Low Token Usage Skill

Use this skill when:

• Token usage is a concern
• You want maximum information density
• Process details aren't needed, just outcomes

Rules

1. No Process Narration

• ❌ "First I'm going to check the file..."
• ❌ "Let me look at the code..."
• ❌ "I need to update..."
• ✅ Direct edits, then summarize what changed

2. No Step-by-Step

• ❌ "Step 1: Open file. Step 2: Find function..."
• ❌ "I'll do this in three parts..."
• ✅ Complete the task, then present final result

3. No Filler

• ❌ "Great idea!", "Let me see...", "Hmm..."
• ❌ Excessive politeness or hedging
• ✅ Neutral, direct statements

4. Results Only

• ❌ "I think we should..." or "Maybe try..."
• ❌ Speculating about approaches
• ✅ "Done." or "Failed: [error]"

5. Batching

• Group related changes into single operations
• Avoid one-tool-call-per-item loops
• Prefer bulk operations

Examples

High Token (Bad):

"Let me take a look at the file first. I'll open it and see what's there. Okay, so I can see the function on line 45. Now I'm going to make an edit to fix the bug. I'll replace the old code with the new code. Let me do that now..."

Low Token (Good):

Fixed. Changed foo() to bar() on line 45.

Activation

This skill is automatically active when:

• User mentions "low token" or "save tokens"
• Token usage exceeds 75% of context window
• User explicitly requests minimal output

Metrics

Target: \x3C50 tokens for simple confirmations Target: \x3C200 tokens for complex results Avoid: >500 tokens for routine tasks