zmoki-astro-starter — Astro 启动器模板

SHA-256: 832473510f69db9df634f8123b4238a909bd146172e8f49a21300bdb63b5a797

2026-10-04T23:40:24.062Z

# zmoki-astro-starter

An Astro starter for personal and project websites — posts, resources, and legal pages — built to be **maintained through an AI coding agent**. The technical SEO is done and guarded by CI, every vendor integration is swappable, and the brand guideline lives inside the site. Built from [zmoki.xyz](https://zmoki.xyz) and genericized into a reusable template.

**Live demo: [starter.zmoki.xyz](https://starter.zmoki.xyz/)**

[![Security Headers](https://img.shields.io/badge/security%20headers-A-brightgreen)](https://securityheaders.com/?q=https%3A%2F%2Fstarter.zmoki.xyz%2F&followRedirects=on)
[![Lighthouse](https://img.shields.io/badge/lighthouse-100%2F100%2F100%2F100-brightgreen)](https://starter.zmoki.xyz/)

## Highlights

- **Astro 7** + **Tailwind CSS 4** + **MDX**, fully static output — deploys to any static host
- **AI-ready**: [`AGENTS.md`](./AGENTS.md) is a canonical technical spec any coding agent reads, plus eight task playbooks in `.claude/skills/`
- **Technical SEO built in and guarded by CI**: sitemap (with images), canonicals, robots on two layers, schema.org JSON-LD, build-time OG images, Google Discover-ready covers
- **No vendor lock-in**: deploy target, analytics, forms, captcha, and image CDN are all swappable behind clean seams
- **A living brand system** at `/-/astro/brand/` — color, typography, components, forms, voice & tone — rendered from the same design tokens the site uses
- **Perfect Lighthouse scores** (100 in all four categories, desktop and mobile) with Lighthouse CI to keep them

## Working with an AI agent

`CLAUDE.md` imports `AGENTS.md`, and Cursor or any other tool reads `AGENTS.md` directly, so any coding agent understands the codebase from the first prompt. The skills in `.claude/skills/` are vendor-neutral Markdown playbooks for the recurring tasks. Conversations look like this:

```
> Re-skin the site to a green accent
  Claude reads the /brand skill, swaps zmoki-indigo → zmoki-emerald across
  src/, verifies WCAG AA contrast on every affected pair, rebuilds, and shows
  you /-/astro/brand/color/ to review.

> Redirect /my-old-post to /blog/new-post
  Claude reads the /redirects skill, adds a row to src/redirects/blog.csv,
  runs npm run build:redirects, and commits the regenerated host artifact.

> Add Plausible analytics alongside PostHog
  Claude reads the /analytics skill, writes one self-contained provider
  component that chains onto window.track, registers it in Analytics.astro,
  and adds the script host to the CSP.

> Make my post about pricing eligible for rich results
  Claude reads the /structured-data skill and adds schema.org JSON-LD per
  Google's docs; npm run check:sd validates it in CI on every build.

> Move the site from Cloudflare to Netlify
  Claude changes platform.deploy in src/site.config.ts, rebuilds the
  redirect and header artifacts, and deletes the stale ones.
```

### Skills (task playbooks)

Each skill is plain Markdown in `.claude/skills/<name>/` — Claude Code loads them as `/name`, other tools read the files directly:

| Skill              | What it covers                                                         |
| ------------------ | ---------------------------------------------------------------------- |
| `/run`             | Launch and verify the dev server                                       |
| `/brand`           | Brand guidelines end to end: colors, typography, voice, brand pages    |
| `/redirects`       | Add or edit URL redirects (compiled per host)                          |
| `/og-images`       | Customize the social-share image cards                                 |
| `/analytics`       | Enable, add, or swap analytics providers; add tracked events           |
| `/structured-data` | schema.org JSON-LD for Google rich results                             |
| `/images`          | Content images: remote originals, optimization, post covers, image SEO |
| `/update-deps`     | Update npm packages and GitHub Actions in staged, verified commits     |

## Features

### Technical SEO, guarded by CI

The invisible work search engines check is wired in, and CI turns regressions into red builds:

- **Sitemap with images** — every indexable page is registered via one list (`src/lib/page-collections.ts`); post covers ship in the image sitemap (`<image:loc>`). CI (`npm run check:sitemap`) fails if an indexable page is missing, a noindex page is advertised, or an entry goes stale.
- **Canonical URLs** — each page declares its own canonical; a cross-posted article can point at the original via frontmatter. Canonical, OG, and structured-data URLs all come from one helper (`src/lib/urls.ts`), so they can't drift apart.
- **Robots on both layers** — per-page `<meta name="robots">` from content frontmatter, plus path-level `X-Robots-Tag` response headers. The sitemap honors both.
- **Structured data** — schema.org JSON-LD following Google's guidelines (BlogPosting, Organization, ImageObject licenses); `npm run check:sd` validates every block in the built site.
- **OG images at build time** — purpose-built 1200×630 social cards rendered by Satori + resvg during `astro build`. No browser, no runtime service, nothing committed. Cards use the site's own design tokens and brand mark, so a re-skin recolors them automatically.
- **Google Discover-ready** — posts can carry a real `cover` photo (≥1200px) that becomes the hero, the `BlogPosting.image`, and the image-sitemap entry; `max-image-preview:large` is set site-wide.
- **Link integrity** — `npm run check:links` fails the build on broken internal links or hard-coded self-origin URLs.
- **Version meta tag** — every page carries the commit hash it was built from, so a crawl maps to an exact version.

