Halflife — Testing skill for Claude Code

halflife

[](https://github.com/shahil1dn/halflife/actions/workflows/ci.yml) [](LICENSE)

Spaced-repetition study tracking where your coding agent runs the test and a script owns the schedule.

It is a folder of markdown files and three small scripts. There is no app, no account, no UI and nothing to install. You point a terminal coding agent at the folder — Claude Code or similar — and it tests you, judges the answer against a bar you wrote in advance, and records the result. Passing pushes a topic further away, failing drags it back.

The part that makes it work: **nothing counts as known until you pass a test on it.** How confident you feel never moves the schedule, because confidence does not predict what you will remember.

Quickstart

git clone https://github.com/shahil1dn/halflife.git
cd halflife

Open your coding agent in that folder and paste this:

read AGENTS.md and set me up

The agent explains how the tool works, then asks about your course, your subjects, what mark counts as a pass, how long a session should be, and any exam dates. It writes those answers into `SETUP.md` and reads them at the start of every session afterwards.

Then it writes your first topics. For each one it asks whether you already know it, learned it recently, or have never learned it. That decides whether the first test is a week away, two days away, or has no date at all and waits in the backlog.

`SETUP.md` ships as a questionnaire and is rewritten in place during setup. To get the blank one back, run `git checkout SETUP.md`.

A day of use

Say `test me`. The agent runs `due.sh`, which reads the topic files and prints what is due:

=== DUE TODAY (2026-09-01): 1 ===
MATH101      long-division                    [A] learning p:1 f:0  topics/math101-long-division.md

=== TO LEARN / RELEARN (no date, your choice)