Prompt Engineering Patterns

1---
2name: prompt-engineering-patterns
3description: Master advanced prompt engineering techniques to maximize LLM performance, reliability, and controllability in production. Use when optimizing prompts, improving LLM outputs, or designing production prompt templates.
4---
5 
6# Prompt Engineering Patterns
7 
8Master advanced prompt engineering techniques to maximize LLM performance, reliability, and controllability.
9 
10## Do not use this skill when
11 
12- The task is unrelated to prompt engineering patterns
13- You need a different domain or tool outside this scope
14 
15## Instructions
16 
17- Clarify goals, constraints, and required inputs.
18- Apply relevant best practices and validate outcomes.
19- Provide actionable steps and verification.
20- If detailed examples are required, open `resources/implementation-playbook.md`.
21 
22## Use this skill when
23 
24- Designing complex prompts for production LLM applications
25- Optimizing prompt performance and consistency
26- Implementing structured reasoning patterns (chain-of-thought, tree-of-thought)
27- Building few-shot learning systems with dynamic example selection
28- Creating reusable prompt templates with variable interpolation
29- Debugging and refining prompts that produce inconsistent outputs
30- Implementing system prompts for specialized AI assistants
31 
32## Core Capabilities
33 
34### 1. Few-Shot Learning
35- Example selection strategies (semantic similarity, diversity sampling)
36- Balancing example count with context window constraints
37- Constructing effective demonstrations with input-output pairs
38- Dynamic example retrieval from knowledge bases
39- Handling edge cases through strategic example selection
40 
41### 2. Chain-of-Thought Prompting
42- Step-by-step reasoning elicitation
43- Zero-shot CoT with "Let's think step by step"
44- Few-shot CoT with reasoning traces
45- Self-consistency techniques (sampling multiple reasoning paths)
46- Verification and validation steps
47 
48### 3. Prompt Optimization
49- Iterative refinement workflows
50- A/B testing prompt variations
51- Measuring prompt performance metrics (accuracy, consistency, latency)
52- Reducing token usage while maintaining quality
53- Handling edge cases and failure modes
54 
55### 4. Template Systems
56- Variable interpolation and formatting
57- Conditional prompt sections
58- Multi-turn conversation templates
59- Role-based prompt composition
60- Modular prompt components
61 
62### 5. System Prompt Design
63- Setting model behavior and constraints
64- Defining output formats and structure
65- Establishing role and expertise
66- Safety guidelines and content policies
67- Context setting and background information
68 
69## Quick Start
70 
71```python
72from prompt_optimizer import PromptTemplate, FewShotSelector
73 
74# Define a structured prompt template
75template = PromptTemplate(
76    system="You are an expert SQL developer. Generate efficient, secure SQL queries.",
77    instruction="Convert the following natural language query to SQL:\n{query}",
78    few_shot_examples=True,
79    output_format="SQL code block with explanatory comments"
80)
81 
82# Configure few-shot learning
83selector = FewShotSelector(
84    examples_db="sql_examples.jsonl",
85    selection_strategy="semantic_similarity",
86    max_examples=3
87)
88 
89# Generate optimized prompt
90prompt = template.render(
91    query="Find all users who registered in the last 30 days",
92    examples=selector.select(query="user registration date filter")
93)
94```
95 
96## Key Patterns
97 
98### Progressive Disclosure
99Start with simple prompts, add complexity only when needed:
100 
1011. **Level 1**: Direct instruction
102   - "Summarize this article"
103 
1042. **Level 2**: Add constraints
105   - "Summarize this article in 3 bullet points, focusing on key findings"
106 
1073. **Level 3**: Add reasoning
108   - "Read this article, identify the main findings, then summarize in 3 bullet points"
109 
1104. **Level 4**: Add examples
111   - Include 2-3 example summaries with input-output pairs
112 
113### Instruction Hierarchy
114```
115[System Context] → [Task Instruction] → [Examples] → [Input Data] → [Output Format]
116```
117 
118### Error Recovery
119Build prompts that gracefully handle failures:
120- Include fallback instructions
121- Request confidence scores
122- Ask for alternative interpretations when uncertain
123- Specify how to indicate missing information
124 
125## Best Practices
126 
1271. **Be Specific**: Vague prompts produce inconsistent results
1282. **Show, Don't Tell**: Examples are more effective than descriptions
1293. **Test Extensively**: Evaluate on diverse, representative inputs
1304. **Iterate Rapidly**: Small changes can have large impacts
1315. **Monitor Performance**: Track metrics in production
1326. **Version Control**: Treat prompts as code with proper versioning
1337. **Document Intent**: Explain why prompts are structured as they are
134 
135## Common Pitfalls
136 
137- **Over-engineering**: Starting with complex prompts before trying simple ones
138- **Example pollution**: Using examples that don't match the target task
139- **Context overflow**: Exceeding token limits with excessive examples
140- **Ambiguous instructions**: Leaving room for multiple interpretations
141- **Ignoring edge cases**: Not testing on unusual or boundary inputs
142 
143## Integration Patterns
144 
145### With RAG Systems
146```python
147# Combine retrieved context with prompt engineering
148prompt = f"""Given the following context:
149{retrieved_context}
150 
151{few_shot_examples}
152 
153Question: {user_question}
154 
155Provide a detailed answer based solely on the context above. If the context doesn't contain enough information, explicitly state what's missing."""
156```
157 
158### With Validation
159```python
160# Add self-verification step
161prompt = f"""{main_task_prompt}
162 
163After generating your response, verify it meets these criteria:
1641. Answers the question directly
1652. Uses only information from provided context
1663. Cites specific sources
1674. Acknowledges any uncertainty
168 
169If verification fails, revise your response."""
170```
171 
172## Performance Optimization
173 
174### Token Efficiency
175- Remove redundant words and phrases
176- Use abbreviations consistently after first definition
177- Consolidate similar instructions
178- Move stable content to system prompts
179 
180### Latency Reduction
181- Minimize prompt length without sacrificing quality
182- Use streaming for long-form outputs
183- Cache common prompt prefixes
184- Batch similar requests when possible
185 
186## Resources
187 
188- **references/few-shot-learning.md**: Deep dive on example selection and construction
189- **references/chain-of-thought.md**: Advanced reasoning elicitation techniques
190- **references/prompt-optimization.md**: Systematic refinement workflows
191- **references/prompt-templates.md**: Reusable template patterns
192- **references/system-prompts.md**: System-level prompt design
193- **assets/prompt-template-library.md**: Battle-tested prompt templates
194- **assets/few-shot-examples.json**: Curated example datasets
195- **scripts/optimize-prompt.py**: Automated prompt optimization tool
196 
197## Success Metrics
198 
199Track these KPIs for your prompts:
200- **Accuracy**: Correctness of outputs
201- **Consistency**: Reproducibility across similar inputs
202- **Latency**: Response time (P50, P95, P99)
203- **Token Usage**: Average tokens per request
204- **Success Rate**: Percentage of valid outputs
205- **User Satisfaction**: Ratings and feedback
206 
207## Next Steps
208 
2091. Review the prompt template library for common patterns
2102. Experiment with few-shot learning for your specific use case
2113. Implement prompt versioning and A/B testing
2124. Set up automated evaluation pipelines
2135. Document your prompt engineering decisions and learnings
214