tokenfficiency

Compare your AI token spend to the code you actually ship.

npx tokenfficiency

Auto-detects your coding agents, pulls your GitHub shipping stats, calculates efficiency metrics, and publishes a shareable report.

What it does

1. GitHub stats via gh CLI — merged PRs, direct commits, lines added/deleted per commit with file-level noise filtering (lock files, generated code, etc.)
2. Token usage via the ccusage family — Claude Code, OpenCode, Codex, Pi, Amp
3. Efficiency grade (S through F) based on cost per line of shipped code
4. Auto-publishes to tokenfficiency.vercel.app for a shareable link

Requirements

At least one of:

• GitHub CLI (gh auth login) — for shipping stats
• A coding agent (Claude Code, OpenCode, Codex, Pi, Amp) — for token usage

Options

--since <YYYYMMDD>   Filter from date (default: 2026-01-01)
--until <YYYYMMDD>   Filter until date
--repo <owner/repo>  Filter to specific repo
--no-post            Don't upload (local only)
--json               Output as JSON (implies --no-post)
--help               Show help
--version            Show version

Examples

npx tokenfficiency                          # full report, auto-publish
npx tokenfficiency --no-post                # local only
npx tokenfficiency --since 20260301         # March 2026 onwards
npx tokenfficiency --repo myorg/myapp       # single repo
npx tokenfficiency --json | jq .efficiency  # pipe to jq

Grading

Based on cost per line of shipped code (after noise filtering):

| Grade | Cost/line | Description | |-------|-----------|-------------| | S | < $0.001 | Hyper-efficient | | A | < $0.005 | Excellent | | B | < $0.01 | Good | | C | < $0.05 | Average | | D | < $0.10 | Costly | | F | > $0.10 | Expensive |

How it works

• Shells out to gh for GitHub data (reuses your existing auth)
• Shells out to ccusage, @ccusage/opencode, @ccusage/codex, @ccusage/pi, @ccusage/amp for token data
• Filters noise files (lock files, .min.js, .d.ts, vendored deps, etc.) at the file level using GitHub's per-file API
• Excludes merge commits, reverts, and bot activity
• Maps token sessions to GitHub repos via local git remotes
• Publishes to tokenfficiency.vercel.app with GitHub token verification (you can only publish your own report)