Stop Wasting Tokens in Claude Code
Most Claude Code projects spend half their context on files Claude doesn't need to be reading. The 3-step fix that cut my token usage by 60% on Cadence's main repo.
Get the full PDF.
Save it, print it, read it offline. Once you unlock you'll see every resource on the site. Unsubscribe whenever, no spam.
Coming from Instagram? Comment TOKENS on the post and I'll DM the link straight to you.
Already on the list? You'll see the download once you submit.
What this is
If you've used Claude Code on a real project, you've seen it. Every prompt, the agent re-reads half the codebase looking for context it already had. You watch the token counter climb. The work gets slower. The bill gets bigger.
The cause is almost never Claude. It's the absence of a navigation system in your repo.
The three rules
You only need three things in place. Once they're in place, the bleed stops.
- Claude has a map. Not a README. A structured index that says "this file does this, that file does that."
- Claude has a rule that says use the map first. This goes in CLAUDE.md and it's the single most expensive sentence I've ever written.
- The map updates when the code does. Otherwise it goes stale, and stale is worse than nothing.
What the bleed looks like
Here's a concrete example from the Cadence repo before I fixed this.
| Metric | Before | After |
|---|---|---|
| Tokens per "fix this bug" turn | ~42,000 | ~16,500 |
| Files read per turn | 14 | 4 |
| Cost per session | $1.80 | $0.65 |
Same model, same task, same repo. The only thing that changed was that Claude stopped reading files it didn't need.
What you need
Three things, all of them text files. No tools to install.
- A repo with at least 20 files where you actually use Claude Code
- 10 minutes
- Willingness to put rules in your CLAUDE.md and trust them
The full PDF walks through the exact prompts, the file map shape Claude needs, what the CLAUDE.md header looks like in practice, and a dummy project you can test on first so you don't blow up your real one.
Drop your email below or comment TOKENS on the IG post and I'll DM it to you.
The complete version goes deeper.
- →Prompt 1: generate a structured map of your repo so Claude can navigate without re-reading
- →Prompt 2: the navigation rule that goes in your CLAUDE.md
- →The CLAUDE.md header template I run on every project
- →A patch prompt to keep the map current as you ship
- →Global CLAUDE.md rules that apply across every repo on your machine
- →A dummy project structure to test the whole thing before you touch real code
- →Back-of-napkin token math: what the bleed actually costs you
Scroll up to grab it. Or comment TOKENS on Instagram.
// work with me
Want this stack inside your business?
I work with a small number of operators per quarter. If your problem looks like the ones I write about, let's talk.
Work with me