home/writing/ch05-token-sinks-2

Ch05 Token Sinks

GPTClaudeGemini··▲ 134 copies·updated 2026-07-14

ch05-token-sinks-2.prompt

# Chapter 5 — Cut Your Token Bill in Half

> **Hours 15-20.** The six token sinks, the three-line formula, and the A/B audit.
> "This chapter pays for the book." (ch05:7)

**The setup:** the author's first month with Claude Code cost **$1,400**. Three
weeks, 2,000 lines of useful code. Seventy cents a line. "The tool was about 2,000
lines of code. The other 218,000 lines of context Claude processed were waste."
(ch05:5)

**A token is roughly 4 characters** (ch05:25). Input tokens are everything Claude
*reads* — your prompt, the files, command outputs, conversation history. Input
dominates, and the biggest input cost is file reading. **The context window is your
biggest lever.**

> Measure your own project before you start:
> `node dist/tools/token-estimator.js src/`

---

## Sink 1 — Conversations that run too long

**ch05:29-37** · "Every message includes all previous messages. Message 50 pays for the full history."

**One conversation per feature.** Start fresh with a one-line handoff:

when to use it

Community prompt sourced from the open-source GitHub repo regardo911/claude-code-beginners (MIT). A "Ch05 Token Sinks" style prompt — adapt the placeholders and specifics to your task. Imported as-is and not independently retested here, so check the output before relying on it.