cursor

Set up Cursor agents in an hour: rules, skills, AGENTS.md, first artifact

You can get a working first pass of a Cursor agent setup in about an hour by doing four things in order: write a short AGENTS.md in the project root, give the agent one small real task, turn the first mistake you see into one project rule, and package one repeat task as a skill. Cursor's docs describe AGENTS.md as a plain markdown alternative to `.cursor/rules`, and advise adding rules "only when you notice Agent making the same mistake repeatedly." The timings below are a suggested plan, not measured results.

Who this is for

You have Cursor installed and a repo open, and you want agents behaving consistently by the end of the hour, not after a week of reading.

The hour

TimeStepDone when
0–10 minWrite AGENTS.mdFile exists in the project root with commands, done-definition, hard stops
10–25 minRun one real taskAgent produces a small, checkable change
25–40 minAdd one project ruleOne `.mdc` file in `.cursor/rules` covers the first mistake
40–60 minAdd one skill, rerunA `SKILL.md` exists and the rerun meets your done-definition

0–10 min: AGENTS.md

Create `AGENTS.md` in the project root. Cursor treats it as "a plain markdown file without metadata," so no frontmatter. Write three things: build and test commands, a definition of done, and hard stops. Keep it under a screen.

10–25 min: one real task

Pick a small bug or a missing test. Ask the agent to fix it and meet your definition of done. Don't correct it mid-run; note what it gets wrong.

25–40 min: one rule from the first mistake

Type `/create-rule` in Agent and describe the fix. Cursor's docs say Agent "generates the rule file with proper frontmatter and saves it to `.cursor/rules`." Project rules must use the `.mdc` extension; a plain `.md` file there is ignored. Choose the rule type that fits:

Rule typeApplies
Always ApplyApply to every chat session
Apply IntelligentlyWhen Agent decides it's relevant based on description
Apply to Specific FilesWhen file matches a specified pattern
Apply ManuallyWhen @-mentioned in chat

40–60 min: one skill, then rerun

If the task you ran is one you'll repeat, make it a skill. A skill is a folder with a `SKILL.md` in `.cursor/skills/` or `.agents/skills/`. Frontmatter needs `name` ("Lowercase letters, numbers, and hyphens only. Must match the parent folder name.") and `description`, which the agent uses to decide relevance. Cursor also ships `/create-skill`, which "Creates Agent Skills, including their structure and `SKILL.md` files." Then rerun a similar task and check it against your done-definition.

Pitfalls

  • Writing ten rules up front. Cursor's guidance: "Start simple. Add rules only when you notice Agent making the same mistake repeatedly."
  • Saving a rule as `.md`. It's ignored in `.cursor/rules`. Use `.mdc`, or put plain markdown in AGENTS.md.
  • Apply Intelligently with no description. Cursor's FAQ says to ensure a description is defined for that type.
  • Skill name doesn't match the folder. The `name` field must match the parent folder name.
  • Expecting rules to change Tab completions. Cursor's FAQ says rules don't impact Cursor Tab.

How to check it

  1. Confirm the rule is listed and active. Cursor's docs say: "From Customize you can see all rules and their status."
  2. Confirm the skill was discovered. Cursor's docs say: "To view discovered skills, open Customize in the sidebar and go to Skills."
  3. Rerun the first task and confirm the original mistake doesn't repeat.
  4. Confirm the output meets your AGENTS.md definition of done.

For the difference between AGENTS.md, CLAUDE.md, and Cursor rules, see AGENTS.md vs CLAUDE.md vs Cursor rules.

Where ShipKits fits

If you'd rather start the hour with the files already written, the Operator Pack ($47) is markdown seats + skills + install.sh — not a hosted app, not a chatbot. Seven named seats. Three skills. Ops playbook + token-cost playbook. Done means a file, PR, or URL. If you want to build it yourself, start from the free files at /free. Cheat Sheet for the skim checklist mindset.

Soft next step

Block an hour and do the four steps in order. Free starter files are at /free. When you want seats, skills, and playbooks already assembled, the Operator Pack ($47) has them.

FAQ

Can I set up a Cursor agent in an hour?+

A working first pass, yes: one instruction file, one rule, and one small real task. Treat the hour as a starting point you refine after the agent makes its first mistakes.

What should I do first?+

Write a short AGENTS.md in the project root with your build and test commands, a definition of done, and hard stops. Everything else builds on that.

When should I add a project rule?+

After the agent's first run, when you see a mistake you'd otherwise correct again. Cursor's own guidance is to add rules only when the agent repeats a mistake.

How do I create the rule quickly?+

Type `/create-rule` in Agent and describe what you want; Cursor saves it to `.cursor/rules` with frontmatter. Keep it to one topic.

What counts as a real first artifact?+

A small, checkable change that meets your definition of done, such as a fixed bug with a passing test. If you can't check it, the setup hasn't been tested yet.

What should I skip in the first hour?+

Long style guides, rules for rare edge cases, and copies of docs the agent can already read. Add those later only if a real run shows you need them.

Get the free starter

Value first, then a kit when you are ready. No fake urgency. Start with Harness Desk Starter at /free.