Claude Token Optimizer — AI skill for Claude Code

token-optimizer

A [Claude Code skill](https://code.claude.com/docs/en/skills) that audits your setup for token waste, applies safe fixes, and helps you cut per-session costs by 30-85%.

**What is a Claude Code skill?** Skills are reusable instruction sets that Claude Code loads on demand. They live in `~/.claude/skills/` (personal) or `.claude/skills/` (per-project) and activate automatically when you mention relevant keywords.

The problem

A typical Claude Code session burns tokens on things that don't help you:

• Terminal noise — npm test dumps 10,000 lines; Claude reads all of them. You needed 12.
• Bloat indexing — Every file search scans node_modules/, dist/, .next/ because there's no .claudeignore.
• Wrong model for the job — Opus on a one-line typo fix costs 10-15× more than Sonnet.
• Fat CLAUDE.md — 500 lines of instructions loaded every session, even when irrelevant.
• No output compression — No PreToolUse hooks means raw terminal output hits context every time.

This skill finds all of these, tells you how much each one costs, and fixes the safe ones automatically.

What it does

You: "Token kullanımını optimize et"

token-optimizer:
  [*] Scanning ~/.claude/
  [*] Scanning project at ./my-app
  [✓] 8 findings:
      🔴 No .claudeignore — node_modules (1.2 GB) being indexed
      🟠 No PreToolUse hooks — test output uncompressed
      🟠 Default model is Opus — 10-15× more expensive than Sonnet
      🟡 CLAUDE.md is 340 lines
      ...
      Estimated savings: ~65%

Then asks how you want to proceed: auto-fix everything safe, walk through each issue, or just keep the report.

Quick start

Option 1: Claude.ai upload

Settings → Capabilities → Skills → Upload `token-optimizer.zip`

Option 2: Claude Code (personal)

# Clone into your personal skills directory
git clone https://github.com/wagnercreative/token-optimizer.git ~/.claude/skills/token-optimizer

Option 3: Claude Code (per-pr