Free kit · 5 skillsClaude Code3 built-in · 1 plugin · 1 custom

Claude is eating your tokens. Stop it with 5 skills.

Claude Code burns context on every file, command output, and message — and once the window fills, answers get slower and more expensive. Five moves keep it lean: see what your setup costs, make Claude write less, undo wrong turns, compress bloated sessions, and carry work into a fresh one without losing the thread.

Get the 5 skills free Why it matters

You commented TOKENS — here's all 5, the guide, and the custom /handoff skill to download. Honest note: the numbers here are from third-party tests and single examples — direction, not guarantees.

Why it matters

A full context window is a slow, pricey one.

Every file read, command output, and back-and-forth stacks into the context window. As it fills, each turn costs more and quality drifts. These five don't all cut raw token counts — some just keep the session clean and reviewable — but together they stop the slow bloat that makes Claude expensive.

5skills, one workflow

3built into Claude Code

1free plugin (Ponytail)

1custom skill in the kit

The five

Run them in this order.

Built in · diagnose

/doctor — see what your setup costs

Type /doctor and it health-checks your setup and estimates what your skills, memory, and extensions cost in context every session — then points at the biggest hogs. Run it first so you know where the waste is before you spend a prompt on it.

/doctor

Source: Claude Code skills docs.

Free plugin · write less

Ponytail — make Claude write less

A free, MIT plugin (98k+ stars) that tells the agent to act like the laziest senior dev in the room: less code, less explanation. Even when total tokens vary, shorter answers are faster to read and easier to review.

/plugin marketplace add DietrichGebert/ponytail
/plugin install ponytail@ponytail

Reddit test: −3% to −11% output tokens JetBrains test: −15% code · −10.3% cost

Directional, not guaranteed. Sources: repo, JetBrains benchmark, Reddit experiment.

Built in · undo

/rewind — undo a wrong path

Claude went down the wrong path? Don't spend ten messages correcting it — each one adds to context. /rewind jumps back to the last clean checkpoint before the mistake so you can try again.

/rewind

Source: Claude Code checkpointing docs.

Built in · compress

/compact — squeeze a bloated session

Summarizes the conversation and continues from the summary instead of the full history. One community example compressed 167K tokens into a ~25K summary — about 85%. Do it at a natural checkpoint, since compacting can lose detail.

/compact

Third-party example · not guaranteed · can lose detail

Sources: context-window docs, the 85% example.

Custom skill · in the kit

/handoff — carry work into a fresh session

Sometimes you want a clean /clear, not a compact — but without losing the thread. /handoff writes a compact brief (goal, decisions, changed files, next step, gotchas) that you copy, clear, and paste into the new session. Start fresh with almost no context, lose nothing that matters.

# install once — drop the handoff/ folder into ~/.claude/skills/
# then:
/handoff  →  copy the brief  →  /clear  →  paste it

Custom skill, included below. Format: Claude Code skills docs.

Download the kit (.zip)/handoff SKILL.md ↗The guide (.md) ↗Ponytail on GitHub ↗

Straight answers

The honest FAQ.

Do these guarantee lower costs?

No — and I won't pretend otherwise. The percentages are from third-party tests and single examples; they show the direction these tools push, not a promise. Your results depend on your codebase and how you work.

Do I have to install all five?

No. /doctor, /rewind, and /compact are built into Claude Code — nothing to install. Only Ponytail (a plugin) and the custom /handoff skill need a one-time setup.

How do I install the /handoff skill?

Download the kit, and copy its handoff/ folder into ~/.claude/skills/ so you have ~/.claude/skills/handoff/SKILL.md. Restart Claude Code if /handoff doesn't appear.

Isn't /compact the same as /handoff?

/compact keeps the same session going on a summary. /handoff is for when you'd rather start a truly clean session (/clear) but carry the goal and decisions across.

Credit

Ponytail is by DietrichGebert (MIT); /doctor, /rewind, and /compact are built into Claude Code. The /handoff skill and the guide are mine — free to use and tweak. Built something with these? Tag me on Instagram or come to the weekly call on Skool.

Get all 5 skills + the guide

Drop your email and I'll send the full guide, the download kit with the custom /handoff skill, and the Ponytail install. Plus every other resource in the library.

Keep going

More ways to make Claude cheaper and better.

Fewer tokens is one lever. These are the others — a cheaper worker model, the model pairing that keeps your limits alive, and the skills that fix how Claude works.

Systems & Agents

Kimi K3 inside Claude Code harness is a cheat code

Kimi K3 just dropped — a frontier-level model at roughly a fifth of the price. The play isn't to replace Claude; it's to make Kimi…

Read guide

Systems & Agents

Stop burning your Claude limits — the Fable + Sonnet setup

You hit Claude limits because your most expensive model does everything, including the grunt work. Three built-in Claude Code sett…

Read guide

Systems & Agents

The Claude skill for ADHD brains

Claude knows how to fix your problem — but buries it under 14 paragraphs first. This free Claude Code skill, "I Have ADHD", forces…

Read guide

Systems & Agents

person-to-skill — turn any expert into your AI advisor

Take any person and all their knowledge from books and podcasts, and turn them into a Claude skill with one line — then they act a…

Read guide

Browse the full library

Keep going

Cheaper context, better answers.

One email a week with the tools and setups that quietly cut your AI bill — the installs, the prompts, and the exact steps.

Join the newsletter Browse all resources