Honey

Honey (I Shrunk the AI) is GreenPT's cross-tool coding skill that cuts AI coding-agent token usage and LLM API costs — making agents emit less code and less prose without losing correctness.

Repository: github.com/Green-PT/honey-for-devs

Works with Claude Code, Cursor, GitHub Copilot, Codex, Gemini CLI, Windsurf, Cline, OpenClaw, and Kiro.

Three levers

1. Less code — YAGNI first. Walk a ladder (does it need to exist? → stdlib → language native → existing dependency → one line → minimum block) and stop at the first rung that works.
2. Less prose — drop the wind-up, the hedging, and narration of code that already speaks for itself. Answer first.
3. Denser agent-to-agent handoffs — when the reader is another agent, hand it the most token-efficient format it parses losslessly (compact / columnar JSON, or ESO). Fires only here — never as a user-facing answer.

Beyond Ponytail + Caveman

Honey combines what Ponytail (minimal code) and Caveman (terse prose) do separately, then adds:

• Auto-intensity — lite / full / ultra chosen reflexively from the request, with no reasoning-token tax for deciding how to comply.
• Safety carve-outs — input validation, error handling, auth, secrets, migrations, deletes, and anything you explicitly asked for are never compressed. Lazy ≠ broken.
• A skill family — an always-on core plus on-demand satellites (review, eco, gain, compress) and a hive of read-only subagents that return compressed handoffs.

Benchmarks

The repo ships a reproducible benchmark: 23 tasks across three kinds of work, scored by a 4-model cross-family judge panel under a neutral rubric. Quality is % of baseline; tokens are generated output vs baseline.

Run cd bench && npm run bench in the repo to reproduce.