LiaPengembang

Mulai

Semua yang dibutuhkan asisten coding AI untuk membangun aplikasi Lia dengan benar. Dokumen dalam Markdown, llms.txt, spesifikasi OpenAPI server API, tipe TypeScript, dan file aturan untuk proyek Anda.

Dokumen ini ditulis agar bisa dibaca oleh orang dan alat AI. Arahkan asisten Anda (Claude, ChatGPT, Cursor, Copilot, Claude Code atau yang lain) ke file di bawah ini, dan tambahkan file aturan ke proyek Anda, sehingga menulis kode Lia dengan benar sejak awal.

#Untuk asisten Anda

ApaDi manaGunakan untuk
Peta dokumen/llms.txtEssensial dan setiap halaman, dengan tautan ke Markdown-nya. Mulai di sini.
Semua dokumen dalam satu file/llms-full.txtTempel atau lampirkan saat alat Anda dapat't menjelajah.
Setiap halaman sebagai MarkdownTambahkan .md ke alamatnya, seperti /concepts/payments.mdSatu topik pada satu waktu. Setiap halaman's Salin halaman menu juga melakukannya.
Server API sebagai OpenAPI 3.1/openapi.jsonMembuat klien berbasis tipe, atau membiarkan agen memanggil API dengan benar.
Sebuah file aturan untuk proyek Anda/AGENTS.mdDi bawah: aturan yang paling penting, dalam file yang dibaca asisten.
Jenis TypeScriptnpm install @lia-labs/sdkSetiap metode'params dan jawaban, setiap peristiwa'data.

#Salin halaman

Setiap halaman memiliki Salin halaman menu di samping judulnya:

  • Salin sebagai Markdown menempatkan halaman tersebut ke papan klip Anda, siap untuk ditempel ke dalam obrolan.
  • Lihat sebagai Markdown membuka Markdown-nya.
  • Buka di Claude dan Buka di ChatGPT mulai obrolan yang membaca halaman.

#Tambahkan aturan ke proyek Anda

Simpan sebagai AGENTS.md di proyek Anda (atau CLAUDE.md untuk Claude Code, atau aturan dalam .cursor/rules/). Asisten membacanya sebelum menulis kode, jadi mereka mengikuti Lia'aturan tanpa diberitahu setiap kali. Itu's juga di /AGENTS.md.

AGENTS.md
# Building a Lia app

This is a web app that runs inside the Lia app (lialabs.org), like a Telegram mini app. Docs: https://docs.lialabs.org (every page as Markdown: add .md; everything: https://docs.lialabs.org/llms-full.txt).

## The SDK
- In a page without a build step: <script src="https://api.lialabs.org/sdk/lia-apps-v1.js"></script> gives window.Lia. Never copy or bundle that file.
- With a bundler: npm install @lia-labs/sdk (or @lia-labs/sdk-react, -preact, -vue, -svelte, -solid, -angular, @lia-labs/elements). Use one or the other on a page, never both.
- Call init() first, then ready() once the first screen is drawn. Outside Lia every call fails with code "not_in_lia"; for local development call mockLia() when isLia() is false.
- Every failure is an Error with a code. "cancelled" from pay() is a normal no, not an error to report.

## Trust
- Never trust the page. Your server verifies a session token (from init() or session(), 15 minutes) with POST https://api.lialabs.org/api/apps/v1/session/verify and takes the player id from Lia's answer, never from the page.
- Secret keys (lia_test_..., lia_live_...) and the webhook secret (whsec_...) live on the server only. The app id is the only public identifier.
- Prices come from Lia's records: the page passes a product id to pay(), never an amount.

## Payments
- pay(productId) opens Lia's pay sheet and answers { chargeId, productId }. The server then checks GET /charges/:id (its player, product and status), gives the item once per charge id, and calls POST /charges/:id/deliver. If it can't deliver, POST /charges/:id/refund.
- On opening, call charges() and deliver any paid, undelivered charges; listen with onCharge() too. Webhooks (charge.paid) are a nudge: check with the API, and verify the Lia-Signature header (HMAC SHA-256 of "<t>.<raw body>" with the whsec_ secret).

## Rewards, ads, notifications
- Pay players in $LIA only from the server: POST /rewards with a key that names the reason (a key pays once, ever). At most 10,000 LIA a reward; handle pool_empty and daily_limit.
- A rewarded video answers { watched, adId }: give the reward only after the server checks POST /ads/verify (watched, and the same player), once per adId.
- Between rounds and games, call Lia.showInterstitialAd() (Lia paces it: about one every 90 seconds and every 2nd round; "too_soon" means carry on). Rewarded videos only behind a button the player chooses (a hint, a second chance), never by themselves between rounds. Test keys always get Google's test ads.
- Banners and native ads (SDK 1.7): Lia.showBanner({ position: "bottom" | "top" }) and Lia.showNativeAd({ placement: "card" | "sheet" }). Lia draws them natively beside the page (the page gets shorter), never inside it: don't build an ad slot in your own HTML for them, just let the layout follow the viewport height. Show the card at pauses (a results screen) and hideNativeAd() when play resumes; both answer { shown: false, reason } when there's nothing to show, so carry on. For an ad inside the page's own content, use the developer's own AdSense or Google Ad Manager tags: Lia lets only Google's ad frames load and opens ad taps in Safari; that revenue is the developer's own, not counted by Lia. For games: pin Lia.showBanner({ position: "top" }) before Lia.ready() and keep it for the visit; put in-page ad rows only on calm screens (home, lobby), never over or under the board; don't use the native card or sheet over gameplay.
- Notifications are opt-in (requestNotifications()), sent by the server with POST /notifications, at most 3 a day per person from one app, an hour apart, 180 characters, no links. Use a startParam to open the right screen.

## Feel like Lia
- Use Lia's theme (CSS variables --lia-bg-color, --lia-text-color, --lia-hint-color...), the safe areas (--lia-safe-area-inset-bottom), Lia.language (13 languages) and Lia.rtl for dir.
- Use Lia's mainButton for the main action, backButton for your own screens, Lia's dialogs and toasts, and play sound only while soundOn.
- Only your own pages load in the frame: open other sites with openLink(url).

## Conventions
- In text people read: "$LIA" for the token, "LIA" next to numbers (5 LIA).
- Retry server POSTs with the same Idempotency-Key header.

#Ajukan pertanyaan yang baik

Beberapa prompt yang bekerja dengan baik dengan file di atas:

#Jika sesuatu'salah

Alat AI masih bisa salah. The Referensi SDK dan server API adalah sumber kebenaran, dan Menguji aplikasi Anda menunjukkan cara memeriksa semua hal sebelum Anda siap rilis.

Diterjemahkan otomatis. Bahasa Inggris adalah sumber kebenaran. Baca dalam Bahasa Inggris

Ketik untuk mencari setiap halaman, metode, dan titik akhir.