`.
See `references/client-islands.md`.
4. **Add middleware** (logger, CORS, cache, secureHeaders, etc.) with `app.use()`. See
`references/core-routing-middleware.md`.
5. **Verify**: `c.html` returns 200 with real markup (curl it), island bundle loads and
hydrates (check the DOM updates), no `hono/jsx` import in client code and no
`hono/jsx/dom` in server code.
## Converting a React SPA → Hono (the big one)
Read `references/react-spa-to-hono-migration.md`. The short version:
- **Pages → SSR JSX.** Each React Router route becomes a Hono route returning `c.html(
)`.
Static content (text, layout, data-driven markup) renders on the server — instant, SEO-friendly,
great Lighthouse.
- **Interactive components → islands.** `react-leaflet` map, `d3` force graph, modals, anything
with `useState`/event handlers → a `hono/jsx/dom` island OR keep the underlying lib
(Leaflet/D3 are framework-agnostic) and mount it from a client entry. D3 and Leaflet don't
need React at all.
- **Icons:** `lucide-react` is React-specific. On the server use inline SVG (or `lucide`'s
framework-agnostic SVG strings); don't import `lucide-react` into `hono/jsx`.
- **Router:** delete `react-router`; routing is server-side in Hono. Client nav can be plain
`
` (full SSR navigation) or a tiny island if you want SPA-like transitions.
- **Data modules** (plain `.ts` exports) port over unchanged — import them into server pages.
- **Don't big-bang it blindly.** Keep the build green at every step; convert page-by-page and
curl each route. A working SSR page with a missing island beats a broken全-rewrite.
## 🛑 "Using Hono" means SSR — a Hono ROUTER in front of static HTML is NOT the default (2026-08-31)
**The most likely way to think you built a Hono site and not have one:** `new Hono()` for the
API, `secureHeaders()`, typed `Bindings` — all correct — and then the actual page is a
hand-written `public/index.html` served through the `ASSETS` binding. Every Hono symbol is
present. Zero HTML is server-rendered. That is **Hono-as-router**, and it satisfies none of the
reasons the Hono default exists (real markup to crawlers/curl/Wayback, no CSR shell, FCP).
It passes every check you would naively run: `tsc` clean, `wrangler deploy --dry-run` rc=0,
`import { Hono } from "hono"` right there at the top of the file.
**The one-command test — run it before claiming a site is Hono:**
```bash
grep -rn 'c\.html(' src/ | grep -v node_modules # ZERO hits => you have a router, not SSR
grep -rln 'hono/jsx' src/ # ZERO files => no server JSX exists
```
Any page a human reads must reach the browser via `c.html()`. `ASSETS.fetch` is for
**assets** — `.js`, `.css`, images, fonts — never for the document.
**Reference incident (2026-08-31, improvecortland).** Built the Worker in Hono, wrote the portal
as `public/index.html`, and reported the site as built on Hono. The user asked *"did you build
this on hono framework btw like /carmack says to"* — the honest answer was *partially*. The
conversion afterwards was ~30 minutes and had one non-obvious constraint worth stealing:
**Converting static-HTML → SSR when tests execute the shipped page.** The harnesses regex-
extracted JS out of `public/index.html` and ran it, so deleting that file would have broken
them. The fix is better than the original on both axes: move the executable blocks into a real
`public/foil-core.js`, `