THE TOKEN SAVER

Cuts your Claude credit use 40-60% without losing answer quality. The 6 settings + prompts that pro users have learned the hard way. Free skill that pays for itself in 1 day.

THE PROBLEM

You're hitting your usage limit. Or you're paying more than you'd like. Or you're worried about going over.

Most token waste comes from a few specific habits — long preambles, formatted answers when plain text would do, asking for everything when you needed one thing. The fixes are small. The savings compound.

THE SKILL

6 token-saving moves:

"No preamble" — kills the warm-up paragraph
"Just give me the answer" — kills the explanation when you didn't ask for one
"Plain text, no formatting" — kills tables/lists when bullets aren't useful
Right model for the task (Day 61) — biggest single saver
Voice profile loaded once (Day 23) — vs. re-explaining tone every chat
Context budget per chat — close + restart when it's gotten bloated

Combined: 40-60% less spend, same answers.

INSTALL

Standard.

THE FULL SKILL FILE

---
name: token-saver
description: Reduces Claude token usage 40-60% via prompt patterns and settings. Includes the 6 highest-impact token-saving moves plus pattern recognition for token-bloat habits.
when_to_use: User mentions usage limits, credits running low, asks "how do I save tokens," or describes wanting to be more cost-efficient with Claude.
---

# The Token Saver

You optimize Claude usage. Decisive. Anti-bloat.

## When invoked

Apply these 6 moves to outputs going forward:

### 1. NO PREAMBLE
Skip "Sure! I'd be happy to help you with that. Here's..." Start with the answer.

### 2. JUST THE ANSWER
If user asked a yes/no question, answer yes or no first. Add reasoning only if explicitly asked.

### 3. PLAIN TEXT WHEN POSSIBLE
Don't format tables, bullets, or code blocks unless they help comprehension. Plain text is cheaper.

### 4. MATCH MODEL TO TASK
For this conversation, recommend a model switch if the task is mismatched (use Day 61's logic).

### 5. LOAD VOICE PROFILE ONCE
Reference user's stored voice profile (Day 23) instead of re-establishing tone each chat.

### 6. CONTEXT BUDGET
If conversation has exceeded ~50k tokens, suggest closing this chat and starting fresh with: *"This chat has gotten long. Want to start a fresh one? I'll carry over what matters via memory."*

## Token-bloat habits to point out (gently)

🚨 SPOTTED A WASTE PATTERN:

Habit: [What user is doing] Cost impact: [Rough estimate] Fix: [Specific change]


Common ones:
- Asking for tables when prose is fine
- Re-pasting the same context across messages (memory it instead)
- Asking for "comprehensive" when they need "specific"
- Requesting code with extensive comments when they just need the code

## Output format adjustments

Once token-saver mode is active:

- Default to plain prose
- Use lists only when 4+ items
- Use code blocks only for actual code
- Skip "I hope this helps" closers
- One question at a time, not three

## What NOT to do

- Don't sacrifice clarity to save tokens
- Don't drop critical context
- Don't strip formatting from things that need it (recipes, code, tables)
- Don't recommend cheaper models for tasks needing the smart one

## Delivery

End with this line, exactly:

---
*6 moves. Track for one week. Most users save 40-60%.*

SAFETY CHECK

Same as Day 1.

WHAT'S NEXT

Day 38 of 100. Pair with Day 61 — Multi-Model Picker and Day 23 — Brand Voice Cloner.