sattle/e2e/AGENTS.md

e2e — Playwright suite

Browser-level tests against the real app. Run: npm run test:e2e (test:e2e:ui for the UI mode).

ENVIRONMENT

• System Chromium, never downloaded browsers: CHROMIUM_PATH env or command -v chromium fallback (flake dev shell provides both). CI installs Playwright's own chromium — the config only forces executablePath when a system browser exists.
• webServer boots quasar dev -m spa --port 9333 (~15-25s ready), reuseExistingServer locally. serviceWorkers blocked.

STRUCTURE

• fixtures.ts — extended test with the mint fixture (MintMocker), routes unrouted after each test.
• helpers/MintMocker.ts — page.route mocks on never-resolving https://mint.test / https://mint2.test origins. Responses must carry Access-Control-Allow-Origin: *. IMPORTANT: mirror the real protocol — mintPubkey is announced on the mint-address (/.well-known/lnurlw/) and note-info responses, NOT the payRequest (a too-generous mock once masked a real bug).
• helpers/wallet.ts — createFreshWallet(page): drives onboarding UI to an unlocked 0-sat wallet.

CONVENTIONS

• No real network, ever — every external call goes through a route mock.
• First-contact trust prompt: receiving a note with a mintPubkey opens the "New mint" dialog — specs must click through it.
• Toasts duplicate in-page text — scope assertions to .q-page or the dialog, never the raw text.
• Dev-only test hooks (e.g. window.__sattleNwcTest) exist for injecting fake transports; never available in production builds.
• WebAuthn ceremonies are NOT e2e-testable headless — assert the honest unsupported state instead.
• Known lint drift: import() type annotations trip consistent-type-imports (pre-existing; the dev checker only lints app code, so CI/e2e are unaffected).