AI Assistant Initialization Prompts

A collection of battle-tested initialization prompts for working with AI coding assistants. These prompts implement systematic methodologies to prevent AI eagerness and ensure high-quality, verified code development.

Philosophy

AI assistants have a fundamental tendency toward over-eagerness - rushing to implement solutions, adding unrequested features, and claiming completion without verification. This repository provides initialization prompts that channel this eagerness into productive, verified progress through systematic human-AI collaboration.

Philosophy Deep Dive

Read universal-principles.md for a complete understanding of the methodology these prompts implement.

Core Principle: "Measure Twice, Cut Once"

Repository Structure

ai-initialization-prompts/
├── README.md
├── universal-principles.md
├── browser-based/
│   ├── 01-initialization-comprehensive.md
│   ├── 01-initialization-condensed.md
│   ├── 02-new-project.md
│   ├── 02-analyze-codebase.md
│   ├── 02-refactor-existing.md
│   └── 02-debug-investigation.md
├── ide-based/
│   ├── 01-initialization.md
│   ├── 02-new-project.md
│   └── 02-analyze-codebase.md
└── specialized/
    ├── rapid-prototyping.md
    └── documentation-writing.md

(some files are missing b/c this project is a wip ☺️)

How to Use

For Browser-Based AI Assistants (Claude, ChatGPT, etc.)

1. Start a new conversation
2. Copy and paste the entire 01-initialization prompt (comprehensive or condensed based on preference)
3. Wait for the AI's acknowledgment
4. Copy and paste the appropriate 02- prompt for your specific task
5. Provide your project details where indicated in the Step 2 prompt

For IDE-Based AI Assistants (Cursor, Cody, Copilot Chat, etc.)

1. Start a new session
2. Copy and paste the appropriate initialization prompt
3. Follow with your task-specific prompt

Key Features

Universal Principles

All prompts implement eight universal principles:

1. Permission-Based Action
2. Evidence Over Assertions
3. Systematic Progression Through Living Documentation
4. Transparent Limitations
5. Critical Thinking Over Agreement
6. Quality Through Review Layers
7. No Time Estimates
8. Collaboration Over Completion

Multi-Step Initialization

• Step 1: Establishes working principles and prevents eagerness
• Step 2: Introduces specific task with appropriate constraints

Investigation-Focused Debugging

Instead of "fixing" problems, the AI investigates systematically to understand issues before proposing changes.

Living Game Plans

Every project begins with a detailed game plan that serves as the single source of truth throughout development.

Prompt Types

Browser-Based Assistants

• Comprehensive: Full principles and detailed workflow (~1500 words)
• Condensed: Essential principles and workflow (~500 words)

Step 2 Variants

• New Project: Starting from scratch with full planning
• Analyze Codebase: Understanding existing code before modifications
• Refactor Existing: Systematic improvement of working code
• Debug Investigation: Scientific approach to troubleshooting

Best Practices

1. Always use Step 1 first - Never skip initialization
2. One task at a time - Complete projects systematically
3. Verify everything - The human verifies, the AI suggests
4. Update game plans - Mark progress and learnings
5. Investigate, don't fix - Understand problems before solving

Common Pitfalls to Avoid

• ❌ Jumping straight to task description without initialization
• ❌ Asking AI to "fix" problems (triggers eagerness)
• ❌ Proceeding without verification
• ❌ Accepting time estimates from AI
• ❌ Skipping the game plan phase

Contributing

These prompts are battle-tested through real development work. Contributions should:

• Follow the established philosophical framework
• Include concrete examples from actual usage
• Maintain the focus on preventing AI eagerness
• Provide clear documentation

License

Copyright (c) 2025 Dennis Porter

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

Acknowledgments

These prompts were developed through extensive real-world AI-assisted development, learning from both successes and failures in human-AI collaboration.

Created by Dennis Porter - dennismars.com