OnStage TW（幕間）

See every theatre performance in Taiwan in one place.

Taiwan's theatre ticketing is fragmented across OPENTIX, udn, KHAM, ERA, KKTIX and Accupass. OnStage TW scrapes all of them into one searchable page, so you can find what's on — and when it goes on sale — without checking each site.

Theatre only (stage plays, musicals, traditional opera, puppetry, children's theatre, crosstalk, comedy…); concerts and music recitals are excluded.

🔗 Live: https://onstage.takalawang.dev · 📦 Repo: https://github.com/TakalaWang/onstage-tw

UI is in Traditional Chinese (the audience is in Taiwan). Code, docs and this README are in English.

Features

• List view, calendar view (/calendar) and map view (/map), plus a prerendered per-show page at /event/[id] (own title, OG tags and schema.org Event JSON-LD, so individual shows are indexable).
• Search & filtering — keyword, city (matches any touring city), category/sub-genre, date range, custom price range, on-sale status, source; sort by date / on-sale / price. Filters collapse behind a toggle.
• Detail modal — description, organizer, running time / age, every session (date + venue + city), price, program-intro images, and a deep link to buy. Plus add-to-calendar (.ics) with an on-sale reminder, and share.
• Favourites — star shows (stored locally), filter to just those.
• Dark / light mode, PWA (installable, works offline), infinite scroll, opening-curtain animation.
• RSS — /feed.xml, plus per-source (/feed-<source>.xml) and per-genre (/feed-genre-<slug>.xml) feeds. No email, no backend; your reader handles notifications.
• Feedback — one click opens a pre-filled GitHub issue.

Sources

Each show is classified into a sub-genre from its title (戲曲 / 偶戲 / 音樂劇 / 兒童親子 / 相聲 / 喜劇 / 舞台劇) for the category filter. tixCraft was evaluated and dropped — its catalogue is concerts/tours, not theatre.

Architecture

Fully static — no database, no backend, no email.

OPENTIX · udn · KHAM · ERA · KKTIX · Accupass
        │  pnpm scrape (tsx)
        ▼
  fetch + parse + enrich ──▶ static/shows.json        ──▶ prerendered SvelteKit pages
                         ├─▶ static/descriptions.json  ──▶ lazy-loaded by the modal
                         ├─▶ static/feed*.xml          ──▶ RSS readers
                         └─▶ static/sitemap.xml        ──▶ search engines

GitHub Actions (twice daily): scrape ─▶ commit data ─▶ Vercel redeploys
  ├─ on scraper failure (a source errors or returns 0), opens an alert issue
  └─ venues that can't be resolved to a city roll up into one tracking issue

The pages read the committed shows.json at build time (prerendered with adapter-static, including every /event/[id]); favourites, calendar, map, .ics and theme are all client-side.

Getting started

Requires Node 24 and pnpm (corepack enable will provide it).

pnpm install
pnpm scrape         # all sources → static/shows.json + descriptions + feeds + sitemap (needs network)
pnpm dev            # http://localhost:5173

All commands

pnpm dev            # dev server
pnpm build          # production build (prerendered static site → build/)
pnpm preview        # serve the production build locally
pnpm check          # svelte-check (types)
pnpm lint           # prettier --check + eslint
pnpm format         # prettier --write
pnpm scrape         # full scrape of every source (re-generates all static data)
pnpm export         # re-generate shows.json + feeds + sitemap from the existing snapshot
ONSTAGE_FAST=1 pnpm scrape   # skip detail-page enrichment (faster, less complete)

See .env.example for the optional ONSTAGE_* overrides (all have defaults).

Tech stack

SvelteKit (Svelte 5 runes) · TypeScript · Tailwind CSS v4 · adapter-static · Leaflet (map) · node-html-parser · html-to-text · tsx · pnpm · ESLint + Prettier.

Data & license

Only factual fields are stored (title, dates, venue, on-sale, link); poster images are hot-linked, and ticket links go back to the official platform, which holds the copyright. Scraping is low-frequency and attributed. Code is MIT.