wix-headless

This skill connects Wix business services to a Wix Headless frontend. Its core job is to configure the Wix backend: infer the needed capabilities, install the Wix apps, seed the backend content, and produce an SDK-integration guide describing how to call Wix from the frontend.

What happens with the frontend depends on the project type and operation:

• Backend-only (self-managed, stripe, or a managed "just set up the backend" run) — the skill emits the SDK guide as its final output; the host owns the frontend, build, and hosting.
• Managed create / connect — the skill also owns the frontend: it scaffolds a new project (create) or attaches Wix to an existing one (connect), wires it to the backend using that same guide, builds, and releases to Wix. (There is no Designer or template library — the frontend is built ad-hoc to intent.)

On a backend-only run the skill does not own the project scaffolding. When the work needs a frontend, scaffold it according to user intent — framework and structure follow the prompt; the skill's job is the Wix backend plus the emitted SDK guide describing how to call Wix from whatever frontend exists, not choosing or generating the app. This is the norm for a stripe run: the project is provisioned via Stripe Projects (credentials land in .env — see stripe/AUTHENTICATION.md), the skill configures the backend and emits the guide, and the live site is finalized per stripe/DEPLOYMENT.md — while the frontend is scaffolded and wired to intent, by the host.

The skill behaves identically across project types except for authentication and deployment, which are isolated in a per-type folder under references/:

Everything else — Discovery, Setup, Seed, the SDK handoff — is project-type-agnostic and refers to "the provided authentication mechanism" rather than any specific method.

1. The project type — managed | self-managed | stripe — provided by the caller, else detected from project signals / user intent (see Project types § "Resolving the project type"), else asked. It selects <TYPE_DIR> (see Path resolution).
2. A Wix metasite + a headless OAuth app exist, and credentials are available — obtained via the project type's AUTHENTICATION.md. The skill needs a bearer token authorized for the metasite and the metasite id.
3. The user intent — free text describing what Wix should power ("add a store", "blog + contact form", "persist my app's data").
4. (Optional) the project on disk — read-only, to sharpen brand/capability inference.

If the credentials are absent, the Wix backend isn't reachable — stop with a clear error.

Always open DISCOVERY.md first (it's agnostic — capability/brand/intent inference + the imagery gate, no auth). Then route by project type + operation:

Backend-only (self-managed, stripe, or managed backend-only) — the lean spine:

1. Discovery (references/DISCOVERY.md) — infer capabilities + brand + intent + imagery.
2. Setup (references/SETUP.md) — install the Wix apps those capabilities need.
3. Seed (references/SEED.md) — create the backend content (+ entity images if imagery is on).
4. Handoff (references/SDK_HANDOFF.md) — after Setup and Seed, emit the integration guide: SDK bootstrap, per-capability call shapes, the seeded IDs, and the @wix/* package list.
5. Finalize deployment (<TYPE_DIR>/DEPLOYMENT.md) — run the project-type's finalize steps.

Throughout any run — if the user asks to send feedback to Wix, complains/gets frustrated, or the run hits any real friction (a confusing error, a doc gap, a tooling dead end, a workaround you had to invent — it doesn't need to repeat), offer to relay it to Wix per references/FEEDBACK.md. Default to offering rather than waiting to be asked. Every flow's finalize step (<TYPE_DIR>/DEPLOYMENT.md / managed/CREATE.md / managed/CONNECT.md) ends with the same checkpoint — treat it as a required self-check, not optional. Send only after an explicit yes — never automatically.

Managed create / connect / iterate — after Discovery, hand the whole run to the managed flow:

• create → references/managed/CREATE.md (scaffold → Setup → Seed → build the frontend → release).
• connect → references/managed/CONNECT.md (init → Setup → Seed → wire the existing UI → release).
• iterate → references/managed/CONNECT.md — the project is already connected, so it reuses the existing wix.config.json (no init). Setup / Seed / wiring are incremental: it may already be set up and seeded from a prior run, so the agent checks current state first (installed apps, already-seeded content, existing wiring) and applies only the delta the new intent needs — never blindly re-installing or re-seeding. Re-release only if the frontend build output changed. These reuse the same SETUP.md/SEED.md/SDK_HANDOFF.md, but apply the SDK guide to build/wire the frontend themselves rather than emitting it, and release via managed/DEPLOYMENT.md.

