every AI coding team eventually asks the same stupid, un-ignorable question: who coded more? they mean the work that moved through the repository, across every agent and terminal. WhoCodedMore reads the local logs, counts lines alongside tokens, converts the burn into water, and gives the argument a public leaderboard
the joke is the competition. the product is an honest meter. your Claude Code, Codex, Cursor, Gemini CLI, Copilot, OpenCode, Windsurf, Cline, and the rest of the stack each hide their own totals. WhoCodedMore is the screen that says: here is everything, across tools, in one number.
how it works
local scan
npx whocodedmore@latest reads usage logs already on the machine. Claude Code, Codex, Gemini CLI, OpenClaw, Cursor, Copilot, OpenCode, Windsurf, Cline, Aider, Amp, Continue, Goose, Kiro, Roo Code, Zed, Warp, Droid, Pi, Kimi CLI, Qwen CLI, Antigravity, Hermes, plus anything ccusage already tracks. zero flags naming tools; the scan discovers whichever logs exist
lines and tokens
token boards are easy and a little fake. a cached prompt looks huge. WhoCodedMore counts real lines too, then tracks tokens, estimated spend, and water. the water number is the part that makes a leaderboard feel like a bill: glasses burned, drinking-days for one human, a live total that climbs when someone syncs more.
sync
the machine sends aggregate totals. code, paths, prompts, and file names stay local. syncing adds those totals to the public board and updates the worldwide counters
the board
public by default once you show up. a friends board if you want the argument in private. lines show what reached code, tokens show the model work around it, and water gives the compute bill a unit you can picture
run it once. it counts, syncs, and you climb. re-run to refresh. or check who else is burning time on this site.