pi-dedumbify

Pi extension for executable spaced repetition with FSRS scheduling.

Write real TypeScript or Python code, run tests in an isolated temp workspace, then rate the review to schedule the next interval.

Install in pi

From npm:

pi install npm:pi-dedumbify

From GitHub:

pi install git:github.com/lukaskawerau/pi-dedumbify

Then restart pi or run /reload.

What it does

• centered review overlay inside pi
• user-authored global card deck
• TypeScript cards graded via Vitest
• Python cards graded via uv + Pytest
• FSRS scheduling via ts-fsrs
• automatic review persistence in SQLite

Commands

• /sr — auto-sync cards, then open review modal
• /sr-sync — force a card rescan + DB sync
• /sr-stats — sync cards, then show DB-backed deck stats
• /sr-validate — structural validation + run solution tests without writing reviews

Keyboard shortcuts in the modal

• tab / shift+tab — switch panes
• ctrl+r — run tests
• ctrl+s — toggle starter/solution
• 1 / 2 / 3 / 4 — Again / Hard / Good / Easy
• esc — close

Card location

Cards live under:

~/.pi/agent/spaced-rep/cards/

DB lives at:

~/.pi/agent/spaced-rep/fsrs.db

Card format

Each card lives in its own directory.

TypeScript example:

sum-array-ts/
  card.yaml
  prompt.md
  starter.ts
  solution.ts
  tests.ts

Python example:

factorial-py/
  card.yaml
  prompt.md
  starter.py
  solution.py
  tests.py

Minimal card.yaml:

id: sum-array-ts
title: Sum an array of numbers
language: typescript
tags:
  - arrays
  - iteration
timeboxSec: 180
files:
  prompt: prompt.md
  starter: starter.ts
  solution: solution.ts
  tests: tests.ts
runner:
  entry: answer.ts

Local development

cd ~/coding/apps/dedumbify
npm install
npm run check
pi

The repo exposes a project-local extension shim at .pi/extensions/dedumbify.ts, so pi auto-discovers it when started from the repo root.

Status

MVP works.

Implemented:

• card discovery + validation
• SQLite deck state + review log
• grading runners for TS and Python
• FSRS scheduling
• review modal with answer buffer and autosave on rating

Still rough:

• result rendering polish
• richer session stats
• nicer reveal-solution flow