Cursor is not magic. It is leverage, and leverage only works when you apply constraints.
I published timbenniks-cursor-rules after the novelty wore off. What I wanted was not a folder of clever prompts. I wanted the agent to behave like the senior engineer I actually want next to me on a SaaS project: plan first, small diffs, root-cause debugging, accessibility before it ships.
The longer argument is in Cursor, rules, and my vibe engineer workflow. This page is the artifact.
What you copy
A .cursor folder with modular rules and slash commands. Copy it into a project and Cursor picks up rules/*/RULE.md and commands/*.md. Or copy AGENTS.md if you want one file.
Rules are scoped by concern:
- nextjs-stack: Next.js 16 App Router, TypeScript, server components, Tailwind v4
- ui-a11y: shadcn/ui, semantic HTML, keyboard and ARIA
- backend-security: validation, secrets, contained side effects
- quality-bar: DRY, no
any, build/lint/typecheck must pass
Commands map to how I work: /plan, /implement, /dod, /prepare-pr, /debug, /refactor, /cleanup, /audit-a11y, /deps.
How to customize without breaking it
Add project-specific rules. Add commands. Change globs. Do not remove the quality bar, weaken accessibility, or add rules that fight the existing ones. If you are not on Next, replace nextjs-stack and keep the rest.
Contentstack Vibe Docs are optional. Add the repo under Cursor Settings → Indexing & Docs if the project actually talks to Contentstack. They are not bundled, because not every repo needs them.
Fork it. Delete what is not yours. Keep the constraints that make the agent trustworthy at speed.
More projects
Open Loops
A local-first macOS tracker for work you hand to coding agents. SQLite on your machine, git-aware tasks, and a full backlog API over MCP.
Open source · macOS · MIT
MCP · Agent skillsContentstack MCP Hub
A hub of agent skills with RBAC, so an agent acting for a user can only reach what that user is allowed to touch, and product managers can propose the next skill without writing it.
Active · Internal + open contributions
MCP · Local-firstTurbo Relay
A local-first intelligence layer for sanctioned AI agents. Agents write evidence-backed knowledge through MCP; humans review, approve, and promote it.
Open source · macOS · MCP