Skip to main content

Anthropic Prompt Engineering Best Practices

Anthropic Prompt Engineering Best Practices

Research Date: December 2025 Purpose: Document prompt engineering best practices for Claude Sources: Anthropic documentation, engineering blogs, community best practices

Executive Summary​

Effective prompt engineering balances clarity, specificity, and brevity. Key principles include progressive disclosure, structured formatting, and iterative refinement.

1. Core Principles​

Clarity​

  • Use specific, concrete language
  • Avoid ambiguity
  • Provide examples
  • Define technical terms

Brevity​

  • Short, declarative statements
  • Bullet points over paragraphs
  • Remove unnecessary words
  • Focus on essentials

Structure​

  • Consistent heading hierarchy
  • Logical organization
  • Visual separation
  • Clear sections

2. Instruction Guidelines​

Effective Instructions​

✅ DO:

  • Be specific: "Use 2-space indentation" not "Format nicely"
  • Provide examples: Show desired output format
  • Use emphasis sparingly: "IMPORTANT" for critical requirements
  • Test iteratively: Remove what doesn't help

❌ DON'T:

  • Overuse emphasis: Reduces effectiveness
  • Write paragraphs: Use bullets instead
  • Add theoretical content: Focus on practical needs
  • Include obvious instructions: Assume basic competence

Instruction Limits​

From Anthropic research:

  • Frontier LLMs can follow ~150-200 instructions reliably
  • Claude Code system prompt contains ~50 instructions
  • Leave headroom for task-specific guidance
  • Quality degrades as instruction count increases

3. Communication Style​

Writing for AI vs. Humans​

"You're writing for Claude, not onboarding a junior dev - use short, declarative bullet points and don't write long, narrative paragraphs."

Preferred Style​

  • Concise: 10 words > 30 words
  • Declarative: "Do X" not "You should consider doing X"
  • Action-oriented: Commands and instructions
  • Specific: Concrete over abstract

Example Comparison​

Poor:

"When working on this project, developers should generally try to follow best practices for code organization and structure, keeping in mind that maintainability is important for long-term success."

Good:

  • Organize code by feature, not file type
  • Use descriptive names
  • Keep functions under 50 lines
  • Document public APIs

4. Progressive Disclosure​

Principle​

Reveal information just-in-time rather than upfront.

Implementation​

## Quick Reference

Common commands: `npm run dev`, `npm test`

**For detailed workflows**, see:
- `docs/development.md` - Development processes
- `docs/deployment.md` - Deployment procedures
- `docs/testing.md` - Testing strategies

Benefits​

  • Reduced token consumption
  • Focused context
  • Scalable documentation
  • Easier maintenance

5. Output Style Configuration​

Available Styles​

  • Normal: Balanced (default)
  • Concise: Direct, minimal
  • Explanatory: Educational, detailed
  • Formal: Polished, professional
  • Custom: User-defined

Custom Output Styles​

Location: .claude/output-styles/[name].md

Example:

# Technical Writer Style

- Use active voice
- Define acronyms on first use
- Include code examples
- Provide step-by-step instructions
- Link to additional resources

6. Best Practices Summary​

Structure​

  • Use Markdown headings consistently
  • Organize content logically
  • Separate sections clearly
  • Include navigation aids

Content​

  • Be specific and concrete
  • Provide examples
  • Use bullet points
  • Avoid paragraphs

Optimization​

  • Test and iterate
  • Remove ineffective content
  • Use progressive disclosure
  • Monitor instruction count

Communication​

  • Write for AI consumption
  • Use declarative style
  • Emphasize sparingly
  • Stay focused

7. Source References​

  1. Anthropic Prompt Engineering Guide
  2. Claude Code Best Practices
  3. Effective Context Engineering

Last Updated: December 2025 Status: Production-validated best practices