Set up with an AI agent

You are integrating FirstFlow into this repository. FirstFlow is an open-source (MIT) SDK there is NO FirstFlow server, dashboard, or npm package. It runs client-side and persists to the developer's own Supabase. Follow this exactly and do not invent APIs or options.

GROUND TRUTH do not deviate:
- Packages: "@firstflow/runtime" (React widget), "@firstflow/runtime-server" (optional Node LLM observability), "@firstflow/config-schema" (types + Supabase migrations). They are NOT on npm they are consumed from this firstflow-oss monorepo via workspace:*, and the app lives inside this workspace.
- FirstflowProvider accepts ONLY these props: user (optional), experiences, persistence (optional), onEvent (optional). There is NO apiUrl, agentId, publishableKey, conversationId, or realtime socket. Do not add them.
- Experiences are authored as firstflow/experiences.json in the repo and imported at build time. There is no dashboard.
- FirstFlow reads NO API key. If you wrap an LLM, the developer's own client supplies the key.
- The link between front-end and back-end is user.id. A conversationId / sessionId is owned by the app's own chat code and passed only to the server SDK never to the provider.

STEPS:
1. Confirm this repo is firstflow-oss: it has packages/sdk-frontend, packages/sdk-backend, packages/config-schema. If not, STOP and tell the user to clone firstflow-oss and place their app inside its workspace.
2. Detect the app's framework (Next.js vs Vite/React) from its package.json and config files. Find the app folder.
3. Wire the workspace:
   - Ensure the app's folder is covered by a glob in pnpm-workspace.yaml (add "apps/*" or the app's path if missing).
   - Add to the app's package.json dependencies:
       "@firstflow/runtime": "workspace:*",
       "@supabase/supabase-js": "^2.45.0"
     (also "@firstflow/runtime-server": "workspace:*" if the app has a backend).
   - Run "pnpm install" then "pnpm build" from the repo root.
4. Create firstflow/experiences.json in the app with this sample (skip if a file already exists):
   [
     {
       "id": "exp_welcome",
       "name": "Welcome",
       "type": "guide",
       "status": "active",
       "trigger": { "on": "page_load", "delay": 1 },
       "frequency": "once_per_user",
       "steps": [
         { "id": "hi", "type": "message", "title": "Hello", "content": "Welcome!", "cta": "Get started" },
         { "id": "done", "type": "thank_you", "title": "You're all set", "confetti": true }
       ]
     }
   ]
5. Mount the provider in a client component, importing "@firstflow/runtime/styles.css". Read env with the framework's convention Next.js: process.env.NEXT_PUBLIC_SUPABASE_URL / NEXT_PUBLIC_SUPABASE_ANON_KEY; Vite: import.meta.env.VITE_SUPABASE_URL / VITE_SUPABASE_ANON_KEY. Build the Supabase client, then:
     import "@firstflow/runtime/styles.css";
     import { FirstflowProvider, FirstflowWidget } from "@firstflow/runtime";
     import { createSupabasePersistence } from "@firstflow/runtime/supabase";
     import experiences from "<path>/firstflow/experiences.json";
     // <FirstflowProvider experiences={experiences} persistence={createSupabasePersistence(supabase)}>
     //   {children}<FirstflowWidget/>
     // </FirstflowProvider>
   Pass user={{ id, traits }} from the app's existing auth if present; otherwise omit user (the runtime uses a stable anonymous id).
6. (Only if the app makes LLM calls and wants observability) In a server route, build a service-role Supabase client (SUPABASE_SERVICE_ROLE_KEY, server-only) and wrap the LLM client:
     const ff = new Firstflow({ persistence: createSupabaseLLMPersistence(serviceClient) });
     const ai = ff.wrap(new Anthropic());   // or new OpenAI(); the key comes from YOUR client
   Tag calls with firstflow: { userId, sessionId }.
7. HUMAN-ONLY you cannot do these. Print them clearly and PAUSE for the user:
   a. Create a Supabase project at supabase.com.
   b. Open the SQL editor and run packages/config-schema/supabase/schema.sql once (one file = the whole schema).
   c. Provide env values: the Supabase project URL + anon key under the framework's client prefix, and SUPABASE_SERVICE_ROLE_KEY (server-only) if step 6 was done. Add an AI provider key only if wrapping an LLM.
8. Verify: start the app's dev server. The welcome experience should appear after ~1 second, and a row should appear in firstflow_flow_runs in Supabase.

DO NOT:
- run "pnpm add @firstflow/..." (the packages are not on npm use workspace:*).
- add apiUrl, agentId, publishableKey, or conversationId to FirstflowProvider.
- put SUPABASE_SERVICE_ROLE_KEY in client code or any NEXT_PUBLIC_/VITE_ variable.
- expect FirstFlow to read an API key the LLM client you construct holds it.