AgentFit tracks your Claude Code and Codex sessions, then scores your Human-AI coding proficiency using the CRAFT framework — five behavioral dimensions you can actually improve.

The Framework
Inspired by DORA Metrics and Microsoft's SPACE framework. Behavioral signals from your logs — no surveys, no guesswork.
How effectively you engineer context for the AI
CLAUDE.md usage, memory writes, cache hit rate
How broadly you leverage available capabilities
Tool diversity, subagent usage, skill adoption
How independently the agent works for you
Message ratio, interruption rate, delegation
How consistently you maintain a coding rhythm
Streak length, daily consistency, active days
How much output you get for your investment
Cost efficiency, output volume, error rate
Scoring: Each dimension is 0–100. Autonomy weighs 25%, Context/Reach/Throughput 20% each, Flow 15%.
What You Get
Reads Claude Code and Codex conversation logs automatically. No setup, no surveys — real data from real work.
Personalized coaching with letter grades, actionable tips, achievement badges, and behavioral insights.
Token breakdowns, tool usage patterns, project analytics, personality fit — every angle covered.
All data stays on your machine in a local SQLite database. Nothing leaves your device.
Extensible architecture lets anyone build custom analysis views and ship insights as a plugin.
Share your metrics to compare with the community. See how your CRAFT score stacks up.
Get Started
Requires Node.js 20+. Works with Claude Code and Codex out of the box.
Or download the desktop app →npx agentfitcurl -fsSL https://raw.githubusercontent.com/harrywang/agentfit/main/setup.sh | bashgit clone https://github.com/harrywang/agentfit.git && cd agentfit && npm start