Dumping the whole monorepo into chat wastes tokens and invites wrong analogies from unrelated packages. I have watched models “fix” admin auth by copying a pattern from a dead experiment folder.
Open the files that matter first. Name them in the prompt. Ask the agent to search before inventing. You are the librarian; Cursor is the scribe.
I separate concerns in the brief on purpose: public JSON content vs admin Prisma vs auth middleware. Those layers fail differently. Mixing them in one prompt creates confident nonsense.
Prefer targeted greps and @file references over “look around.” Vague exploration feels productive and burns the budget you needed for the actual edit.
When the repo grows, a short architecture note helps — entry points, content source behavior, where uploads live. Future you and the agent both benefit.
Large does not mean you need longer prompts. It means you need sharper ones.
On this portfolio codebase I also keep marketing sections and blog content out of random refactors. If the task is “fix contact SLA,” the agent does not need to rewrite blog covers. Scope is a kindness.
If search returns too much noise, narrow by folder before you ask for an edit. The cheapest bug to avoid is the one never introduced in the wrong package.
Next step
Ready for a site that books patients or wins jobs?
Dental practices and home-service contractors — tell me your services, area, and timeline. I’ll reply with a practical plan.