Claude Token Optimizer

🚀 Claude-Code-Enhancements

Custom Skills for the Claude Code CLI Agent to stop it from "thinking out loud", prevent context window bloat, and force professional engineering logic.Reduce Claude token usage by 60–80% while getting faster, cleaner outputs .

🧠 Why these skills?
Standard AI agents act like chatty assistants. They guess when a request is ambiguous, dump thousands of tokens of redundant code into your terminal, and delete things without confirming.

These custom .skill enhancements rectify that by introducing a structured, API-driven software development loop.

🛠️ The Protocols

1. Antigravity Protocol (V1 & V2.0)
The gold standard for daily coding.
*   Intent Rectification: If your request is vague, it stops and asks options instead of guessing.
*   Artifact-First Editing (V2.0): It stops terminal "yapping" by writing plans and long blocks to local .md files instead of printing to your chat window.
*   Persistent Memory: Saves architectural notes to prevent "session amnesia."

2. Ultimate Protocol Simulator
For maximum token efficiency in automated pipelines.
*   Zero-English Mode: Shuts off all conversational output.
*   JSON Only: Communicates status via minified JSON blocks.
*   Internal Sandbox: Runs local tests silently before reporting success.

📊 Performance & Selection Guide

Mode Selection Guide

Token Savings vs No Skill

*(See BENCHMARK.md for the raw logs of our 5-task stress test).*

The screenshot after i run some tests :

📥 Installation

1. Create a folder in your Claude skills directory:
   * Windows: C:\Users\YourName\.claude\skills\antigravity
   * Mac/Linux: ~/.claude/skills/antigravity
2. Download the SKILL.md file from the relevant folder in this repository.
3. Place it in the folder created above.
4. Open Claude Code and say:…