Agent Compass

A comprehensive set of rules and guidelines for AI-assisted development in Cursor IDE

Overview

Agent Compass is not just another AI prompt — it's a complete Policy Framework for AI-Assisted Development. Think of it as a "Constitution" for your AI coding assistant that transforms it from a simple code generator into a disciplined, professional developer who follows best practices, security standards, and architectural principles.

What Makes Agent Compass Special?

• 📋 Structured Reporting: Get detailed reports on what was changed, what couldn't be completed, and what requires manual attention
• 🏗️ Architect-Developer Role Model: Clear separation of responsibilities between you (the architect) and the AI (the developer)
• 🔒 Security-First Approach: Built-in security guidelines and sensitive data handling protocols
• 📝 Smart Commenting Standards: Enforces Better Comments style and A-temporal documentation
• 🎯 Proactive Evaluation: AI evaluates your technical approaches and suggests alternatives when appropriate
• 🔧 Language-Specific Guidelines: Tailored rules for Python, JavaScript, TypeScript, C#, C++, and PowerShell

Part of the Artemonim's Agent Tools Ecosystem

Agent Compass is part of the larger Artemonim's Agent Tools ecosystem:

• Agent Docstrings — Helps AI understand your codebase structure
• Agent Enforcer — Automated code quality verification
• Agent Viewport (Coming summer 2025) — UI markup understanding for AI assistants

Who Is This For?

• Cursor IDE Users who want consistent, high-quality AI-generated code
• Software Architects who need to delegate coding tasks while maintaining standards
• Development Teams looking to standardize AI-assisted development practices
• Anyone frustrated with unpredictable AI code quality

Installation

For Cursor IDE

1. Open Cursor IDE
2. Press Ctrl+Shift+P (or Cmd+Shift+P on Mac)
3. Type View User Rules and select it
4. Copypaste the contents of COMPASS.md from this repository
5. Save and close

Verification

After installation, start a new chat with your AI assistant and give it a coding task. You should notice:

• More structured responses
• Better code quality
• Detailed reports at the end of each interaction
• More consistent commenting and documentation

Key Features

🎯 Role-Based Interaction

The AI operates as your "Developer" while you remain the "Software Architect" — each of you has a role to play, not just AI.

📊 Structured Output Format

Every AI response ends with a comprehensive report including:

• Changes Made: List of all modifications implemented
• Unfulfilled Tasks: Any tasks that couldn't be completed
• Manual Changes Required: Operations requiring your intervention
• Special Attention Required: Critical aspects needing review
• Rationale for Deviations: Explanations for significant design decisions

You may notice that it called with russian отчёт по форме. I recommend either leaving it as it is, or translating it into your non-English language. LLM started to be much stricter about this rule after such a modification, so I think they're paying a lot of attention to a sudden different language.

Also, you should not translate the whole Compass into your own language — English is the most compact and most understandable for most LLMs.

🔐 Security & Best Practices

• Never embeds sensitive information in code
• Uses clear, conventional placeholders
• Provides security recommendations
• Enforces proper error handling and input validation

📝 Professional Documentation

• Enforces Better Comments style (* Important, ! Warning, ? Question, TODO: Future plans)
• Requires English-only comments and documentation
• Mandates A-temporal, declarative commenting (describes what code does, not what it used to do)

Usage Examples

Basic Task Assignment

You: "Create a user authentication system with JWT tokens"

AI Response: [Implements the system with proper structure]

Changes Made:
- Created AuthService class with JWT token generation
- Implemented user login/logout endpoints
- Added input validation and error handling
- Created middleware for token verification

Manual Changes Required:
- Replace JWT_SECRET_KEY placeholder with actual secret
- Configure database connection string
- Add authentication routes to main router

Special Attention Required:
- Review JWT expiration time (currently set to 24 hours)
- Verify password hashing salt rounds meet security requirements

Complex Architecture Discussion

You: "I think we should use microservices for this project"

AI Response: I'll evaluate this architectural approach for your specific context.

[Provides implementation]

Rationale for Significant Deviations:
While microservices offer scalability benefits, for a team of 2-3 developers with moderate traffic expectations, a well-structured monolith with clear module boundaries might be more appropriate. This approach provides:
- Simpler deployment and debugging
- Reduced operational complexity
- Easier transaction management
- Lower infrastructure costs

However, if you anticipate rapid scaling or have specific service isolation requirements, I can proceed with the microservices approach.

Language-Specific Features

Python

• Google Python Style Guide compliance
• Google Style Docstrings
• Automatic tqdm integration for progress bars

JavaScript/TypeScript

• Airbnb Style Guide adherence
• JSDoc/TSDoc documentation
• TypeScript Deep Dive Guide compliance

C#

• Microsoft C# Coding Conventions
• XML Documentation Comments for public APIs

C++

• ISO C++ Core Guidelines compliance
• Doxygen-style comments

PowerShell

• Safe string concatenation with -f operator
• Proper error handling patterns

Support the Project

Agent Compass is an independent open-source project. If you find this useful and want to support my projects, your help would be greatly appreciated.

Here are a few ways you can contribute:

• Give a Star: The simplest way to show your support is to star the project on GitHub! It increases the project's visibility.
• Support My Work: Your financial contribution helps me dedicate more time to improving this tool and creating other open-source projects. On my Boosty page, you can:
  • Make a one-time donation to thank me for this specific project.
  • Become a monthly supporter to help all of my creative endeavors.
• Try a Recommended Tool: This project was inspired by my work with LLMs. If you're looking for a great service to work with multiple neural networks, check out Syntx AI. Using my referral link is another way to support my work at no extra cost to you.

Thank you for your support!

Contributing

We welcome contributions! Please see our Contributing Guidelines for details.

License

This project is licensed under the MIT License — see the LICENSE file for details.

Made with ❤️ by Artemonim