What you get
Claude Code is a coding agent that works in your terminal, inside your repository. With Orwel it can do both halves of the measure step in the vibe coding loop: instrument the app correctly, and later read real usage before you plan the next change.
You need an Orwel workspace (one Pro plan with a 14-day trial, see pricing) and Claude Code running in your project folder.
Step 1: Install the Orwel skill
From the project root:
npx skills add hyperlabs-ai/hyper-skills/orwel
The skill is a set of instructions Claude Code loads when you ask about Orwel. It covers the full SDK and its rules: key in an environment variable, one wrapper module, a single client side init, catalog codes for track(), identify with an email and an awaited conversion(). It also proposes an event catalog from your service layer and verifies with Dev mode.
If you prefer project instructions over a skill, paste the CLAUDE.md analytics template into your CLAUDE.md. Claude Code loads ./CLAUDE.md or ./.claude/CLAUDE.md at the start of every session.
Step 2: Put the key in an environment variable
Copy the development key (orwel_dev_...) from your workspace and add it to .env under the name your framework expects:
| Framework | Variable |
|---|---|
| Next.js | NEXT_PUBLIC_ORWEL_KEY |
| SvelteKit, Astro | PUBLIC_ORWEL_KEY |
| Vite (React, Vue, Svelte) | VITE_ORWEL_KEY |
| Nuxt 3 | NUXT_PUBLIC_ORWEL_KEY |
You can also paste the key into the chat. The skill writes it to .env for you and never into code. In production, set the same variable to the production key (orwel_pk_...) in your hosting provider.
Step 3: Ask Claude to install and instrument
Start with the install, then the business events:
Install Orwel in this project, identify users when they sign in and record a lead when they
submit the contact form. Then propose an event catalog from the service layer and wait for my
confirmation before instrumenting it.
Claude detects the framework, installs orwel@latest, creates the wrapper, adds the init and wires your forms. When it proposes the catalog, check that each code fires after the action succeeds and that track() codes come from the catalog. Ten more prompts for signup funnels, feature usage, purchases and audits are in vibe coding prompts.
If the catalog includes custom codes, register them before shipping: add them to a journey or feature in Journeys setup or Features setup in the dashboard and save. Unknown codes are rejected.
Step 4: Verify in Dev mode
Ask Claude to run the app with the development key and debug: true and trigger one of each event. Then open Dev mode in the dashboard. Every request arrives live with its type, code, visitor and HTTP status, plus the errors and warnings found in it and a suggested fix.
Clear every error (event_not_found, invalid_field, missing_session_id) and every warning that matters (conversion_no_email, missing_attribution). If something is rejected, copy the warning from Dev mode into Claude Code. The skill knows what each one means.
When Dev mode is clean, deploy with the production key.
Step 5: Connect the Orwel MCP server
This is what turns analytics into input for the next prompt. In Orwel, open Settings, API keys and create a key with the mcp scope, set to server only. Then, in your project folder:
claude mcp add --transport http orwel https://api.orwel.io/mcp \
--header "Authorization: Bearer orwel_sk_..."
Claude Code asks you to approve the orwel server the first time it starts. The server is scoped to the workspace that owns the key and needs an active subscription or trial. The production key and the development key are not accepted here; only an API key with the mcp scope works.
Step 6: Ask Claude about real usage
Now Claude can query your workspace. Useful questions:
- "Who are my top leads this month and why do they score high?"
- "How did my campaigns perform in the last 30 days?"
- "Pull conversions and the issues Orwel detected in the last 14 days, then propose the next change."
The tools cover leads, lead analytics, visitors, conversions, campaigns, detected issues and signals, and lists. Feature adoption and journey reports are in the dashboard, so paste those numbers into the prompt when the change depends on them. This is the vibe coding feedback loop running inside Claude Code.
Use a separate API key per agent so you can revoke each one on its own, and never commit an orwel_sk_ key: it reads customer data.