Linux Master
A Linux command-line trainer that runs entirely in the browser, in English and Hebrew. It drills commands with spaced repetition, checks typed answers by running them in a simulated shell, and ends with 12 terminal missions shaped like the first levels of OverTheWire Bandit. It is an installable, offline-capable PWA with no server, no account and no build step.
Live: https://saharhalili95.github.io/linux-master/
Features
- 8 levels, about 1,100 questions: 46 commands asked in both directions, all 450 usage lines from the Bandit guide, 75 pipe and workflow scenarios, and 30 Bandit-specific tools.
- Choice or typing, one tap to switch. Typed answers are normalised (flag order, merged flags like
-la, quoting, spacing). If the text still differs, the command runs in the simulated shell and is accepted when its output matches the reference answer.
- An explanation for every wrong answer: what the command you picked actually does, and why it doesn't solve this task.
- Report a problem on any question: one tap opens a pre-filled GitHub issue with the level, question id, options, correct answer and what the learner answered.
- Leitner review: mistakes come back 4 questions later and at the start of later rounds until answered correctly.
- Terminal missions: a simulated shell with pipes, redirection, globs,
find, awk, sed, strings, base64, tr and more. Every visit plants a fresh random password, so there is nothing to memorise.
- Habits: daily goal and streak, a calendar reminder (
.ics), per-level and per-command accuracy.
- Your data stays yours: progress lives in
localStorage, with export to file or clipboard and restore.
- Full English and Hebrew (RTL) interface and content, switchable at any time with shared progress.
Architecture
Plain HTML, CSS and classic scripts that share one global scope, loaded in order. No framework and no bundler.
flowchart LR
subgraph data
C[content.js<br>Hebrew questions] --> E[content.en.js<br>English overlay]
M[missions.js<br>12 mission builders]
end
S[shell.js<br>lexer, parser, virtual FS,<br>~30 commands] --> A
E --> A[app.js<br>i18n, Leitner engine,<br>typing check, UI]
M --> A
A --> L[(localStorage)]
W[sw.js<br>network-first service worker] -. offline cache .- A
- shell.js parses a POSIX-like command line (quotes,
|, &&, ||, ;, >, 2>/dev/null, globs, variables) and runs it on an in-memory file system with modes, owners and locked directories. sameOutput(a, b, fixture) returns true, false, or "can't tell" for unsupported commands, so the checker never guesses.
- Offline and updates: the service worker serves navigations network-first and falls back to the cache, and every asset URL is versioned, so phones pick up a new release on the next open.
Testing
Every push runs three CI jobs, and GitHub Pages deploys only after all of them pass.
| Job | What it checks |
|---|
| browser | About 380 Playwright tests on 4 devices (desktop Chrome, iPhone 13 WebKit, Pixel 7, iPhone SE): game flow, review scheduling, typing checks with hundreds of generated variations, both languages, offline and update flow, layout overflow, size budget, and axe WCAG 2.1 AA on every screen in both languages |
| ubuntu-answers | Runs every correct and wrong answer of levels 3-8 on real Ubuntu 24.04 in Docker. Then a differential test runs 570 commands both in the simulator and in Ubuntu and requires identical output, plus a safety check that none of the 190 wrong options is accepted by output |
| quality | ESLint, html-validate, and Lighthouse CI with hard thresholds (performance 90, accessibility, best practices and SEO 100; currently 98/100/100/100) |
A weekly job also checks that every link to the official Bandit pages is still alive.
Bugs the tests caught
- Wrong options that were actually right. The Ubuntu sandbox found a ROT13 distractor that decoded correctly, and a
| option that worked by race condition.
- Simulator drift. The differential test caught
file column alignment, wc field widths, ls with several operands, cp into a missing directory, and a missing /tmp.
- Output checking too lenient. The safety check found 15 wrong options that produced the "right" output in a too-simple fixture (for example
cat - waiting on the keyboard, or grep missing a line at the end of a file). The fixtures were made richer until it reached 0.
- Phones. The stats row overflowed on iPhone SE, and long answers overflowed narrow screens.
- Mixed languages. A scan for Hebrew characters in the English UI found sentences assembled from Hebrew fragments.
Run locally
cd tests && npm ci && npx playwright install chromium webkit
npx playwright test # browser tests (serves the site itself)
npm run sandbox # answers on Ubuntu in Docker
npm run diff # simulator vs Ubuntu
npm run quality # lint + HTML validation
npx lhci autorun # Lighthouse
node screenshots.js # regenerate README screenshots (site on :8766)
To add a language, add a dictionary to I18N in app.js, a content overlay to CONTENT, and the code to LANGS.
Install on a phone
Open the site in Safari or Chrome and choose Add to Home Screen. After the first load it works without a network.
