Cursor CLI Tutorial for Beginners (should you use it?)

Key Concepts

• Cursor CLI: A command-line tool from Cursor that leverages Cursor's core code editing technology.
• Clawed Code: A similar tool, likely from a competitor, with custom slash commands.
• Cursor Agent: The executable for the Cursor CLI, used to interact with the Cursor AI.
• Feature Planning: Using the AI to create a detailed plan before implementing a new feature.
• Model Selection: Choosing between different AI models (e.g., GPT-4, Sonnet 4) for different tasks.
• Context Clearing: Resetting the AI's context to avoid confusion from previous files and commands.
• Code Review: Using the AI to review code changes and identify potential issues.
• Parallel Execution: Running multiple Cursor Agent instances in separate terminals to work on different parts of a project simultaneously.
• Token Usage & Cost: The amount of tokens used by the AI, which directly impacts the cost of using the Cursor CLI.
• Max Mode: A potentially more powerful but also more expensive mode for the AI.
• API Key Parameter: Using an API key to integrate the Cursor Agent into custom scripts and workflows.

Getting Started with Cursor CLI

• Installation: Run the provided command in a terminal (WSL required on Windows).
• Authentication: Authenticate via a URL using your Cursor account.
• Interface: The Cursor CLI interface is similar to Cursor chat.
• Basic Usage: Tag files and provide instructions to the AI.

Feature Planning and Implementation

• Planning a Feature: The presenter aims to build a Trello-like board for Vibe Scan feature tracking.
• Prompting: Using a pre-written prompt (from a file) to guide the AI in planning the feature.
• Model Selection: Switching from GPT-4 to Sonnet 4 for better results based on experience.
• Clarification: Providing follow-up messages to refine the AI's behavior (e.g., emphasizing planning before implementation).
• Plan Review: Reviewing the generated plan and providing feedback.
• Phase Implementation: Implementing the feature in phases, starting with phase one.
• Code Review: Using the control+r shortcut to review the changes made by the AI.

Parallel Development

• Running Multiple Agents: The presenter demonstrates running two Cursor Agent instances in parallel to implement different phases of the project (2A and 2B).
• Potential Issues: Encountering issues with file updates and needing to revert to previous states using git checkout.

Code Review and Bug Fixing

• Code Review Instructions: Asking the AI to follow code review instructions to identify issues.
• UI Testing: Checking the UI to identify issues not caught by the AI.
• Error Handling: Providing compilation errors to the AI for debugging.
• Iterative Improvement: Iterating on the UI based on feedback and AI suggestions.

Vibe Scan Feature Example

• Implemented Feature: A Trello-like board for tracking feature ideas in Vibe Scan.
• Functionality: Admins can move ideas across columns, and regular users can suggest new ideas.

Cost Considerations

• Unexpected Costs: The presenter spent $30 on the CLI without realizing it due to high token usage.
• Max Mode Default: The Cursor CLI seems to default to using the "max mode," which is more expensive.
• Lack of Transparency: The cost is not clearly surfaced while using the CLI.
• Mitigation: Switching to Sonnet 4 and using the regular Cursor Agent to reduce costs.

Output Quality and Workflow Integration

• Output Quality: The presenter found the output quality slightly better when using the Cursor Agent inside the Cursor editor.
• Workflow Integration: The presenter suggests that these CLIs are most effective when integrated into custom workflows using the API key parameter.
• Example: The presenter references a previous video demonstrating how they integrated Clawed Code into Vibe Scan for an autofix feature.

Conclusion

The Cursor CLI is a promising tool for leveraging AI in code editing and development. It allows for feature planning, implementation, and code review directly from the command line. However, users should be aware of the potential costs associated with token usage, especially when using the "max mode." The CLI is most effective when integrated into custom workflows using the API key parameter.