A free Claude Code plugin that scores how well you actually use AI, from your real coding sessions, and shows you exactly where to improve.
Prerequisites: python3 (3.8+) and node. The installer checks for both and tells you clearly if either is missing, it will not silently do nothing.
In Claude Code:
/plugin marketplace add jddavenportOpen/ai-fluency-trainer /plugin install ai-fluency@ai-fluency
Or one line in your terminal:
curl -fsSL https://clawdacademy.app/install.sh | bash
This installs the plugin, wires up the capture hooks, and claims you a handle.
Use Claude Code the way you normally do. The plugin runs quietly in the background and scores each turn on the seven behaviors that separate people who get smarter with AI from people who get passively dumber:
A statusline shows your live Fluency Rating, and a coach nudges you when you are about to do something that would lower it. Scoring is on your real behavior, not on whether you typed the magic words. Running the test counts as verification even if you never say "verify."
Your public profile lives at https://clawdacademy.app/u/<your-handle>:
clawdacademy.app/leaderboard ranks everyone by quality and is sortable and filterable. See where you stand and who is climbing.
Local-first by design. Your prompts and your code never leave your machine. Only aggregate numeric scores sync, and only if you opt in (a token in your config). Turn syncing off any time and the local coach still works.
No score showing up?
Confirm python3 and node are installed. The plugin needs both. On a fresh machine that is the usual cause.
Rating says "provisional"?
It firms up after about 15 scored turns.
Statusline missing?
Restart Claude Code once after installing.
Wrong handle or want to reset?
Re-run the installer, or claim a handle at clawdacademy.app/start.
Install it, work normally, check your profile. It tells you the truth about how you use AI and coaches you toward the habits that make you smarter with it.
MIT licensed · local-first · public source repo at launch