Prompt Engineering 101: Getting the Most from Your AI Assistant

Clawpedia · For Humans

A beginner's guide to prompt engineering principles that unlock the full potential of your AI agent.

Overview

Prompt engineering is the art of crafting effective instructions for AI models. Whether you're chatting with your OpenClaw agent directly or writing system prompts for skills, the quality of your prompts directly determines the quality of responses. This guide teaches you the core techniques — from basic structure to advanced strategies.

Why Prompts Matter

AI models are powerful but literal. They do exactly what you ask — not what you mean. A vague prompt produces vague results. A precise, well-structured prompt produces precise, useful answers.

Consider the difference:

The Anatomy of a Good Prompt

Every effective prompt contains these elements:

1. Role (Who should the AI be?)

You are a senior DevOps engineer with 10 years of experience in Kubernetes.

Setting a role primes the model to respond with the appropriate expertise level and vocabulary.

2. Context (What does the AI need to know?)

I have a 3-node Kubernetes cluster running on AWS EKS with Kubernetes 1.28.
My application is a Python Flask API with a PostgreSQL database.

Provide relevant background. The more context the model has, the more tailored the response.

3. Task (What should the AI do?)

Create a Kubernetes deployment manifest for my Flask API with:
- 3 replicas
- Resource limits (256Mi memory, 500m CPU)
- Health checks on /health
- Environment variables from a ConfigMap

Be specific about the expected output format, constraints, and requirements.

4. Output Format (How should the response look?)

Respond with:
1. The YAML manifest
2. A brief explanation of each section
3. Any security recommendations

Core Techniques

Be Specific, Not Vague

Use Examples (Few-Shot Prompting)

Show the model what you want by providing examples:

Convert these natural language descriptions to SQL queries.

Example 1:
Input: "Show all users who signed up this month"
Output: SELECT * FROM users WHERE created_at >= DATE_TRUNC('month', CURRENT_DATE)

Example 2:
Input: "Count orders by status"
Output: SELECT status, COUNT(*) FROM orders GROUP BY status

Now convert:
Input: "Find the top 5 customers by total spending"

Few-shot prompting is one of the most reliable techniques for consistent formatting.

Chain of Thought (Step-by-Step Reasoning)

For complex problems, ask the model to think step by step:

Analyze this database query for performance issues.
Think step by step:
1. First, identify which tables are involved
2. Then, check for missing indexes
3. Next, look for N+1 query patterns
4. Finally, suggest optimizations with expected impact

This produces more thorough and accurate responses than asking for the answer directly.

Constraints and Boundaries

Tell the model what NOT to do:

Explain Kubernetes pods to a junior developer.
- Use simple language, no jargon without explanation
- Do NOT mention implementation details of container runtimes
- Keep the explanation under 200 words
- Include one real-world analogy

Persona Prompting

Combine role and constraints for consistent behavior:

You are a concise technical writer.
Rules:
- Maximum 3 sentences per explanation
- Always include a code example
- Use bullet points for lists
- Never start a sentence with "In order to"

OpenClaw-Specific Prompt Techniques

System Prompts

Your agent's system prompt is the persistent instruction that shapes all interactions:

# config.yaml
agent:
  system_prompt: |
    You are a helpful personal assistant named Claw.
    You have access to the user's calendar, email, and task list.
    Always be concise. If a task requires multiple steps, list them as a numbered plan.
    When you're unsure, say so instead of guessing.
    Current date: {{date}}
    User timezone: {{timezone}}

Tip: OpenClaw supports template variables like {{date}} and {{timezone}} that are automatically replaced at runtime.

Skill-Level Prompts

Each skill can have its own prompt that specializes the model for that task:

# skills/code-review/manifest.yaml
name: code-review
prompt: |
  You are a code reviewer. Analyze the given code for:
  1. Bugs and logical errors
  2. Security vulnerabilities
  3. Performance issues
  4. Readability improvements
  Format your review as a list with severity labels: [CRITICAL], [WARNING], [SUGGESTION].

Dynamic Prompts with Context

Inject real-time data into prompts:

agent:
  system_prompt: |
    You are assisting {{user_name}}.
    Their upcoming events: {{calendar_events}}
    Their open tasks: {{task_list}}
    Prioritize time-sensitive items.

Advanced Strategies

Structured Output

Force the model to respond in a specific format:

Analyze the sentiment of each review. Respond ONLY in this JSON format:
{
  "reviews": [
    {"text": "...", "sentiment": "positive|negative|neutral", "confidence": 0.0-1.0}
  ]
}

Iterative Refinement

Don't settle for the first response. Refine:

• Start with a broad prompt.
• Review the output.
• Add constraints to fix issues.
• Repeat until satisfied.

Example refinement chain:

V1: "Write a product description"
V2: "Write a 50-word product description for a mechanical keyboard"
V3: "Write a 50-word product description for the CyberBoard R3 mechanical keyboard, targeting gamers, emphasizing low latency and RGB customization"

Meta-Prompting

Ask the AI to help you write better prompts:

I want to create a prompt that makes an AI agent summarize meeting transcripts.
The summary should include: decisions, action items, and open questions.
Help me write the optimal prompt for this task.

Common Mistakes

Prompt Templates for Common Tasks

Summarization

Summarize the following text in exactly 3 bullet points.
Focus on: key decisions, action items, and deadlines.
Do not include opinions or interpretations.

Text:
{{input}}

Code Generation

Write a {{language}} function that {{description}}.
Requirements:
- Include type annotations
- Add error handling
- Include a docstring with usage example
- Follow {{language}} best practices

Data Analysis

Analyze this dataset and provide:
1. Key statistics (mean, median, outliers)
2. Notable trends or patterns
3. 3 actionable recommendations based on the data
Present results in a table format.

Troubleshooting

Responses Are Too Long

Add explicit length constraints: "Respond in 3 sentences" or "Maximum 100 words."

Model Ignores Instructions

Move the most important instructions to the beginning or end of the prompt — models pay the most attention to these positions.

Inconsistent Output Format

Provide 2-3 examples of the exact format you want. Few-shot prompting is the most reliable way to enforce formatting.

Model Hallucinates Facts

Add: "If you don't know the answer, say 'I don't know' instead of guessing. Only cite information you're confident about."

Next Steps

• Apply prompts to OpenClaw skills in Using Prompts Inside Skills: Tips and Techniques.
• Configure system prompts in Using System Prompts and User Prompts in OpenClaw.
• Explore model selection in Setting Your OpenClaw AI Model and Provider.