AI Agent Guide
How to Use Cursor with VerdantStack
VerdantStack starters ship with AGENTS.md and generated API references — so Cursor reads the architecture, follows your patterns, and starts productive from the first edit.
What VerdantStack gives Cursor
Every starter includes context files that Cursor reads automatically:
AGENTS.md
General context — architecture, commands, rules. Cursor reads it automatically.
TypeDoc API ref
Generated spec — point Cursor at it for accurate module-level edits.
Live demo
Click through the running app to see the patterns in action before asking Cursor to extend them.
Step by step
1. Open the starter in Cursor
Clone the starter, open the folder in Cursor. Cursor reads AGENTS.md at the repo root automatically — this is the general-purpose context file that tells the AI the architecture, the commands, and the rules.
2. Cursor reads AGENTS.md on first interaction
When you first open Cursor's AI pane (Cmd+L or Ctrl+L), Cursor scans AGENTS.md and loads the context. It now knows the file map, the tech stack (SvelteKit, Drizzle, Postgres/SQLite), and the patterns the starter uses. You don't need to explain the project — it's already loaded.
3. Cursor reads the starter's context files
The starters ship AGENTS.md (the general-purpose context file, read automatically by Cursor and every major AI coding tool). AGENTS.md encodes the patterns: "use Drizzle for all DB queries, never raw SQL", "all server-side code goes in src/lib/server/", "RBAC is enforced in hooks.server.ts". This prevents Cursor from inventing parallel patterns. You can add a .cursorrules file yourself for tool-specific constraints.
4. Ask for features in the Composer
Cursor's Composer (Cmd+I) lets you describe a feature in natural language and get a multi-file change. Because it knows the architecture, you can say: "Add a password reset flow with email verification" and it will use the existing session model, the D1 database, and the auth patterns.
5. Verify with the test suite
After Cursor generates code, run the test suite. The 259–314 tests are your safety net. If Cursor broke an invariant, the tests catch it. Open the terminal in Cursor (Ctrl+`) and run: npm test.
Example prompts
"Add an invite flow where an admin can invite a member by email, the member gets a single-use link, and clicking it adds them to the org"
Cursor will use the existing service layer, RBAC system, and audit log — the same patterns the starter uses for org membership.
"Add rate limiting to the login endpoint — 5 attempts per minute per IP"
Cursor will use the existing rate-limiting pattern (if present) or add one that follows the starter's conventions — not a standalone middleware that conflicts.
"Write tests for the invite flow following the existing test patterns"
Cursor will follow the existing vitest conventions, use real DB tests, and match the HTTP-level flow patterns in the test suite.
Frequently asked questions
Does Cursor work with SvelteKit?
Yes. Cursor has strong SvelteKit support. The VerdantStack starters add AGENTS.md to give it the specific patterns (Drizzle ORM, scrypt + hashed-session auth, RBAC middleware) so it follows them instead of inventing new ones.
Does Cursor need any extra config files?
No. AGENTS.md is the general-purpose context file — all AI tools read it (Claude Code, Cursor, Copilot, Replit). The starters ship it at the repo root. If you want Cursor-only constraints, add a .cursorrules file yourself — Cursor will read it alongside AGENTS.md.
Will Cursor try to replace the auth system?
Not if AGENTS.md says "do not modify the auth model." Cursor respects the repository context files. Without them, Cursor may suggest "improvements" that break the session model. The context file prevents that.
Can I use Cursor to add features to the starter?
Yes — that's the primary use case. The starter gives Cursor a known, tested foundation. You add features on top. The tests verify nothing broke. The audit log tracks what changed. The RBAC system controls who can do what.
How does this compare to Claude Code?
Both work with the same starters. Cursor and Claude Code both read AGENTS.md. The starters ship the context file plus the API reference. The underlying code, tests, and architecture are identical.
Try it
Every VerdantStack starter includes Cursor support out of the box:
Related reading
Get in touch
Questions about the product, team licenses, or anything else? We'll respond within 48 hours.