sup3x/claude-code-eco

README preview

# claude-eco — eco mode for Claude Code

**Claude Code spends your money on work you never asked for: re-reading files it just edited, pasting back diffs it already applied, writing three-paragraph reports for two-line fixes. I measured every token of it, deleted it, and published all 82 raw runs — including the ones where I lose.**

*Measured hardest on Claude Fable 5 — the hungriest configuration that ever existed — and tuned for every effort-based Claude. `/eco`: −31% to −73% output tokens depending on task and effort level (−63% mean on the flagship n=5 study), with critical findings intact and all produced fixes executed and verified. `/eco-max`: up to −75% by dialing reasoning effort down — opt-in, labeled.*

[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE) [![Claude Code](https://img.shields.io/badge/Claude%20Code-skill%20%2B%20plugin-blueviolet)](https://code.claude.com/docs/en/skills) [![Benchmarks](https://img.shields.io/badge/benchmarks-82%20raw%20runs-success)](benchmarks/results.md)

[Results](#measured-results) · [Install](#install) · [Usage](#usage) · [Benchmarks](benchmarks/results.md) · [FAQ](#faq)

![Same answer at half the tokens — /eco −31% to −73% depending on task and effort, critical findings intact; /eco-max up to −75%](assets/social-preview.png)

The waste isn't occasional — it repeats on **every turn of every session**, which is why usage limits evaporate and API bills surprise you. **claude-eco** deletes it with two skills:

- **`/eco`** — frugality rules with a non-negotiable quality floor. Full reasoning depth preserved.
- **`/eco-max`** — the same rules plus a **low reasoning-effort override** for routine chores. Effort is the single biggest token lever on modern Claude models (Fable 5, Sonnet 5, Opus 4.8), and of every terse-mode tool we surveyed, this is the only one we found that touches it.

**What that means in practice:** verbose answers shrink 2–4×, replies come back up to 3× faster (30s → 10–15s, measured), and the same weekly limit simply lasts longer — same model, same fixes, minus the padding.

## See the difference

View full repository on GitHub →