Each Wix call uses the universal call shape (SETUP.md §1) with $TOKEN/$SITE_ID obtained per <TYPE_DIR>/AUTHENTICATION.md. The skill runs non-interactively except for the one imagery question (and asking the project type if it can't be resolved).

Don't smoke-test the frontend locally unless the user asks to verify. The deliverable is the built-and-released site (managed) or the SDK guide (backend-only) — not a local test report. By default do not start a dev server (wix dev / astro dev / npm run dev) to curl pages, drive the cart / booking / login flow, or launch a headless browser: correctness comes from following the recipes, real errors surface at wix build / wix release, and a headless run can't complete an interactive login or a real payment anyway — so these loops routinely burn minutes of wall for little signal (this is distinct from, and in addition to, the "release once at the very end" rule — that one only bans extra build+release cycles, not dev-server testing). Only spin up a dev server and smoke-test when the user's prompt explicitly asks to verify / test / confirm it works — and then keep it to a single lightweight pass (pages compile and render), not a full purchase/booking/login drive. The one post-release check in <TYPE_DIR>/DEPLOYMENT.md is the sanctioned verification.

Compute <SKILL_ROOT> from this file (<SKILL_ROOT>/SKILL.md — strip /SKILL.md); hold the absolute path in scratch. Then resolve the project type to its folder and hold it too:

<TYPE_DIR> = <SKILL_ROOT>/references/<projectType>/ — one of managed/, self-managed/, stripe/.

Start a run by opening DISCOVERY.md. The flow files (CAPABILITIES, DISCOVERY, SETUP, SEED, SDK_HANDOFF, IMAGE_GENERATION, AI_FEATURES) are project-type-agnostic; the per-type specifics live under <TYPE_DIR>/ (AUTHENTICATION.md, DEPLOYMENT.md, and — managed only — CREATE.md/CONNECT.md).

This skill has no skill upstream — the how is read from the live Wix docs at dev.wix.com/docs. Read-priority: a page this skill links is read by curl-ing its .md twin directly (first priority — don't re-discover a curated link with search); the Wix MCP doc/search tools are second priority, for finding pages the skill doesn't link or as a fallback if a fetch fails. (Append .md to any docs URL for raw markdown; menu pages list child links, content pages carry the schema.)

Doc discovery is the shared fallback for both tracks — never the first move. Each track's primary source is its pinned material (below); when that doesn't cover what you need, fall back to references/DOC_DISCOVERY.md — a semantic doc-search + schema lookup that works with or without the Wix MCP.

• Seed reads each capability's create flow from its inline recipe — a self-contained local inline-recipes/setup-*.md (mapped per capability in SEED.md § "What to seed per capability") that inlines the calls and supersedes the REST doc pages, so read it and seed from it alone. Only a capability with no inline recipe (e.g. coupons) falls back to doc discovery (DOC_DISCOVERY.md).
• Handoff links the SDK docs for each capability's API shape, and supplies the runtime package set from the inlined map in SDK_HANDOFF.md (the SDK .md pages don't expose @wix/* import strings to navigation, so packages are mapped, not navigated).

Setup carries its app-install call (and the appDefId constants) inline in SETUP.md; CAPABILITIES.md is the vertical index that lets Discovery match intent and declares, per built vertical, the Required site features + Implementation checklist that Seed enables (backend-backed features) and the Handoff carries into the guide (so the host builds a complete site, not a bare data dump). This skill carries the what (which capabilities, how much content, what a finished site includes) and reads the how off the docs.