### Platform-agnostic integrations

Every integration point ships with a working default and a documented seam to swap it:

- **Deploy target** — Cloudflare Pages (default), Netlify, Vercel, or AWS Amplify. Redirects (`src/redirects/*.csv`) and response headers (`src/headers/headers.config.ts`) are authored once, platform-neutral, and compiled to the host's format from a single `site.config.ts` setting. The artifacts are committed and drift-checked in CI.
- **Analytics** — call sites fire `window.track(event, props)` / `window.identify(id, props)`, never a vendor SDK. Providers stack: PostHog and Google Tag Manager are built in; adding another is one component. Off by default (`PUBLIC_ANALYTICS_ENABLED`), so local dev and Lighthouse runs stay clean.
- **Forms** — the resource lead-magnet gate posts through one dispatcher (`Form.astro`); Brevo is built in, and the backend handles captcha validation, contact storage, and asset delivery.
- **Captcha** — single-select and self-gating (`Captcha.astro`); Cloudflare Turnstile is built in, off by default (`PUBLIC_CAPTCHA_ENABLED`).
- **Image CDN** — host content-image originals on any bucket/CDN (R2, S3, …) behind a custom domain; Astro downloads and optimizes them at build. Decoupled from the deploy host, and the same setting drives the CSP `img-src`.

### Content

Three content collections (Content Layer API, zod-validated schemas in `src/content.config.ts`):

- **`blog`** — posts with per-post authors (byline + bio + schema.org author), prev/next navigation, optional cover photo
- **`resources`** — downloadable resources and external links; a resource with a `form` + `asset` becomes an email-gated lead magnet with a thank-you page fallback download
- **`legal`** — privacy and terms placeholders

The Markdown pipeline adds anchor links on definition terms, styled external links, and copy buttons on code blocks. RSS is included.

### Brand system

An internal, noindex style guide at `/-/astro/brand/` covering color, typography, forms, components, voice & tone, and the OG cards. It renders from `src/design-tokens.mjs` — the same token file Tailwind uses — so changing a token moves the site and the guideline together. All of Tailwind's palettes are pre-generated under the `zmoki-` prefix, and custom scales (like the shipped `zmoki-cream`) slot in beside them.

Fonts are self-hosted via Astro's Fonts API (downloaded and subset at build, served same-origin, zero-CLS fallbacks) with the family names in the same token file.

### Performance & security

- Lighthouse 100/100/100/100 on desktop and mobile, asserted by a Lighthouse CI workflow on every push and PR
- Security headers (CSP, HSTS, `X-Frame-Options`, …) authored in `src/headers/headers.config.ts`, earning an A on securityheaders.com
- Fully static output — no server, no runtime dependencies

## Create a site from this template

**With `npm create astro`** — scaffold the files into a new folder:

```bash
npm create astro@latest my-new-site -- --template Zmoki/zmoki-astro-starter
```

**With GitHub** — click **Use this template → Create a new repository** on the [repo page](https://github.com/Zmoki/zmoki-astro-starter). Best when you want a hosted repo from the start (e.g. to wire up push-to-deploy).

**By cloning** — grab it and detach the history:

```bash
git clone https://github.com/Zmoki/zmoki-astro-starter.git my-new-site
cd my-new-site && rm -rf .git && git init
```

Then:

```bash
cd my-new-site
npm install
npm run dev   # http://localhost:4321
```

…and follow **[SETUP.md](./SETUP.md)**: edit `src/site.config.ts`, replace the brand mark, drop in your content, re-skin the palette. Or open your AI agent and ask it to do the checklist with you.

## Scripts

- `npm run dev` — dev server
- `npm run build` — production build (also compiles redirects/headers and emits OG images to `dist/og/`)
- `npm run preview` — preview the built site
- `npm run check` — type check (`astro check`)
- `npm run lint` — ESLint
- `npm run format` / `format:check` — Prettier
- `npm run favicons` — regenerate the favicon set from `public/brand-mark.svg`
- `npm run build:redirects` / `check:redirects` — compile redirect CSVs to the host artifact / CI drift guard
- `npm run build:headers` / `check:headers` — compile header rules to the host artifact / CI drift guard
- `npm run check:sd` / `check:links` / `check:sitemap` — post-build guards: structured data, internal links, sitemap coverage
- `npm run lhci:mobile` / `lhci:desktop` — Lighthouse CI
- `npm run timeline:blog` — generate `blog-timeline.csv`

## Project structure

```
src/
├── site.config.ts     # everything site-specific (edit this first)
├── design-tokens.mjs  # color palette + font families (re-skin here)
├── content.config.ts  # content collection schemas
├── components/        # reusable Astro components (+ analytics/, forms/, captcha/)
├── content/           # blog/, resources/, legal/
├── headers/           # platform-neutral response-header rules
├── images/            # committed content images
├── layouts/           # Base, Post, Resource, Legal, Brand layouts
├── lib/               # URLs, page registry, robots, dates
├── og/                # OG card templates, manifest, theme, fonts
├── pages/             # routes (incl. /-/astro/brand/ design system)
├── redirects/         # platform-neutral redirect CSVs
└── styles/            # Tailwind entry point
```

## CI

`.github/workflows/ci.yml` runs on every push and PR to `main`: format check, type check, lint, redirect/header drift checks, build, then the post-build guards (structured data, internal links, sitemap). A separate `lighthouse.yml` workflow asserts the Lighthouse scores. No secrets are required — provider config is committed constants, and analytics/captcha are enabled only for production builds of `main`.

## Docs

- **[SETUP.md](./SETUP.md)** — the checklist for spinning up a new site
- **[AGENTS.md](./AGENTS.md)** — the full technical spec (read by AI tools, useful for humans too)
- **`.claude/skills/`** — per-task playbooks (redirects, brand, analytics, OG images, …)