Prompt Engineering

Prompt Engineering Guide

Create effective prompts for AI coding assistants across multiple platforms (Claude, GPT-4, GitHub Copilot, Cursor, Windsurf).

Overview

Master prompt engineering fundamentals, advanced techniques, platform-specific optimizations, and proven templates. Learn to craft prompts that elicit accurate, relevant, and high-quality code and responses.

When to Use This Skill

• Creating/updating system prompts for AI agents
• Improving prompts producing inconsistent results
• Designing instruction files (.instructions.md, SKILL.md)
• Optimizing agent behavior for specific tasks
• Debugging problematic prompts
• Migrating prompts between platforms

Before You Write: 9-Dimension Analysis

Before writing any prompt, silently extract these 9 dimensions from the request:

Missing dimensions → add reasonable defaults or ask the user before writing.

Framework Selector

Choose the architecture before writing. See prompt frameworks for full reference.

Core Principles

1. Clarity and Specificity

Be explicit and unambiguous:

Poor:

Do something with this text

Good:

Extract all dates, names, and organizations from this text.
Return as JSON with fields: dates[], names[], organizations[]

Rules:

• Use concrete examples
• Define exact success criteria
• Avoid ambiguity
• Specify output format precisely

2. Recommend Structure

<role>AI role and primary task</role>

<context>Necessary background information</context>

<instructions>
Step-by-step guidelines
1. First step
2. Second step
</instructions>

<output_format>Precise format description</output_format>

<examples>
<example id="1">
<input>Sample input</input>
<output>Expected output</output>
</example>
</examples>

<constraints>Task boundaries and what NOT to do</constraints>

3. Effective Delimiters

Markdown (simple prompts):

# Main Instructions
## Section 1
- Point A
- Point B

XML (complex structures):

<instructions>
  <task>Main task</task>
  <constraints>
    <constraint>Limitation 1</constraint>
  </constraints>
</instructions>

See prompt engineering fundamentals for comprehensive theory.

Core Techniques

Chain of Thought (CoT)

Encourage step-by-step reasoning:

Solve step by step:
1. Analyze input data
2. Identify patterns
3. Formulate hypothesis
4. Test with examples
5. Provide final answer with justification

Use for: Complex problems, multi-step analysis, debugging, architecture decisions

Few-Shot Learning

Provide 2-5 examples:

<examples>
<example id="1">
Input: "Buy iPhone 13"
Output: {"intent": "purchase", "product": "iPhone 13"}
</example>

<example id="2">
Input: "This phone is terrible!"
Output: {"intent": "complaint", "product": "unknown"}
</example>
</examples>

Now process: "Looking for good smartphone under $500"

Use for: Specific output formats, pattern learning, classification

Self-Consistency

For critical tasks:

Solve using three different methods.
Compare results and choose the most justified answer.
Explain your selection.

Use for: High-stakes decisions, complex calculations, architecture validation

See advanced techniques for more methods.

Platform-Specific Notes

• Claude: Excels with XML structure, supports prefilling, extended thinking
• GPT-4.1: Optimized for agentic workflows, long context (1M tokens), tool use
• Copilot/Cursor: Workspace-aware, instruction files, inline suggestions
• Local models: Simpler structures, more examples, shorter context

See platform-specific techniques for detailed optimizations.

Quick Reference

Prompt Creation Workflow

1. Define goal clearly
2. Choose structure (markdown/XML)
3. Add examples (1-5)
4. Specify output format
5. Add constraints
6. Test and iterate

Common Mistakes

• Ambiguous instructions
• Missing output format specification
• Too few/many examples
• Ignoring platform differences
• No constraints specified

Prompt Length Guidelines

• Simple tasks: 100-500 tokens
• Complex tasks: 500-2000 tokens
• System prompts: 1000-5000 tokens
• Max recommended: 8000 tokens

Examples Repository

See comprehensive examples for:

• System prompts for agents
• Code generation prompts
• Code review prompts
• Testing prompts
• Documentation prompts
• Refactoring prompts

See prompt templates for ready-to-use templates.

Further Reading

• Prompt Frameworks - Named frameworks (RTF, CO-STAR, RISEN, CRISPE, ReAct), Memory Block, PAC2026 positional structure
• Prompt Engineering Fundamentals - Theory and principles
• Advanced Prompting Techniques - Expert methods
• Platform-Specific Techniques - Claude, GPT-4, Copilot optimizations
• Practical Guidelines - Real-world best practices
• Comprehensive Examples - Copy-paste ready prompts
• Prompt Templates - Reusable templates

Source: ziltorian/ContextOrchestration — distributed by TomeVault.