prompt-engineering

Prompt Engineering

Patterns and techniques for crafting effective Claude prompts. Covers system prompts, few-shot learning, chain-of-thought reasoning, structured output, and prefilling.

When to Use

• User is writing or optimizing prompts for Claude
• User needs structured JSON output from Claude
• User is designing system prompts or instructions
• User wants chain-of-thought reasoning or few-shot examples
• User is building prompt templates for reuse

Core Patterns

System Prompts

System prompts set Claude's persona, constraints, and output format. Place stable instructions here; they are cached separately and can use prompt caching.

message = client.messages.create(
    model="claude-sonnet-4-6-20250514",
    max_tokens=1024,
    system="""You are a senior code reviewer. Follow these rules:
1. Focus on bugs, security issues, and performance problems.
2. Rate severity as CRITICAL, HIGH, MEDIUM, or LOW.
3. Provide a fix for each issue found.
4. If the code is clean, say "No issues found." and nothing else.""",
    messages=[{"role": "user", "content": f"Review this code:\n```\n{code}\n```"}]
)

Few-Shot Examples

Provide 2-4 input/output examples to demonstrate the exact format and reasoning you expect. Use the messages array with alternating user/assistant turns.

messages = [
    {"role": "user", "content": "Classify: 'My order never arrived and nobody responds to emails'"},
    {"role": "assistant", "content": '{"category": "shipping", "sentiment": "negative", "priority": "high"}'},
    {"role": "user", "content": "Classify: 'Love the new feature update, works great!'"},
    {"role": "assistant", "content": '{"category": "feedback", "sentiment": "positive", "priority": "low"}'},
    {"role": "user", "content": f"Classify: '{user_input}'"}
]

message = client.messages.create(
    model="claude-sonnet-4-6-20250514",
    max_tokens=256,
    messages=messages
)

Chain-of-Thought Reasoning

Ask Claude to reason step-by-step before giving a final answer. Use extended thinking for complex problems that benefit from deep reasoning.