gitmyhub

atrophy

TypeScript ★ 127 updated 26d ago

Measure what your brain is losing while AI does your work - unaided-skill baselines, decay curves, reliance drills

A command line tool that measures whether relying on AI coding assistants is making your own unaided coding skills decline over time.

TypeScriptNode.jsSQLitePythonsetup: easycomplexity 2/5

Atrophy is a command line tool for developers who want to know if relying on AI coding assistants is quietly making them worse at coding on their own. It works like a fitness tracker for coding skill: you take a short exercise without any AI help, it grades your answer, and it keeps a rating for you that moves up or down over time, similar to a chess rating.

You start with a baseline session that takes about 25 minutes, solving one exercise for each of five skills: syntax recall, debugging, reading code and predicting its output, remembering standard library calls, and outlining a design. After that, you run short 5 to 10 minute drills two or three times a week, and the tool automatically picks whichever skill you have practiced least recently. Once a month you can also run a drill with AI help turned on, and those results are tracked separately so you can compare your unaided performance against your AI assisted performance over time.

A local dashboard shows one chart per skill, plus a chart comparing your unaided scores to your AI assisted scores side by side. If your unaided line drops while your AI assisted line stays steady, that gap is meant to show how dependent you have become on AI help. Exercises come in Python and JavaScript, and the tool tries to pick a difficulty where your chance of success is around 65 percent, since the README says that is where practice teaches the rating the most.

The README points to several outside studies, including ones from The Lancet, PNAS, and Anthropic, as reasons this kind of skill decay is worth measuring. All your data is stored locally in a single SQLite file on your own machine, with no account or syncing required unless you choose to opt into a public leaderboard. The tool is installed through npm and requires Node.js version 22 or higher, plus Python 3 if you want to use the Python exercises. It is open source under the MIT license.

Where it fits