AI Prompting Guide

Practical techniques for getting better results from AI coding assistants when building with OnchainKit.

Get better results from AI coding assistants when building with OnchainKit. Whether you’re using Cursor, GitHub Copilot, or other AI tools, these techniques will improve your workflow.

Understanding Context Windows

Why Context Matters

AI coding assistants have a “context window” - the amount of text they can process at once. Think of it as working memory:

• Most assistants process thousands of tokens (roughly 4-5 words per token)
• Everything you share consumes this limited space
• Once full, older conversation parts may be lost

Provide relevant context upfront since AI can only work with what’s in its current window.

Optimizing for Context Windows

Optimize for context windows:

1. Share relevant info first - Most important details upfront
2. Skip unnecessary content - Avoid irrelevant code or docs
3. Structure requests clearly - Use sections and formatting
4. Reference smartly - Share only relevant files for large codebases

For larger projects, create a central documentation file rather than repeatedly explaining context.

Setting Up AI Tools

Cursor Rules

Cursor Rules provide consistent context to help AI understand your codebase better.

1. Open Command Palette: Cmd + Shift + P (Mac) or Ctrl + Shift + P (Windows/Linux)
2. Search “Cursor Rules” and create/edit rules
3. Add project-specific rules:
   • Next.js
   • Astro
   • Vite
4. Save to apply rules to AI suggestions

OnchainKit Projects

Create a new OnchainKit project:

npm create onchain@latest

After setup, generate comprehensive documentation for your project.

Creating Project Documentation

A project instructions file helps AI understand your codebase. Create this early and update regularly.

Instructions.md prompt:

Create a detailed instructions.md file for my OnchainKit project:

1. Overview: Project goals and core functionality
2. Tech Stack: Technologies, libraries, frameworks with versions
3. Project Structure: File organization with explanations
4. Coding Standards: Style conventions, linting rules, patterns
5. User Stories: Key functionality from user perspective
6. APIs and Integrations: External services and connections

For architecture planning and complex design decisions, use reasoning-focused AI models. They excel at analyzing tradeoffs, edge cases, and long-term planning.

Effective Prompting Strategies

Be Specific

AI tools respond best to clear, direct instructions.

❌ “Help me with my code” ✅ “Refactor this authentication function to use async/await instead of nested then() calls”

Provide Context

Template:

I'm working on an OnchainKit project using [frameworks/libraries]. I need help with:

1. Problem: [describe specific issue]
2. Current approach: [what you've tried]
3. Constraints: [technical limitations]
4. Expected outcome: [what success looks like]

Relevant docs: @https://onchainkit.xyz

Relevant code:
[paste your code]

Iterate

Start simple and refine rather than trying to get everything perfect at once.

Template:

Let's approach this step by step:
1. First, implement basic [feature] with minimal functionality
2. Review and identify improvements
3. Add error handling and edge cases
4. Optimize for performance

Start with step 1.

Working with OnchainKit

Using Documentation

Reference OnchainKit documentation directly in your prompts:

1. Browse OnchainKit docs
2. Find your component
3. Reference the docs URL in your prompt

Example:

I'm implementing a swap component with OnchainKit. 

Docs: @https://onchainkit.xyz

Show me how to implement a swap that:
1. Swaps USDC to ETH on Base
2. Handles connection states
3. Includes error handling
4. Follows UX best practices

Component Integration

Token balance example:

Implement a token balance display feature:

1. Shows connected wallet's balance of [TOKEN_SYMBOL]
2. Updates when balance changes
3. Handles loading and error states
4. Follows project coding standards
5. Update instructions.md with this implementation

Debugging with AI

Debugging Prompts

Bug analysis template:

I'm encountering an issue:

1. Expected behavior: [what should happen]
2. Actual behavior: [what's happening]
3. Error messages: [any errors]
4. Relevant code: [problematic code]

Analyze step by step and help me:
1. Identify potential causes
2. Suggest debugging steps
3. Propose solutions

Debug logging template:

Add comprehensive logging to this function to trace:
1. Input values and types
2. Function execution flow
3. Intermediate state changes
4. Output values or errors

Code:
[paste your code]

When Stuck

Clarification template:

I'm unsure how to proceed with [specific task]. Here's what I know:
1. [problem context]
2. [what you've tried]
3. [specific areas needing guidance]

What additional information would help you assist better?

Advanced Techniques

1. Step-by-step reasoning

   Analyze this code step by step and identify potential issues.
   

2. Format specification

   Structure your response as a tutorial with code examples.
   

3. Length guidance

   Provide a concise explanation in 2-3 paragraphs.
   

4. Clarify ambiguities

   You suggested two approaches. I prefer the first with TypeScript.
   

Best Practices

1. Understand context limitations - AI has finite memory, prioritize important info
2. Provide relevant context - Share code, errors, and project details that matter
3. Be specific - Clear instructions beat vague questions
4. Break down complex tasks - Iterative approaches work better
5. Request explanations - Ask AI to explain generated code you don’t understand
6. Use clear formatting - Structure prompts with sections
7. Reference docs - Include OnchainKit documentation links
8. Test and validate - Always review AI-generated code before implementing
9. Build on context - Reference earlier conversation parts when iterating
10. Provide feedback - Tell AI what worked and what didn’t