Grok long run

---
name: grok-long-run
description: Brief a long Cursor run on Grok 4.7 — outcome, slice, self-check, stop condition. Use when the user picks Grok, asks for a long coding or knowledge job, or says to use the strongest long-run model. Stay inside Cursor; do not write a Grok or ChatGPT product.
icon: sparkle
color: orange
---

# Grok long run

Grok 4.7 is Cursor’s long-run coding and knowledge model. It is a picker choice. It is not a second product, and it is not permission to skip a first slice.

## When to pick it

- The job will take many steps (a migration, a stubborn test suite, a research-then-patch pass)
- You need the model to check its own work before it declares done
- The user named Grok, Grok 4.7, or “the long model”

Stay on the repo default for a one-screen copy edit.

## Brief

1. Outcome a user can see
2. First slice that could ship alone
3. How you will verify (command, URL, empty/loading/error)
4. Stop condition you could check without the transcript
5. What not to expand into

Pair with `/goal` if the objective is not one file. Pin a playbook Custom Mode if there is a procedure (polish, security pass, project coordinator).

## While it runs

- Check your own diffs before you say done
- Do not narrate model news or SpaceXAI context unless the user asked
- Do not write a parallel kit, prompt pack, or restyle for Grok Bot / ChatGPT / Claude

## Done when

The slice is verifiable and you stopped. A longer context window is not a reason to keep going.