Token Optimization Kit
Find out where your Claude context actually goes — then get it back.
Most people running Claude every day have no idea how much of the context window is gone before they type a word. It is rarely the work — it is an oversized CLAUDE.md, rules with no path scoping, MCP servers loading whether you use them or not. This kit measures that first and reports real numbers off your real files, then restructures the setup so the numbers come down. It is not a list of tips.
What you can produce with it
- A per-source breakdown of what is loaded before your first message
- Findings ranked by severity, each with evidence and a named fix
- A lean CLAUDE.md, with the bulk moved to skills that load on demand
- Rules that stop loading at session start because they are path-scoped
- A .claudeignore written for your stack
What is inside
3 Contents
-
01
token-audit
Reads your real configuration — CLAUDE.md, rules, agents, skills, MCP servers, settings — and reports how much of the context window is consumed before you type anything. Ranks findings CRITICAL to LOW with evidence and a fix for each. Read-only; it never modifies a file.
-
02
context-diet
Applies the fixes. Splits an oversized CLAUDE.md into a lean core plus on-demand skills, adds paths frontmatter so rules stop loading at session start, writes a .claudeignore, and can add output-filtering hooks. Shows a plan and waits for confirmation, backs up every file before writing, and never deletes content.
-
03
Token Kit Guide
Ten pages — install, where tokens actually go, a catalog of 18 techniques ranked by impact, compact-versus-clear, config restructuring, the Claude.ai and Desktop variant, and a one-page reference card.
Specification
| Type | Skills · Claude Skill (token-audit), Claude Skill (context-diet), PDF guide, 10 pages |
|---|---|
| Contents | 3 |
| Install | Under 5 minutes |
| Runs in | Claude Code, Claude Desktop, Claude.ai |
| Made for | Claude Code and Claude Desktop daily users |
| Delivery | Instant download — all sales final |
Token Optimization Kit measures where your Claude context is going, then gives you the tooling to take it back. Two skills and a ten-page guide.
token-audit scans your global and project CLAUDE.md, your rules and whether
each is path-scoped, your agents, installed skills, MCP servers, .claudeignore
and settings — then reports the always-loaded token total, its share of the
window, a per-source table, and a prioritised action list.
context-diet does the remediation, with a confirmation step and a backup
before every write.
Buy this if
- You use Claude Code or Claude Desktop most working days
- You have watched quality degrade partway through a long session
- Your CLAUDE.md has grown and you have stopped reading it
- You want measurement before you change anything
Do not buy this if
- You use Claude occasionally and have never hit a limit
- You want a billing or spend dashboard — this works at the context level
- You want savings without changing how your project is set up
What is not included
Listed here rather than discovered after payment. Sales are final, so the scope has to be unambiguous before you buy.
- It does not reduce usage by itself — token-audit measures, you decide, context-diet applies
- No percentage saving is claimed, because the result depends entirely on your current setup. A clean setup gets a short report saying so.
- Token counts are estimates at roughly four characters per token, not tokenizer output. Claude Code's own /context and /usage are authoritative and the kit tells you to run both.
- No billing or invoice integration; it never reads your account spend
- No proxy or middleware — nothing sits between you and Claude
Questions
How much will it actually cut?
We do not put a number on it, and the kit does not either. The result depends entirely on what your setup looks like now — someone with a 40KB CLAUDE.md and six always-on MCP servers has far more to recover than someone already lean. The audit gives you your number before you change anything, and if your setup is already tight it says so in a short report rather than inventing work.
Will it change my files?
token-audit never writes anything — it is read-only by design. context-diet does write, but it shows you the plan first and waits for confirmation, backs up every file before touching it, and moves or archives content rather than deleting it. It also refuses to restructure a setup that is already lean.
Does it work in Claude.ai, or only Claude Code?
Both. In Claude Code it reads your actual config files. In Claude.ai and Claude Desktop there are no config files to read, so it runs a behavioural variant instead — the guide covers that difference in section 06.
What do I need before installing?
A Claude plan with code execution enabled — Settings, Capabilities, Code execution and file creation. That is it for the skills. Python 3.8 or newer only if you want to run the audit script standalone outside Claude, which is optional and has no dependencies.
Why is this the cheapest kit?
It is the front door. The pain is close to universal among daily Claude users, it is the fastest kit to demonstrate, and its job is to show you whether these kits are worth buying at all — not to make money.
Is this a list of prompting tips?
No. The skills read your actual configuration and report real numbers off your real files. The 18 techniques in the guide are there to explain what the findings mean and what to do about them, not as the product.
Token Optimization Kit
3 contents, instant download, all sales final.