pnpm monorepo. Node >= 22.12 (see .nvmrc); pnpm 10.34.1.

Root scripts target apps/site. Use pnpm dev:dashboard / build:dashboard for the Next app, pnpm dev:studio for the Studio.

Moving from Next.js on Vercel to Astro 7 SSR on Cloudflare Workers, with Sanity upgraded to v6 and the video/AI pipeline (Remotion, Gemini, ElevenLabs, NotebookLM, GCS, YouTube upload) dropped. Auth moves from Supabase to better-auth on Cloudflare D1.

Branch note: dev still holds the pre-Next Astro app and main holds the Next app — main is ahead. Harvest old Astro code with git show dev:<path>, never by checking dev out over the working tree.

• output: "server" — there is no getStaticPaths anywhere. Pagination is a runtime bounds check on Astro.params, not a build-time param list.
• GROQ queries must live in .ts files. Sanity TypeGen cannot parse .astro frontmatter, so defineQuery calls in .astro files are silently skipped.
• Never read secrets from import.meta.env. Vite inlines non-PUBLIC_ vars into the server bundle, baking them into the deployed Worker script. Read them from import { env } from "cloudflare:workers" — @astrojs/cloudflare v14 reduced Runtime to { cfContext } and deprecated locals.runtime.
• Cloudflare bindings are not inherited into env.production. Anything added to the top level of wrangler.jsonc must be repeated there.
• The environment is chosen at BUILD time, via CLOUDFLARE_ENV. The adapter flattens wrangler.jsonc into dist/server/wrangler.json and writes .wrangler/deploy/config.json, which redirects wrangler away from the source config — so wrangler deploy --env production on a default build silently ships the dev dataset under the dev worker name. Use pnpm --filter @codingcatdev/site build:production.
• Astro.site is baked in at build time, so it cannot carry a per-env origin. Canonicals, og:url, feeds and the sitemap read Astro.locals.siteUrl, resolved per request from the SITE_URL var in middleware.
• stegaClean anything bound to an attribute, <meta>, JSON-LD, or XML. Stega's zero-width characters are harmless in text nodes and corrupting in href, src, and feeds.
• @sanity/icons v5 has no root-entry icon exports. Import from subpaths: import {UserIcon} from "@sanity/icons/User".
• astro-portabletext uses singular type / block / mark keys, and mark components read the mark definition from node.markDef, not node. Both fail silently — the component simply never runs.
• A self-closing <script /> in an .astro file breaks props inference, so Astro.props degrades to Record<string, any>. Use paired tags.
• apps/sanity/plugins/podcast-rss/ is vendored, not an npm dep — see its README.

Sanity guidance is vendored in .agents/skills/ (pinned by skills-lock.json). The Astro-relevant ones for this migration:

• .agents/skills/portable-text-serialization/rules/astro.md — astro-portabletext usage. Note it uses singular type / block / mark keys, unlike @portabletext/react's plural types / marks.
• .agents/skills/sanity-best-practices/references/astro.md
• .agents/skills/content-modeling-best-practices/, seo-aeo-best-practices/

.agents/skills/sanity-live-cache-components/ documents the Next.js cacheComponents + next-sanity pattern being removed. It applies only to apps/web and should be deleted with it.

Cloudflare guidance comes from the Claude Code cloudflare plugin — invoke the cloudflare, workers-best-practices, and wrangler skills rather than guessing Workers APIs.

pnpm lint, pnpm typecheck, pnpm build cover apps/site; CI also builds the Studio and fails if apps/sanity/extract.json has drifted from the schema.

Bindings only resolve under wrangler dev, not astro dev — verify anything touching locals.runtime.env, D1, or OG image generation against a real Worker.

apps/site/baseline/ holds pre-migration production sitemap.xml and RSS snapshots (472 URLs, 50 items per feed). Diff against them before cutover; a changed podcast GUID re-publishes every episode to Apple/Spotify.