4.9 KiB
name, description
| name | description |
|---|---|
| engineering-prompts | Engineers effective prompts using systematic methodology. Use when designing prompts for Claude, optimizing existing prompts, or balancing simplicity, cost, and effectiveness. Applies progressive disclosure and empirical validation to prompt development. |
Engineering Prompts
LEVEL 1: QUICKSTART ⚡
5-Step Prompt Creation:
- Start Clear: Explicit instructions + success criteria
- Assess Need: Does it need structure? Examples? Reasoning?
- Add Sparingly: Only techniques that improve outcomes
- Estimate Cost: Count tokens, identify caching opportunities
- Test & Iterate: Measure effectiveness, refine based on results
LEVEL 2: CORE PHILOSOPHY 🎯
The Three Principles
Simplicity First
- Start with minimal prompt
- Add complexity only when empirically justified
- More techniques ≠ better results
Cost Awareness
- Minimize token usage
- Leverage prompt caching (90% savings on repeated content)
- Batch processing for non-urgent work (50% savings)
Effectiveness
- Techniques must improve outcomes for YOUR use case
- Measure impact, don't just apply best practices
- Iterate based on results
LEVEL 3: THE 9 TECHNIQUES 🛠️
Quick Reference
| Technique | When to Use | Cost Impact |
|---|---|---|
| 1. Clarity | ALWAYS | Minimal, max impact |
| 2. XML Structure | Complex prompts, instruction leakage | ~50-100 tokens |
| 3. Chain of Thought | Reasoning, analysis, math | 2-3x output tokens |
| 4. Multishot Examples | Pattern learning, format guidance | 200-1K tokens each |
| 5. System Role | Domain expertise needed | Minimal (caches well) |
| 6. Prefilling | Strict format requirements | Minimal |
| 7. Long Context | 20K+ token inputs | Better accuracy |
| 8. Context Budget | Repeated use, long conversations | 90% savings with cache |
| 9. Tool Docs | Function calling, agents | 100-500 tokens per tool |
LEVEL 4: DESIGN FRAMEWORK 📋
D - Define Requirements
Questions to Answer:
- Core task?
- Output format?
- Constraints (latency/cost/accuracy)?
- One-off or repeated?
E - Estimate Complexity
Simple:
- Extraction, formatting
- Simple Q&A
- Clear right answer
Medium:
- Analysis with reasoning
- Code generation
- Multi-step but clear
Complex:
- Deep reasoning
- Novel problem-solving
- Research synthesis
S - Start Simple
Minimal Viable Prompt:
- Clear instruction
- Success criteria
- Output format
Test first. Add complexity only if underperforming.
I - Iterate Selectively
Add techniques based on gaps:
- Unclear outputs → More clarity, examples
- Wrong structure → XML tags, prefilling
- Shallow reasoning → Chain of thought
- Pattern misses → Multishot examples
G - Guide on Cost
Cost Optimization:
- Cache system prompts, reference docs (90% savings)
- Batch non-urgent work (50% savings)
- Minimize token usage through clear, concise instructions
N - Note Implementation
Deliverables:
- The optimized prompt
- Techniques applied + rationale
- Techniques skipped + why
- Token estimate
- Caching strategy
LEVEL 5: ADVANCED TOPICS 🚀
Tool Integration
When to use MCP tools during prompt engineering:
Need latest practices?
└─ mcp__plugin_essentials_perplexity
Complex analysis needed?
└─ mcp__plugin_essentials_sequential-thinking
Need library docs?
└─ mcp__plugin_essentials_context7
Context Management
Prompt Caching:
- Cache: System prompts, reference docs, examples
- Savings: 90% on cached content
- Write: 25% of standard cost
- Read: 10% of standard cost
Long Context Tips:
- Place documents BEFORE queries
- Use XML tags:
<document>,<source> - Ground responses in quotes
- 30% better performance with proper structure
Token Optimization
Reducing Token Usage:
- Concise, clear instructions (no fluff)
- Reuse examples across calls (cache them)
- Structured output reduces back-and-forth
- Tool use instead of long context when possible
Anti-Patterns
❌ Over-engineering - All 9 techniques for simple task ❌ Premature optimization - Complexity before testing simple ❌ Vague instructions - "Analyze this" without specifics ❌ No examples - Expecting format inference ❌ Missing structure - Long prompts without XML ❌ Ignoring caching - Not leveraging repeated content
Stop here if: You need advanced implementation details
LEVEL 6: REFERENCES 📚
Deep Dive Documentation
Detailed Technique Catalog:
reference/technique-catalog.md- Each technique explained with examples, token costs, combination strategies
Real-World Examples:
reference/examples.md- Before/after pairs for coding, analysis, extraction, agent tasks
Research Papers:
reference/research.md- Latest Anthropic research, benchmarks, best practices evolution