#Quick Start

The official Quickstart path is: sign in and open a folder → ask Agent to explain the repo → make one small change → review the diff and run checks → switch to Plan Mode for bigger work. The prompts below are ready to paste.

Open Agent with Cmd+I on macOS or Ctrl+I on Windows / Linux (sidepane).

#1. Explain the codebase

With the workspace root open, send this as-is:

Explain this codebase. Point me to the main entry points, key modules,
and anything I should read before making changes.

Agent searches the repo, reads relevant files, and summarizes how it fits together. That is the fastest way to orient in an unfamiliar codebase. If the answer is vague, follow up:

List the commands in package.json / Makefile / pyproject.toml that
run tests, lint, and the local dev server. Quote the exact scripts.

#2. Make one small change

Ask for options and choose—do not say “just improve it”:

Suggest three small, safe improvements in this codebase.
Explain the tradeoffs and wait for me to choose one.

Good first tasks: copy tweaks, small UI fixes, a missing aria-label, or a README line that matches real start commands. After you pick:

Implement option 2 only. Do not refactor unrelated files.
When you finish, show the diff and tell me how to verify it.

If you already know the change, describe the result, not the keystrokes:

In the README, add a "Windows (PowerShell)" section that shows how
to create a venv and run the existing test script. Match the repo's
current commands; do not invent new npm scripts.

#3. Review the diff and verify

When Agent finishes, the sidepane / editor shows a diff. Do three things yourself:

1. Read the diff: did it stay inside the files you approved? Did it “clean up” unrelated modules?
2. Ask it to run checks the project already uses (tests, types, lint, local build). Do not let it invent a toolchain:

Run the checks this project already uses (tests, typecheck, lint, or
the local build). Paste the exact commands you ran and their exit codes.

3. Run them again yourself (PowerShell examples—adapt to the repo):

# Node example
npm test

# Python example
python -m pytest

# macOS / Linux
npm test
# or
python -m pytest

Official docs also cover a deeper review workflow. For a first pass, being able to repeat the commands and exit codes is enough.

#4. Use Plan Mode for bigger changes

When the task spans many files, needs research, or must be approved before code, press Shift+Tab in the agent input to toggle Plan Mode. Officially, Cursor will:

1. Research the codebase for relevant files
2. Ask clarifying questions
3. Write a detailed implementation plan
4. Wait for your approval before building

A solid opener:

We need to add a /health JSON endpoint that returns { "ok": true }
without changing authentication. Research the existing HTTP stack,
ask me anything unclear, and produce a plan. Do not edit files yet.

Details are in Plan Mode. If the build drifts: revert files with a checkpoint or Git, refine the plan, and build again—usually faster than stacking prompts on a wrong implementation.

#A 20-minute drill

flowchart TD
  A[Open folder] --> B[Explain repo]
  B --> C[Pick one small change]
  C --> D[Review diff]
  D --> E[Run existing checks]
  E --> F{Need many files?}
  F -->|No| G[Commit or discard]
  F -->|Yes| H[Shift+Tab Plan Mode]
  H --> I[Approve, then build]

#Prompt habits you can use now

For queueing work while Agent runs, or long-lived objectives, see the queue and /goal in Agent.

#Next

• Agent
• Tab and Inline Edit
• Plan Mode