honey-for-devs
Honey (I Shrunk the AI) by GreenPT: a cross-tool coding skill that cuts AI coding-agent token usage and LLM API costs — write less code, less prose, and denser
What is honey-for-devs?
Honey (I Shrunk the AI) by GreenPT: a cross-tool coding skill that cuts AI coding-agent token usage and LLM API costs — write less code, less prose, and denser agent-to-agent handoffs (−53%, lossless in benchmarks) with no loss of quality. Works with Claude Code, Cursor, GitHub Copilot, Codex, Gemini CLI, Windsurf, Cline & Kiro.
Overview of honey-for-devs
What it is
Honey (I Shrunk the AI) by GreenPT is a cross-tool coding skill that reduces the output tokens AI coding agents generate, lowering LLM API costs. It works with Claude (claude.ai and API), Claude Code, Cursor, GitHub Copilot, Codex, Gemini CLI, Windsurf, Cline, OpenClaw, oh-my-pi, Kiro, Kilo Code, Hermes Agent, Kimi Code, Devin CLI, Grok Build, Qoder, Antigravity, and CodeBuddy. The skill applies three levers: writing less code (YAGNI ladder), less prose (removing unnecessary narration), and denser agent-to-agent handoffs (using compact JSON or ESON). The summary reports a −53% reduction in handoff tokens (lossless in benchmarks). Tests on Claude Opus 5.5 (23 tasks × 3 runs) show Caveman v4.0.1 achieving −29% output, −27% lines of code, −26% cost, with a judge result of tie (3/8/10) and 100% tests. Ponytail v5.3.2 achieved −28% output, −52% lines of code, −26% cost, with a judge result of worse (1/16/4).
Key features
- Three levers: less code (YAGNI ladder: does it need to exist? → stdlib → language native → existing dependency → one line → minimum block), less prose (drop wind-up, hedging, narration), denser agent-to-agent handoffs (compact/columnar JSON or ESON; −49% on Opus 4.8, −15% on Opus 5.5; two tasks, no p-value).
- Auto-intensity: lite / full / ultra chosen reflexively from the request, without spending reasoning tokens on deliberation.
- Safety carve-outs: input validation, error handling, auth, secrets, migrations, deletes, and explicitly requested items are never compressed.
- 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.
- Works with a wide range of AI coding tools (see list above).
- Results from Claude Opus 5.5 benchmarks: Caveman v4.0.1 reduces output by 29%, lines by 27%, cost by 26%; Ponytail v5.3.2 reduces output by 28%, lines by 52%, cost by 26%.
Use cases
Reducing token usage and API costs when using AI coding agents. Writing less code and less prose without losing correctness. Efficient agent-to-agent handoffs in multi-agent workflows.
Getting started
See the repository README for setup instructions.
License and pricing
MIT license. Open source.
Based on the project's public README and repository metadata.