Extend .mdc rule frontmatter with a model: field for cost-efficient agentic workflows

Feature request for product/service

Cursor IDE

Describe the request

Add a model: field to .cursor/rules .mdc frontmatter so that individual rules can declare which model should be used when executing the tasks they govern:

---
description: Summarise notes files and write index entries
alwaysApply: false
model: claude-3-5-haiku-20241022
---

Cursor’s own fast / default aliases would work equally well as values, keeping it model-agnostic.

Use case: note vault indexing

I use Cursor to manage a ~650-file Obsidian markdown vault. A set of .mdc rules tell the AI how to index notes — reading each file and writing 2-3 sentence summaries to index files. This is pure summarisation: read → summarise → write. No architectural reasoning required.

When the agent runs a re-index pass it uses the default (full) model. Haiku is entirely sufficient for this and costs ~20× less. The only current workaround is manually switching models in the UI before starting - friction that’s easy to forget and impossible to enforce in a shared or automated context.

Why it matters beyond this use case

Most agentic workflows have a natural split between tasks that need deep reasoning and tasks that are bulk, low-complexity work (summarisation, formatting, extraction, boilerplate generation). Rules are already the mechanism for scoping behaviour - letting a rule declare its preferred model would make it possible to build cost-efficient, self-documenting agent workflows without manual model switching.

A max_tokens: field would be a natural companion for the same reason.

Hey @user544!

Curious if subagents would be the right fit for this! Have you tried them out?