Get the RTK Operator Playbook

A 30-minute Claude Code session quietly burns 118,000 tokens — most of it noise. Boilerplate. Whitespace. The 2,000-line cargo test output where only three lines matter. A single Rust binary called RTK — 52,000+ stars in four months, Apache 2.0, by Patrick Szymkowiak and team at github.com/rtk-ai/rtk — cuts that to 23,900. Same Claude, same session, 80% of the bill — gone. This is the plain-English operator playbook: what RTK actually is (a CLI proxy that sits between Claude Code and your shell, not a model and not a wrapper), the 4 strategies it uses (smart filtering, grouping, truncation, deduplication), the one command that installs it (brew install rtk), the one command that turns it into a transparent auto-rewrite hook (rtk init -g — and the per-agent flags for Cursor, Codex, Gemini CLI, Windsurf, Cline), the 5 commands where the savings are biggest (cargo test 25,000 → 2,500 minus ninety; pytest 8,000 → 800 minus ninety; git push 1,600 → 120 minus ninety-two, RTK literally rewrites it to "ok main"; git diff 10,000 → 2,500 minus seventy-five; aws lambda list-functions verbose JSON → three columns), the two telemetry commands that tell you it is actually working (rtk gain for last-30-days savings, rtk discover for missed opportunities — run both weekly), and the two leaks that quietly kill your savings even after you install (Mistake 1: Claude Code's built-in Read, Grep, and Glob bypass the hook because it only catches Bash tool calls — fix by telling Claude in CLAUDE.md to use cat and rg for large reads; Mistake 2: treating RTK as a Rust-only tool because of the name — it is polyglot, pytest, go test, jest, vitest, rspec, rubocop, and golangci-lint are all covered with the same minus-ninety on every test runner).

By submitting, you'll be subscribed to the Hyperautomation AI Report newsletter — one practical AI email per week. Unsubscribe anytime.

Free. No spam. Unsubscribe anytime.