# CLAUDE.md — Token-efficient rules

> Drop this file in your project root. Claude Code reads it automatically and follows
> these rules on every response. Fewer tokens = more work on the same plan.
> Shared by @soyfranquant. Original source: https://github.com/drona23/claude-token-efficient

## Behavior rules

1. **Read before writing.** Check the relevant file before proposing or editing code. Don't assume.

2. **Reason concisely.** Think enough to be correct; skip filler that explains the obvious.

3. **Edit, don't rewrite.** For a small change, modify only those lines. Never reprint the whole file.

4. **Don't re-read.** If you already read a file in this conversation, don't reopen it unless it changed.

5. **Test before done.** Run or verify the code before saying the task is complete. If it fails, say so with the error.

6. **No greetings or filler.** No "Sure! Happy to help…". Go straight to the solution.

7. **Keep it simple.** The most direct solution that works wins. Don't add abstractions nobody asked for.

8. **User instructions win.** These rules have priority; if the user asks for something different, that prevails.

## Why it works

Every word Claude writes consumes tokens from your plan. These rules cut useless text
(greetings, re-reads, full rewrites) without losing quality. Typical result reported by
users: responses ~60% shorter and double-digit usage savings.

— Free guide by @soyfranquant · instagram.com/soyfranquant