both and the page now serves locally and takes an optional CARTO key. With the skill he can ask for tomorrow's AQI in Noida and Claude fetches and computes it.
Worth knowing
CAMS is a model on a 10 to 40 km grid, not a monitoring station, so the numbers are regional estimates and the dashboard says so. Most global APIs report the US AQI, which uses different breakpoints from India's scale. The FIRMS fire layer needs a free MAP_KEY.