ezy-template-community-astro-ssr

Imported from AppifyLab/ezy-template-community-astro-ssr (AGENTS.md). Install upstream with npx skills add AppifyLab/ezy-template-community-astro-ssr. Copyright stays with the author.

Rules for AI agents working in this app

This repo renders ONE community's public Site with Astro on a Cloudflare Worker (output: 'server', @astrojs/cloudflare): /, /<slug…>, /blog, /blog/<slug> and a 404 for anything else. Every other path on the hostname (/login, /join, /feeds, /dashboard, /api, /_next, /robots.txt, /sitemap.xml…) is served by the EzyCommunity App and never reaches this Worker — link to those with a plain <a href>.

• Keep the adapter @astrojs/cloudflare (this Template pins 14.x; the platform needs 13 or later). @astrojs/node|vercel|netlify are refused by the platform.
• This Template uses npm: keep package-lock.json committed (the Build then runs npm ci) and add no second lockfile. That is the Template's choice — the platform picks npm, pnpm or yarn from the lockfile, and refuses bun.
• Commit NO wrangler config (wrangler.json[c]/.toml): the platform writes the deploy config at build time.
• NO Astro.session and NO on-demand image transforms (imageService stays 'passthrough'): the platform strips the SESSION KV and IMAGES bindings, so both fail at request time.
• Never create pages under an App path (src/pages/api/, src/pages/login.astro…) or a robots.txt / sitemap.xml (as a page or in public/) — the App owns those URLs and visitors never see them.
• This Template does ALL data access in src/lib/api.ts, called from page frontmatter at request time, so api.ts decides field by field what reaches the page. Keep it that way — never fetch from a browser <script>. (The platform itself also allows browser fetches to relative /api/public/...; server-only is this Template's choice.)
• The API base is EZY_SITE_URL from import {env} from 'cloudflare:workers' (set by the platform; .env in dev), falling back to Astro.url.origin. Astro.locals.runtime no longer exists. Relative fetch URLs fail on the server.
• The community hostname is behind Cloudflare, which 403s a request with no User-Agent: server-side fetches send one plus Accept: application/json.
• NEVER let a member email reach the page — src/lib/api.ts re-projects every API response field by field for exactly this reason; keep that shape rather than spreading raw JSON.
• Brand = name + logo + favicon from /api/public/site/v1/initial-data. That payload has no description and no colours; don't invent fields for them.
• An unknown post sets Astro.response.status = 404 and renders NotFound.
• Do not add a base or asset prefix: the site is served at / on its own origin.
• No .github/workflows/ — the site deploys from the platform, not Actions.

Platform rules

True for any code on an EzyCommunity Site, not only this Template:

• What is built is detected from the files, and the verdict is final: TanStack Start (@tanstack/react-start), Astro (static, or SSR with @astrojs/cloudflare only), or a root index.html published as-is. Next.js, Nuxt, SvelteKit, Remix, React Router (framework mode), Gatsby, Angular and un-built Vite / CRA / Eleventy source are refused. The project sits at the repository root (a zip may wrap it in one folder); a Node project has one lockfile (npm, pnpm or yarn; bun is refused). Server code runs on Cloudflare Workers (nodejs_compat), not Node.
• The App owns /api, /login, /join, /feeds, /dashboard, /settings, /_next, /robots.txt, /sitemap.xml, /<product>/<id>/checkout and more: a file or route there is never served (the dashboard lists it as a shadowed path). Link to App pages with a plain <a href>.
• Community data comes only from its public API on its own hostname (/api/public/...): server code builds the URL from EZY_SITE_URL (the only environment variable), browser code uses relative URLs. Deployed server code can reach no other host. Data fetched during the Build is frozen until the next Build (a Redeploy reuses the old output) — fetch at request time or in the browser.
• There are no secrets. No secret store exists: never put keys, tokens or passwords in code, .env or config — assume anything in the project can end up public. Never send a member's email address to the browser.
• No deploy config. Don't commit wrangler.*: the platform writes the Worker config and drops any KV, R2, D1 or other binding.