For developers on Claude Code and Codex
Stop hitting your Claude Code and Codex limits halfway through the day
Ready-made files that stop your coding agent wasting tokens on stale context, giant file reads, noisy test logs and an expensive model doing simple work. Set up in about 10 minutes.
One-time payment. Instant download. Use it in every project.
- Works with Claude Code and Codex
- Checked against the official docs, October 2026
- Plain text files you can read before you run them
- Full refund if a file is broken
token-saver-kit/
- ├─ START-HERE.md
- ├─ GUIDE.md
- ├─ claude-code/CLAUDE.md
- ├─ claude-code/settings.json
- ├─ claude-code/hooks/block-big-reads.sh
- ├─ claude-code/hooks/filter-test-output.sh
- ├─ claude-code/agents/scout.md
- ├─ claude-code/agents/test-runner.md
- ├─ claude-code/skills/codebase-map/
- ├─ claude-code/glm-settings.example.json
- ├─ codex/AGENTS.md
- ├─ codex/config.toml
- ├─ codex/deep.config.toml
- ├─ prompts/
Where your tokens actually go
Most usage limits aren't hit by hard work. They're hit by waste the agent repeats every message. The kit fixes the five biggest leaks.
01
Your CLAUDE.md or AGENTS.md is hundreds of lines long, and it's re-sent with every single message.
The fix: Lean templates under 100 lines, with long instructions moved into skills that load only when needed.
02
The agent opens lockfiles, build output, node_modules and 5,000-line files in full.
The fix: A settings file that blocks those reads, and a hook that turns huge-file reads into a search or a line range.
03
One test run dumps thousands of lines of passing tests into your context.
The fix: A hook that trims test output to the failures, and a cheap Haiku test-runner that reports back a summary.
04
Every session starts by re-exploring your whole repo with Glob, Grep and Read.
The fix: A codebase-map prompt and skill, so the agent reads one short map instead of searching from scratch.
05
The most expensive model does everything, including renaming a variable.
The fix: A Codex config that defaults to gpt-6-luna with a "deep" profile for hard tasks, and model guidance for Claude Code.
What's inside
Plain files you copy into your project. No app to install, no account to create.
- START-HERE.md
- The 10-minute setup, in order of impact, with a checklist
- GUIDE.md
- Why each step works, with sources, and when not to use it
- claude-code/CLAUDE.md
- Lean memory template (fill in the blanks)
- claude-code/settings.json
- Blocks reads of dependencies, build output, lockfiles and .env files
- claude-code/hooks/block-big-reads.sh
- Stops whole-file reads of files over 1,500 lines
- claude-code/hooks/filter-test-output.sh
- Trims test-runner output to failures only
- claude-code/agents/scout.md
- Read-only explorer on Haiku that returns a short summary
- claude-code/agents/test-runner.md
- Runs your tests on Haiku and returns only the failures
- claude-code/skills/codebase-map/
- Makes Claude read your map before it searches
- claude-code/glm-settings.example.json
- Optional: route to the cheaper GLM Coding Plan
- codex/AGENTS.md
- Lean memory template for Codex
- codex/config.toml
- Defaults to gpt-6-luna, OpenAI's most efficient model
- codex/deep.config.toml
- A "deep" profile for the hard tasks
- prompts/
- Make a codebase map, plan-then-build, and fresh-session handoff
See a file before you buy
This is the read-blocking part of claude-code/settings.json, exactly as it ships. The kit adds the two hooks it points to.
claude-code/settings.json
{
"permissions": {
"deny": [
"Read(./node_modules/**)",
"Read(./dist/**)",
"Read(./build/**)",
"Read(./.next/**)",
"Read(./coverage/**)",
"Read(./package-lock.json)",
"Read(./pnpm-lock.yaml)",
"Read(./yarn.lock)",
"Read(./.env)",
"Read(./.env.*)"
]
},
"hooks": {
"PreToolUse": [
{ "matcher": "Read", "hooks": [{ "type": "command",
"command": "${CLAUDE_PROJECT_DIR}/.claude/hooks/block-big-reads.sh" }] },
{ "matcher": "Bash", "hooks": [{ "type": "command",
"command": "${CLAUDE_PROJECT_DIR}/.claude/hooks/filter-test-output.sh" }] }
]
}
}Set up in about 10 minutes
Step 1
Measure first
Run /usage and /context in Claude Code, or /status in Codex, so you can see the difference afterwards.
Step 2
Copy the files in
Drop the templates, settings, hooks and subagents into your project, and the Codex config into ~/.codex. START-HERE.md walks you through it in order.
Step 3
Work as usual
Fresh sessions every 15–20 messages, the cheap model by default, and a map instead of re-exploring. Check /usage again after a few days.
It's for you if
- You're on Claude Pro or ChatGPT Plus and hit your limit before the day is over
- You use Claude Code, Codex, or both, on real projects
- You'd rather copy working files than piece settings together from docs
Skip it if
- You only use the chat apps, not Claude Code or Codex
- Your projects are a handful of files (there isn't much to save)
- You want a guaranteed saving: it depends on how you work, so we don't promise a number
Get the Token Saver Kit
$4.99
One-time payment. Instant download. Use it in every project.
- Every file listed above, as one ZIP
- Claude Code and Codex versions
- START-HERE setup and the full GUIDE
- Use it in all your projects, including client work
Secure checkout by Whop. If a file is broken, email contact@aieatingtheworld.com and we'll fix it or refund you. See our refund policy.
Checkout not loading? Pay on Whop instead.
Questions
Will it actually save tokens?
It removes the most common sources of waste: oversized memory files, reads of files the agent doesn't need, noisy test output, re-exploring the repo and using the top model for everything. How much you save depends on your projects and habits, so we don't promise a number. The guide shows you how to measure before and after with /usage, /context and /status.
Does it work with Codex as well as Claude Code?
Yes. There's a Codex folder with an AGENTS.md template, a config.toml that defaults to gpt-6-luna and a deep profile for hard tasks. The prompts work in both.
Mac, Linux or Windows?
The templates, settings, subagents and prompts work everywhere. The two hooks are shell scripts that need bash and jq, so on Windows run them in WSL or Git Bash, or skip them.
Is it safe to install?
Every file is plain text you can read first. The hooks only block or shorten output; they don't send anything anywhere. Nothing is installed globally except the Codex config, and only if you copy it.
Couldn't I set this up myself from the free guides?
Yes, and the guides on this site explain how. The kit saves you the evening of reading docs and writing the settings, hooks and subagents yourself, and it's checked against the current Claude Code and Codex docs.
How do I get the files?
Right after you pay, Whop takes you to your purchase, where the ZIP is ready to download. To download it again later, sign in to Whop with the same email.
Can I get a refund?
It's a download, so it's non-refundable once downloaded, unless a file is broken. If something doesn't work, email contact@aieatingtheworld.com and we'll fix it or refund you.
Not ready? Start with 3 free files
Get the lean CLAUDE.md and AGENTS.md templates and the fresh-session handoff prompt from the kit, free. They fix the two biggest leaks: oversized memory files and long sessions.
- claude-code/CLAUDE.md
- codex/AGENTS.md
- prompts/handoff.md