diff --git a/openspec/changes/add-mobile-h5/design.md b/openspec/changes/add-mobile-h5/design.md
new file mode 100644
index 0000000..0abefae
--- /dev/null
+++ b/openspec/changes/add-mobile-h5/design.md
@@ -0,0 +1,30 @@
+# Design: mobile H5 (add-mobile-h5)
+
+## Decision: Nuxt 3 H5 now, Taro deferred
+
+Chosen: a first-party Nuxt 3 H5 app (`apps/mobile`) reusing `@vmall/shared`,
+`theme.css`, and `@vmall/ui`.
+
+Considered and deferred: **Taro (Vue 3) for H5 + WeChat mini-program from one
+codebase**. Taro is only worth its cost when the mini-program is a committed
+target: `@vmall/ui` is DOM-based and does not run in a mini-program, Tailwind
+needs plugin+rpx workarounds there, `createApi` needs a transport adapter
+(`fetch` does not exist in MP), and MP login needs a wx.login → openid
+backend endpoint. For H5 alone, Nuxt is strictly better. Deferring Taro later
+loses little: it would share neither the UI kit nor page code with the Nuxt
+app anyway.
+
+## If the mini-program becomes a target later
+
+Split into three changes instead of one big-bang:
+
+1. `add-mobile-taro-scaffold` — Taro 4 + Vue 3 workspace app, `createApi`
+   transport injection (fetch on H5, `Taro.request` on MP), tab-bar shell,
+   account+password login (works on both, no wx APIs). Acts as the
+   compatibility spike (Taro 4 + pnpm monorepo + CSS variables in WXSS).
+2. `add-mobile-h5-flows` — the page inventory from this change, delivered on
+   the Taro H5 build.
+3. `add-mobile-miniprogram` — MP build target, wx.login session exchange
+   endpoint, share/subscribe messages. Payments can ride the existing wallet
+   (balance payment) to avoid `wx.requestPayment` merchant onboarding; top-up
+   stays on H5/PC.