Agentic Architect: .NET Cursor Rules Kit

Cursor is brilliant at .NET. Until it isn't.

You're on Pro, you've configured it, and the first week feels like a cheat code. Then it starts throwing exceptions where you've used Result<T> for six months. It injects DbContext into a singleton. It writes a read-only EF query with no AsNoTracking. You spend 20 minutes in review undoing what it confidently wrote.

That's not a Cursor bug, it's a context gap. Cursor doesn't know your architectural decisions, so it defaults to the average .NET code on the internet.

This kit closes the gap. It's the system I run on production C#/.NET: 9 scoped .mdc rule files, ADR templates, and a Learning Log the model re-reads at the start of every session. Setup takes about 2 minutes and works whether you use Rider or Visual Studio alongside Cursor.

What's inside:

 rules/

  result-pattern.mdc   Result<T> over throw for business failures

  di-scoping.mdc     Never capture Scoped into Singleton

  ef-core-reads.mdc   AsNoTracking on every read-only IQueryable

  clean-architecture.mdc Layer boundaries: no EF in the domain

  mediatr-pipeline.mdc  Command/handler/validator standards

  async-patterns.mdc   CancellationToken propagation

  minimal-api.mdc    Minimal API + Swagger doc standards

  xunit-conventions.mdc Test naming + fixture patterns

  mcp-tools.mdc     MCP server wiring patterns

 templates/

  LEARNING_LOG.md    Seed file Cursor re-reads on session start

  ADR-template.md    Architecture Decision Record starter

  ADR-examples/     3 worked examples

 prompts/

  hydrate-learning-log.md Prompt: scan codebase, propose ADRs

 QUICKSTART.md      5-minute setup, any .NET solution

Who it's for:

 Senior C#/.NET developers using Cursor Pro

 Teams on Clean Architecture / DDD / MediatR

 Anyone who's fixed Cursor's "confidently wrong" pattern in review

Not for:

 Beginners still learning C# (these rules enforce patterns you should already understand)

 JS/TS-only developers (this is .NET-specific)

The maths:

Senior .NET devs bill at £60-120/hour. Cursor regressions cost a conservative 2 hours per sprint on an active codebase. That's £120-240 of your time. This kit is £9. It pays for itself the first time it catches a DbContext-in-singleton before it ships.

One-time purchase. MIT licensed. No subscription.

v1.0 launch price: £9. When the 30-minute video walkthrough ships (Q3 2026) the bundle goes to £29. Buy now and the video lands as a free update.

30-day no-questions refund. If the rules don't stop at least one regression on your codebase in 30 days, email me and I'll refund every penny. Keep the files.