Luzmo Agent Skills: AI assistants that know Luzmo

2:00 min

Medium

AI coding assistants such as Claude Code, Cursor, Codex, Windsurf, Google Antigravity, and GitHub Copilot can help you create and maintain Luzmo integrations. Luzmo Agent Skills give supported assistants Luzmo-specific instructions, best practices, and implementation patterns, working alongside the public API and developer documentation rather than replacing them.

In this article, you will learn:

What Agent Skills are and how to install them
How to prompt an AI coding assistant effectively for Luzmo work
Worked examples: building an embed, troubleshooting one, and prototyping a chart
What context to give your assistant, and what never to give it

Installing Luzmo Agent Skills

Install the skills interactively with:

npx @luzmo/agent-skills@alpha

Once installed, describe the outcome you want and provide the relevant context. A useful prompt is specific about the framework, authentication method, and expected behavior.

Example: build a multi-tenant customer dashboard

Build a React page using the Luzmo embed package. Retrieve the embed authorization from POST /api/luzmo-authorization, pass the returned key and token to the dashboard component, and display a useful error when authorization fails. The backend must scope each request to the signed-in customer's tenant ID. Do not expose the master API credentials in the browser.

The assistant can generate the frontend component, the server-side authorization endpoint, and a test plan. Verify afterward that two test users from different tenants cannot access each other's data. This assumes a dashboard already exists in the Luzmo platform — the assistant builds the embed, not the dashboard itself.

Example: troubleshoot a failing embed

This dashboard works in the Luzmo application but returns a permission error when embedded. Here is the sanitized authorization request, the browser error, and the dashboard ID. Compare the requested access rights with the dashboard access configuration and list the most likely causes.

Giving the assistant a specific environment, expected result, and sanitized evidence gets a far more useful answer than "why doesn't my dashboard work."

Example: prototype a chart with Flex

Create a Luzmo Flex column chart showing monthly recurring revenue by month, split by customer segment. Add a date filter, loading state, responsive sizing, and a currency formatter. Use placeholder dataset and column IDs and label where I need to replace them.

The result is a starting point you can connect to the correct dataset and validate against real data.

Give your assistant the right context

An AI assistant produces better results when it knows:

Which Luzmo feature you're using
Your frontend and backend frameworks
Whether the experience is embedded and multi-tenant
The relevant dashboard, dataset, or chart configuration
The expected result and the current behavior
Any error messages or sanitized request/response payloads

Never paste production secrets, API tokens, personal data, or unrestricted customer data into a prompt. Keep API credentials on the server and use short-lived embed authorization for frontend experiences.

Treat generated code as a starting point: review it, test it in a non-production environment, and confirm behavior against the Luzmo developer documentation.