Skip to main content
Lovable’s web agent does not read skill files from .agents/skills/ or .cursor/skills/ — it only reads Project knowledge, Workspace knowledge, AGENTS.md, and CLAUDE.md from your GitHub repo. The npx skills add CLI has no direct effect on what Lovable sees.However, Lovable projects are GitHub-backed. If you also work on the same project locally in Cursor (a common hybrid workflow), you should install the skills there too — see the Cursor guide.

How Lovable reads context

Lovable’s agent considers several context sources when generating code:
AGENTS.md and CLAUDE.md at the repo root are always read by Lovable, even in very long sessions where project or workspace knowledge may not always be followed consistently. Use them for rules that must never be lost.
For Notifyer integrations, the most reliable approach is to put the critical API context in Project knowledge and keep the AGENTS.md file for structural rules that must persist across long sessions.
Add the following to your project’s Project settings → Knowledge. This tells Lovable’s agent exactly how to call the Notifyer API when building your app.
Project Knowledge supports up to 10,000 characters. The block above is intentionally concise. Expand it with specific endpoint shapes or payload examples if your project needs precise API call generation.

Option 2 — AGENTS.md in your GitHub repo

For rules that must survive long Lovable chat sessions, add an AGENTS.md file to the root of the GitHub repository linked to your Lovable project. Lovable always reads root-level AGENTS.md files regardless of session length. Create AGENTS.md in your repo root:

Example Lovable prompts

With the project knowledge or AGENTS.md in place, use Lovable’s Agent mode to build Notifyer-powered features:

Send a WhatsApp message from a form

Send a template when window is closed

Broadcast from the dashboard

Webhook receiver

Label and handoff from the app UI


Using Plan mode before building

For complex integrations, click Plan next to the message input to switch to Plan mode before any code is written. Plan mode never modifies your code — it only reasons and proposes an approach.
When you’re happy with the plan, approve it and Lovable automatically switches to Agent mode to implement each piece. The approved plan is saved to .lovable/plan.md in your project.

Environment variables by backend

Lovable supports two distinct backends. The env var workflow is different for each.
Lovable Cloud is Lovable’s own full-stack hosting platform — no external Supabase account needed. It handles the backend automatically.When Lovable generates an Edge Function that calls the Notifyer API, instruct it to read secrets from environment variables:
Lovable Cloud manages the secret injection automatically as part of the Cloud environment. You can review and manage Cloud settings via Connectors → App connectors → Lovable Cloud → Manage permissions.

Important constraints

These operations cannot be done via Lovable or any API — they require the Notifyer Console web UI:
  • Initial WhatsApp number connection (embedded signup / QR scan)
  • Billing and Stripe plan changes
  • Meta template approval (timing is controlled by Meta, not the API)