Does Token Coach support both Claude Code and Codex?

Yes, Token Coach is designed to detect your runtime and provide native advice for both environments, translating concepts between Claude.md and AGENTS.md automatically.

What is a Context Quality Score?

It is a metric provided by the underlying token-optimizer that measures how much of your current session is useful content versus overhead; scores below 70 usually trigger immediate optimization suggestions.

What are ghost tokens in AI context windows?

Ghost tokens are hidden overhead in your AI context, such as unused skills, redundant prompts, or excessive metadata that consume your context window without adding value to the current task.

How does the coaching session work?

It starts with a quick diagnostic of your current setup, asks about your specific goals (new build, fixing lag, or agent design), and then provides a conversational guide through the most impactful optimizations.

Token Coach

Name: Token Coach
Author: alexgreensh

byalexgreensh

•

862

•

분석 및 모니터링

Optimizes AI context windows by identifying token-wasting patterns and providing proactive architecture guidance for Claude Code and Codex projects.

Token Coach acts as an interactive performance consultant for AI-driven development, helping engineers maintain high context quality while minimizing token overhead. By analyzing skill architecture, session-level metrics, and multi-agent setups, it provides data-driven advice to prevent common pitfalls like the '50-Skill Trap' and context decay. Whether you are bootstrapping a new project or troubleshooting a sluggish existing environment, Token Coach offers actionable steps and estimated token savings to ensure your AI assistant remains fast, accurate, and cost-effective.

주요 기능

01862 GitHub stars

02Detection of 'ghost tokens' and high-overhead architectural patterns

03Specialized architecture advice for multi-agent system design

04Interactive performance coaching based on real-time project metrics

05Prioritized action plans with estimated token savings for every fix

06Context quality scoring with specific session-level health insights

사용 사례

01Optimizing Codex environments for better response speed and accuracy

02Designing token-efficient multi-agent architectures from the ground up

03Diagnosing and fixing context window bloat in existing Claude Code skills

주요 기능

01862 GitHub stars

02Detection of 'ghost tokens' and high-overhead architectural patterns

03Specialized architecture advice for multi-agent system design

04Interactive performance coaching based on real-time project metrics

05Prioritized action plans with estimated token savings for every fix

06Context quality scoring with specific session-level health insights

사용 사례

01Optimizing Codex environments for better response speed and accuracy

02Designing token-efficient multi-agent architectures from the ground up

03Diagnosing and fixing context window bloat in existing Claude Code skills