Prompt Writer

Overview

This skill provides expert guidance for crafting effective prompts for LLMs and AI agents. It applies research-backed techniques from cognitive science, prompt engineering research, and production LLM systems to create prompts that are clear, reliable, and minimize hallucinations.

When to Use This Skill

Activate this skill when users request:

• "Write a prompt for [agent/chatbot/system]"
• "Help me improve this prompt"
• "Review my prompt for [task]"
• "Create a system prompt for [application]"
• "I need a prompt template for [use case]"
• Any request involving prompt engineering, instruction writing, or LLM configuration

Workflow

1. Load the Prompting Guide

ALWAYS start by reading the comprehensive prompting guide:

Read references/prompting-guide.md

This guide contains:

• Mental models for how LLMs process prompts
• Architecture of effective prompts
• Research-backed reasoning techniques (Chain of Thought, Tree of Thoughts, etc.)
• Agent-specific prompting strategies
• Anti-patterns and common mistakes
• Complete examples and templates

2. Understand the Use Case

Clarify the prompt's purpose by asking:

• What task should the LLM perform?
• Is this for a chat interface or an AI agent with tools?
• What are the inputs and expected outputs?
• Are there quality/reliability requirements?
• What failure modes should be prevented?

3. Apply Prompting Principles

Based on the guide in references/prompting-guide.md, apply relevant techniques:

For All Prompts:

• Use clear, specific instructions
• Provide concrete examples
• Define the output format explicitly
• Include relevant constraints and guardrails

For Complex Tasks:

• Apply Chain of Thought reasoning
• Break down multi-step processes
• Use self-verification steps
• Include error handling instructions

For AI Agents:

• Define tool usage patterns clearly
• Specify when to use which tools
• Include decision trees for complex workflows
• Add verification steps before actions

4. Structure the Prompt

Follow the architecture from the prompting guide:

1. Context/Role - Who is the LLM? What's the scenario?
2. Task - What exactly should be done?
3. Format - How should the output be structured?
4. Examples - Concrete demonstrations (when helpful)
5. Constraints - What to avoid, quality requirements
6. Reasoning Process - How to think through the task (for complex prompts)

5. Review and Iterate

Check the prompt against common failure modes:

• Is it specific enough to prevent hallucinations?
• Does it include examples for clarity?
• Are edge cases handled?
• Is the output format unambiguous?
• Would this work reliably across different inputs?

Resources

references/prompting-guide.md

Comprehensive guide covering:

• Part I: How LLMs Actually Work
• Part II: Architecture of Effective Prompts
• Part III: Engineering Reliable Outputs
• Part IV: Research-Backed Advanced Reasoning Techniques
• Part V: Mastering AI Agent Prompting
• Part VI: Advanced Techniques for Complex Tasks
• Part VII: Complete Examples and Templates
• Part VIII: Anti-Patterns and Common Myths
• Part IX: Failure Modes and Recovery
• Part X: Advanced Operations
• Part XI: Quick Start and Reference

Usage: Read this file at the start of every prompt-writing task. Reference specific sections as needed during the writing process.

Examples

Example 1: User Request

"Write me a system prompt for a code review agent"

Process:

1. Read references/prompting-guide.md
2. Identify this as an AI agent task (Part V)
3. Apply agent-specific patterns
4. Include tool usage instructions
5. Add verification steps

Example 2: User Request

"Help me improve this prompt: 'Summarize this document'"

Process:

1. Read references/prompting-guide.md
2. Identify missing elements (output format, length, focus areas)
3. Apply prompt architecture principles (Part II)
4. Add specificity and examples
5. Define clear success criteria

Example 3: User Request

"Create a prompt for extracting structured data from unstructured text"

Process:

1. Read references/prompting-guide.md
2. Apply Chain of Thought reasoning (Part IV)
3. Define explicit output schema
4. Include examples of input → output
5. Add validation steps to prevent hallucinations