Honey Agent Skill by GreenPT MCP Server

Open-source GreenPT skill that cuts coding-agent output 29% across mixed tasks and up to 70% in focused review workflows.

Write less code and say less about it.Honey (I Shrunk the AI) byGreenPTis a cross-tool coding skill that cuts AI coding-agent token usage and LLM API costs — making agents emit less codeandless prose without losing correctness. It works withClaude (claude.ai and the API), Claude Code, Cursor, GitHub Copilot, Codex, Gemini CLI, Windsurf, Cline, OpenClaw, oh-my-pi, Kiro, Kilo Code, and Hermes Agent. Three independent levers, applied reflexively:
- 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. The cheapest line is the one you never write.
- Less prose— drop the wind-up, the hedging, the narration of code that already speaks for itself. Answer first.
- Denser agent-to-agent handoffs— when the reader is another agent, not a human, hand it the most token-efficient format it parses losslessly (compact / columnar JSON, or ESON). Cuts handoff size ~in half at zero loss of recovery. Fires only here — never as a user-facing answer.

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

- Auto-intensity—lite/full/ultrachosen reflexively from the request, with no deliberation tax (it never spends reasoning tokens decidinghowto comply — that would defeat the purpose on reasoning models).
- Safety carve-outs— input validation, error handling, auth, secrets, migrations, deletes, and anything you explicitly asked for arenevercompressed. Lazy ≠ broken.
- A skill family, not one prompt— an always-on core plus on-demand satellites (review, eco, gain, compress) and ahiveof read-only subagents that return compressed handoffs. See Skills & subagents.

Volume is cost. In agentic coding sessions, the volume of generated code and prose is what runs up the bill — and most of it is waste.

This repo ships areproducible benchmark(bench/) so you don't have to take the numbers on faith: 23 tasks across three kinds of work — baseline vs Cavemanvs Ponytailvs Honey — same model, same prompts, only the skill changes. Correctness is objective (unit tests, structural / accessibility checks, and lossless round-trip recovery for agent handoffs); quality is scored by a4-model cross-family judge panel(median of Opus 4.8 + Sonnet 4.6

- Haiku 4.5 + GPT-5.5) under aneutral rubricthat says nothing about length, so a terse skill gets no thumb on the scale. The figures below are the committed results (Claude Opus 4.8, 3 runs each) — runcd bench && npm run benchto reproduce.

Every number is apaired per-task deltavs baseline — runs collapse by median, tasks pair up, and the figure is the median of those paired deltas with a two-sided Wilcoxonp. Not a ratio of arm totals: that is dominated by whichever task happens to be longest, and it is how token-saving tools end up publishing numbers nobody can reproduce. Endpoints and the run ladder are pre-registered in bench/METHODOLOGY.md.

OnClaude Opus 5(23 tasks × 3 runs, 207 cells, zero refusals or truncation —full-opus5-lean):

Honey is the only arm with no failing cell — the no-skill baseline fails four. And the cut islarger on the newer model, not smaller: −71% LOC on Opus 5 against −39% on Opus 4.8. That runs against the 2026 prompting guidance that newer models need less instruction, which we tested directly and rejected — see METHODOLOGY.md.

A single blended number hides the story, because the levers fire differently per task type. Honey on Opus 4.8, where the full competitor set was run —Δ LOCmeasures Lever 1 directly,Δ outputmeasures the tokens (codeandthe prose around it):

Against the competitors on the whole suite (judge win/loss/tie by exact sign test):

- Code— the deepest cut (−39%) at 100% unit-test pass. Ponytail's mandatory self-checkinflatestrivial code (+60% on Opus, +92% on GPT-5.5).
- User-facing— the carve-out keeps Honey from compressing polish: the output delta here is astatistical tie, and Honey holds the only 100% accessibility pass while Ponytail drops to 81% on the structural/a11y checklist.
- Agent-to-agent— under adversarial relay queries (ordinal, nested, absence, cross-field count) Honey is theonly variant that stays 100% losslesswhile roughly halving handoff size; Caveman and Ponytail compress harderandlose recovery (67% / 50%). Its biggest, cleanest win — on 2 tasks, so no p-value.
- Quality is a tie overall(p=0.648) — fewer tokens at no measurable quality cost, not higher quality. But the whole-suite tie is two opposing effects cancelling: on Opus, Honeywins user-facing 6/0/1 (p=0.031)andloses the code judge 2/11/1 (p=0.022)— on tasks where every variant passes 100% of the unit tests, so that is a stylistic penalty for terseness, not a correctness one. Neither effect replicates on GPT-5.5 (p=0.375 / p=1.000), so treat the code-judge dip as suggestive, not established. Caveman's judgemeanalso ties baseline exactly — but paired, it loses 16 of 23 tasks (p=0.004). Means hide that; sign tests don't.
- The dollar saving is unproven at this sample size.−21% on Opus is p=0.104 — not significant on 23 tasks. Output volume is down; the bill is not yet a claim.