Cursor, Day-to-Day
Why this guide exists
You can use Cursor every day and still miss what it added last month. You may never try the features that would help you most. Cursor ships fast. This guide keeps you current, a few new techniques a day, so you do not get left behind and you use Cursor the way it is meant to be used.
Who it is for
Cursor, Day-to-Day is an 18-day guide to Cursor's agent, for developers who have used Cursor for completions or chat but never run its agent day to day.
How it works
You build one small project, Ticket Triage, across every day. Each day starts where the day before ended and takes 20 to 40 minutes. Start here is a real lesson, in the CLI and most days the editor, that ends in a result you can check, such as a passing test. Going deeper holds the details and edge cases for when you have more time.
What you will learn
- Set which commands the agent may run without asking (Day 4).
- Keep rules in
AGENTS.mdthat load every session (Day 5). - Block a dangerous command with a hook (Day 11).
- Hand a task to a Cloud Agent and get a pull request (Day 15).
- Let Bugbot review your pull request (Day 16).
- Run the agent headless in GitHub Actions (Day 17).
The 18 days fall into five stages: getting started (Days 1 to 4), context and cost (Days 5 to 8), extending the agent (Days 9 to 13), away from the desk (Days 14 to 17), and making it yours (Day 18).

How it was tested
The Start here steps ran live on Cursor CLI 2026.11.13-eacade1 in November 2026, with Cloud Agents and Bugbot on a legacy Ultra plan and one GitHub Actions run. Key editor steps on Days 1, 2, 4, 9, and 13 ran in the Cursor editor 3.30.14. Steps not run say "written from the docs, not yet run."
Get the code
Day 1 builds Ticket Triage by hand. The code is in the ticket-triage-day-by-day repository on the cursor-guide branch, with one tag per day, cursor/day-00 to cursor/day-18. Use a tag to catch up on any day.
Start
Start at Day 1, one part a day.
Also in this series: Claude Code, Day-to-Day, OpenAI Codex, Day-to-Day, OpenCode, Day-to-Day, Pi, Day to Day, and GitHub Copilot, Day-to-Day.
The parts
- The Agent Said 4 Tests Passed. Run Them Yourself.
- Stop Patching Your Agent's Mistakes. Rewind Them.
- Ask First, Plan Second, Build Third: The Three Modes of Cursor's Agent
- Stop Clicking Run Once: Allow, Deny, and Sandbox the Cursor Agent
- Stop Repeating Yourself to Cursor: Three Lines in AGENTS.md That Load Every Session
- Why Your Cursor Agent Forgot: Four Runs That Show What It Can See
- One Question, Three Models: Choosing a Model in the Cursor CLI
- Set the Cap Before You Need It: What a Day of Cursor Actually Costs
- Stop Retyping Your Best Prompt: Turn It Into a Cursor Skill
- Ready Is Not Proof: Connect an MCP Server to Cursor and Watch the Agent Call It
- A Rule Asks, a Hook Says No: Block
rm -rfin Cursor With Your Own Script - The Reviewer That Cannot Edit: A Read-Only Cursor Subagent in One File
- One Folder, One Flag: Package Your Cursor Skill and Subagent as a Plugin
- Two Agents, One Repository, Zero Collisions: Cursor Worktrees in the CLI
- Close the Laptop, Keep the Agent Working: Your First Cursor Cloud Agent
- I Left a Test Out on Purpose. Cursor Bugbot Found It in 26 Seconds.
- Success Is Not Done: Running the Cursor Agent Headless, Locally and in GitHub Actions
- One Ripped Jacket, One Ordinary Day: The Cursor Agent Workflow From Casual to Elite