Honey (I Shrunk the AI) plugin
A token-saving skill for coding agents: shorter code, tighter replies and compact handoffs between agents
- 346 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The session-start hook runs a script and reads the repository folder structure
- The rules change agent behavior: it writes less code and explanation, so verify the result with tests
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/honey-i-shrunk-the-aiDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The session-start hook runs a script and reads the repository folder structure
- The rules change agent behavior: it writes less code and explanation, so verify the result with tests
This is third-party code. Review the repository files before installing.
What it does
Honey from GreenPT cuts the output tokens a coding agent spends without losing correctness. The skill applies three levers. First, less code: the agent walks a ladder of questions from «does this need to exist» through the standard library to one line and stops at the first rung that solves the task. Second, less prose: no wind-up, hedging or narration of obvious code, with the answer first. Third, denser handoffs between agents: when the reader is another agent, data goes in a compact form it parses without loss. The skill picks the lite, full or ultra intensity itself from the request, spending no tokens on deliberating the choice. Security and error-handling code, migrations, deletes and anything you asked for explicitly are never compressed. The bundle has an always-on core and on-demand satellites such as review, a savings report and carbon accounting. Separate read-only subagents return compressed handoffs. A session-start hook adds a map of top-level folders and shared helpers of about 50 tokens, so the agent reuses existing code.
Who it is for. For developers who pay for Claude Code, Cursor or Codex tokens and want less extra code and chatter in agent replies.
Good fit when
- The token bill grows and the agent writes long replies narrating its own code
- The agent adds extra abstractions and dependencies where the standard library is enough
- Several agents pass large data to each other and the handoff should be compact
- You want a report on how many tokens and how much carbon the skill saved
- You need the same rules for several clients at once: Claude Code, Cursor, Copilot, Codex
Not a fit when
- You need detailed code explanations and teaching detail: the skill shortens them
- You need to compress tool output in a stream: separate proxies such as snip exist for that
- The task is entirely about security, migrations or deleting data: the skill leaves those parts alone, so there is nothing to gain
Example request
/honey ultra, then add an access token check to the API and return only the changesLimitations
For Claude Code you install the plugin from the marketplace, and for other clients the rules sit in files under .cursor, .clinerules, .kiro and similar folders, with details in the docs/platforms.md document of the repository. After installing, the skill must be switched on once with `/honey`. The mode is saved per project, and `/honey default <mode>` sets the value for every repository without its own.
The repository map is turned off by the `HONEY_MAP=0` variable. The cost saving is for a single call: in a real Cline agent loop on Opus 5.5 output tokens drop by 39%, but the bill stays flat because input dominates a session. The author tested handoff compression on only two tasks.
How to disable. Run `/honey off` to stop the skill or remove the plugin through `/plugin`. The repository map at session start is turned off by the `HONEY_MAP=0` variable.
FAQ
How do I install Honey in Claude Code?
Run `/plugin marketplace add Green-PT/honey-for-devs` and `/plugin install honey@greenpt`, then `/honey` once. A badge in the status line shows the active mode.
How do the lite, full and ultra modes differ?
They are three compression intensity levels. The skill picks a level itself from the wording of the request, and you set one by hand with `/honey lite`, `/honey full` or `/honey ultra`.
Will Honey break error handling or input validation?
No, these parts, like auth, secrets, migrations and deletes, are left uncompressed by the carve-out rules.
How is Honey different from Caveman?
Caveman compresses prose, and Honey adds code compression and denser handoffs between agents. In the author's run on Claude Opus 5.5 Caveman cuts output tokens by 29% and Honey by 40%.
Which clients does Honey work with?
With Claude Code, Cursor, GitHub Copilot, Codex, Gemini CLI, Windsurf, Cline, Kiro, Kilo Code, Hermes Agent and others. For all of them except Claude Code you install a rule file or use a one-line installer.
Discussion
No questions or reviews yet.
Related
PluginsA skill plus hooks that keep a coding agent on task and cut speculative defenses, unread checksums and scope creep
A skill that cuts an agent's prose output, plus a local proxy that compresses what the agent reads
A parallel divergent ideation skill for agents: fans out under cognitive frames, prunes traps and deepens the best branches
Token Savior indexes code by symbol, keeps memory between sessions and rewrites noisy Bash commands so the agent spends fewer tokens
A Go CLI that configures the AI coding agents you already use: it gives them shared memory, a workflow and deterministic steps without replacing the agents them
28 skills that remove instead of add: rewrite a document clean, check the read of a request, compress text and challenge a decision
Comparison with similar tools
Entries that share the main search query, and how they differ.
README in short
Honey builds on two other skills, minimal code and terse prose. Against no skill on Claude Opus 5.5 it cuts output tokens by 40%, lines of code by 52% and single-call cost by 31%, with every test passing.