Cursor is a code editor built around AI. Its Agent can search your codebase, edit multiple files, and run terminal commands. These prompts use its rules files, its modes (Agent, Ask, Plan, and Debug), and @-mentions.
Copy the prompt, paste it into the tool's chat, and replace everything in [BRACKETS] with your own details.
1 Set Up Project Rules
What it's for: Giving Cursor's Agent standing instructions so you don't repeat yourself every chat.
When to use it: You're starting a project, or the Agent keeps ignoring your stack or style.
Create project rules for this codebase.
Tech stack: [TECH STACK]
Rules for every chat: [e.g. "ask before adding dependencies", "keep components under 150 lines"]
Rules for specific files: [e.g. "for files in /app/api: always check the user is logged in"]
Please:
1. Put rules that apply everywhere in an "Always Apply" rule.
2. Put file-specific rules in separate rules scoped to those file patterns.
3. Save them as .mdc files in .cursor/rules (or a simple AGENTS.md if one file is enough).
4. Keep each rule short and specific, and explain what you created.
Tips: Plain .md files inside .cursor/rules are ignored; rules there need the .mdc extension. Use AGENTS.md if you prefer plain markdown.
What it's for: Learning how part of your app works without risking any edits.
When to use it: You're about to change code the AI wrote and don't understand it. Switch to Ask mode, which doesn't edit files.
Explain how [FEATURE, e.g. login] works in this app. @[FILE OR FOLDER]
Please:
1. Summarize the flow in plain English, step by step.
2. List the files involved and what each one does.
3. Point out anything that looks fragile or insecure.
4. Tell me which file I'd change to [GOAL], and what could break.
Tips: Use @-mentions for files and folders when you know where to look. If you don't, leave them out and let Cursor search.
What it's for: Getting a reviewable plan before the Agent edits lots of files.
When to use it: The feature touches several files or has more than one way to build it. Press Shift+Tab to switch to Plan Mode.
Plan how to add [FEATURE] to my app.
Must have: [REQUIREMENTS]
Don't touch: [FILES OR AREAS]
Ask me clarifying questions first, then give me a plan with small steps that each end in something I can test, plus the risks.
Tips: Edit the plan before building. If the result misses the mark, revert, sharpen the plan, and run it again. That's often faster than patching.
What it's for: Giving the Agent the real error output instead of your summary of it.
When to use it: A command failed in Cursor's terminal (a build, a test, or npm run dev).
This failed: @Terminals
Please:
1. Explain the first real error in plain English.
2. Give me 3 possible causes and a quick check for each, before changing code.
3. Fix the most likely cause, re-run the same command, and show me the new output.
4. List every file you changed.
Tips: If a fix makes things worse, restore a checkpoint from the chat timeline to roll your files back. For bugs that are hard to reproduce, try Debug mode.
What it's for: A second look at your uncommitted work for bugs and security slips.
When to use it: Before every commit, especially after a long Agent session.
Review my uncommitted changes: @Commit (Diff of Working State)
Look for:
1. Bugs or edge cases I missed.
2. Secret keys, debug logs, or test code left in.
3. API routes or database queries that don't check who the user is.
4. Changes to files that weren't part of the task.
Sort what you find by severity. Don't fix anything until I say so.
Tips: Use @Branch (Diff with Main) to review everything on your branch at once.
5 free + 104 in packs5 free prompts on this tool's page, plus 104 prompts in 7 paid packs
Cursor (Anysphere)
Cursor cloud agents can now build shareable interactive canvases (dashboards, reports), and a beta Environments API manages agent cloud environments in code.
Best for: Cursor users who run cloud agents and want shareable outputs or scripted agent setups
