Stop Burning Claude Credits
Every git status, every grep, every paragraph of polite prose costs tokens. Two tools cut a normal session's bill by 90%. Setup takes ten minutes.
Instagram Carousel
Two tiny tools, one habit. RTK filters bash output before it reaches the model, Caveman mode compresses the model's replies. Together they take 90% off a typical session.
Export
7 PNGs in one ZIP
Every git status, every grep, every paragraph of polite prose costs tokens. Two tools cut a normal session's bill by 90%. Setup takes ten minutes.
Where Tokens Go
Look at any long session and the cost lives in two places. Bash output piped back to the model, and the model's own verbose explanations.
Tool 1
Rust Token Killer is a CLI proxy. It runs your commands, strips noise, and hands the model only the parts that matter. Same workflow, smaller output.
# Install once
brew install rtk
# Verify
rtk --version
rtk gain # see lifetime savingsTool 2
Caveman is a skill that tells Claude to drop filler. Same answers, 75% fewer words. Plain English on demand for the few times you want it.
# Turn on
/caveman
# Or save forever in
~/.claude/CLAUDE.md
# Levels: lite, full, ultraCombined Effect
RTK cuts the input side. Caveman cuts the output side. Together they drop a typical session from 200k tokens to under 25k. Same work done.
Gotchas
RTK is opinionated about find syntax and a few other commands. Caveman keeps tech accuracy but can read terse to newcomers. Both fixable.
Closing
Install both tonight. Tomorrow's first session uses 10% of what last week's did. The savings compound every session you run after that.
Same outputs, smaller bill. The cheapest performance upgrade you can ship today.
Slide 1 of 7