Hi, I have been working on this project for a while now, and i think i found the perfect way to setup my cursor-tasks file the Agent can follow. from a youtube video i saw that they fed an already created repo to get lists of tasks. however, since the project i was working on doesn’t have a clear competitor i decided to go ahead and create a different way of setting up cursor-tasks.
First, i will discuss my idea with ChatGPT/Claude, then i will ask it to create a readme with the best tech stack, at first to be honest i didn’t check the tech stack carefully, and it ended up with a lot of recently made tech which claude doesn’t really grasp. thus, i decided to ask it directly to use tech stack an AI Coder like claude sonnet or chatgpt comfortably good at. the readme should also include folder structure. you can also ask the AI model to create cursor rules file based on popular cursor rule for your specific project with globs.
After it created the readme, i will ask it to to create a task list with the following prompt :
"Create a comprehensive development guide for [PROJECT NAME], following the KISS (Keep It Simple, Stupid) methodology. The guide should be divided into [NUMBER] logical parts, with each part focusing on different phases of development. For each task in the guide: 1. Add priority levels (Critical, High, Medium, Low) 2. Include effort estimates in days 3. Add clear success criteria that define when a task is complete The guide should be structured with checkboxes for each task and subtask, making it easy to track progress. Avoid code snippets and focus on clear, actionable descriptions. Each part should include: - An introduction explaining that phase’s focus - Major components to be developed in that phase - Detailed task breakdowns with subtasks - Dependencies between tasks (if applicable) - A brief overview of what’s coming in the next part The project involves [BRIEF DESCRIPTION OF YOUR PROJECT, TECHNOLOGIES, AND GOALS]. Start with Part 1 focusing on [INITIAL PHASE FOCUS, e.g., “setting up the core infrastructure and foundation”]. After completing each part, wait for my confirmation before proceeding to the next part.
Then on cursor I will ask the agent to do the following :
Process each task in the @.cursor-tasks file sequentially. Do not take screenshots after completing tasks or stories. Follow our security best practices for all implementations, especially Row Level Security policies in Supabase and proper authentication checks in Workers. When working with AI APIs, implement cost-efficient patterns with proper caching and error handling. Mark completed task with [x] in the .cursor-tasks file Maintain existing formatting and indentation.
honestly, the results are amazing, for someone who hasn’t developed projects on this scale, it feels like having a team of developers working with you.
Anyways, what is your way with working with Cursor AI Agent?
credit to kleneway (Kevin Leneway) · GitHub for the idea and guide to creating cursor-task