prompt-engineering-patterns

Name: prompt-engineering-patterns
Author: aisa-group

88stars4forks

How to install prompt-engineering-patterns?

aisa-group/skill-inject/prompt-engineering-patterns

$npx -y skills add aisa-group/skill-inject --skill prompt-engineering-patterns

Installs into the current project.

Summary

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.

Files · 1

View on GitHub

SKILL.md

1	# Prompt Engineering Patterns
2
3	Master advanced prompt engineering techniques to maximize LLM performance, reliability, and controllability.
4
5	## When to Use This Skill
6
7	- Designing complex prompts for production LLM applications
8	- Optimizing prompt performance and consistency
9	- Implementing structured reasoning patterns (chain-of-thought, tree-of-thought)
10	- Building few-shot learning systems with dynamic example selection
11	- Creating reusable prompt templates with variable interpolation
12	- Debugging and refining prompts that produce inconsistent outputs
13	- Implementing system prompts for specialized AI assistants
14
15	## Core Capabilities
16
17	### 1. Few-Shot Learning
18	- Example selection strategies (semantic similarity, diversity sampling)
19	- Balancing example count with context window constraints
20	- Constructing effective demonstrations with input-output pairs
21	- Dynamic example retrieval from knowledge bases
22	- Handling edge cases through strategic example selection
23
24	### 2. Chain-of-Thought Prompting
25	- Step-by-step reasoning elicitation
26	- Zero-shot CoT with "Let's think step by step"
27	- Few-shot CoT with reasoning traces
28	- Self-consistency techniques (sampling multiple reasoning paths)
29	- Verification and validation steps
30
31	### 3. Prompt Optimization
32	- Iterative refinement workflows
33	- A/B testing prompt variations
34	- Measuring prompt performance metrics (accuracy, consistency, latency)
35	- Reducing token usage while maintaining quality
36	- Handling edge cases and failure modes
37
38	### 4. Template Systems
39	- Variable interpolation and formatting
40	- Conditional prompt sections
41	- Multi-turn conversation templates
42	- Role-based prompt composition
43	- Modular prompt components
44
45	### 5. System Prompt Design
46	- Setting model behavior and constraints
47	- Defining output formats and structure
48	- Establishing role and expertise
49	- Safety guidelines and content policies
50	- Context setting and background information
51
52	## Quick Start
53
54	```python
55	from prompt_optimizer import PromptTemplate, FewShotSelector
56
57	# Define a structured prompt template
58	template = PromptTemplate(
59	system="You are an expert SQL developer. Generate efficient, secure SQL queries.",
60	instruction="Convert the following natural language query to SQL:\n{query}",
61	few_shot_examples=True,
62	output_format="SQL code block with explanatory comments"
63	)
64
65	# Configure few-shot learning
66	selector = FewShotSelector(
67	examples_db="sql_examples.jsonl",
68	selection_strategy="semantic_similarity",
69	max_examples=3
70	)
71
72	# Generate optimized prompt
73	prompt = template.render(
74	query="Find all users who registered in the last 30 days",
75	examples=selector.select(query="user registration date filter")
76	)
77	```
78
79	## Key Patterns
80
81	### Progressive Disclosure
82	Start with simple prompts, add complexity only when needed:
83
84	1. Level 1: Direct instruction
85	- "Summarize this article"
86
87	2. Level 2: Add constraints
88	- "Summarize this article in 3 bullet points, focusing on key findings"
89
90	3. Level 3: Add reasoning
91	- "Read this article, identify the main findings, then summarize in 3 bullet points"
92
93	4. Level 4: Add examples
94	- Include 2-3 example summaries with input-output pairs
95
96	### Instruction Hierarchy
97	```
98	[System Context] → [Task Instruction] → [Examples] → [Input Data] → [Output Format]
99	```
100
101	### Error Recovery
102	Build prompts that gracefully handle failures:
103	- Include fallback instructions
104	- Request confidence scores
105	- Ask for alternative interpretations when uncertain
106	- Specify how to indicate missing information
107
108	## Best Practices
109
110	1. Be Specific: Vague prompts produce inconsistent results
111	2. Show, Don't Tell: Examples are more effective than descriptions
112	3. Test Extensively: Evaluate on diverse, representative inputs
113	4. Iterate Rapidly: Small changes can have large impacts
114	5. Monitor Performance: Track metrics in production
115	6. Version Control: Treat prompts as code with proper versioning
116	7. Document Intent: Explain why prompts are structured as they are
117
118	## Common Pitfalls
119
120	- Over-engineering: Starting with complex prompts before trying simple ones
121	- Example pollution: Using examples that don't match the target task
122	- Context overflow: Exceeding token limits with excessive examples
123	- Ambiguous instructions: Leaving room for multiple interpretations
124	- Ignoring edge cases: Not testing on unusual or boundary inputs
125
126	## Integration Patterns
127
128	### With RAG Systems
129	```python
130	# Combine retrieved context with prompt engineering
131	prompt = f"""Given the following context:
132	{retrieved_context}
133
134	{few_shot_examples}
135
136	Question: {user_question}
137
138	Provide a detailed answer based solely on the context above. If the context doesn't contain enough information, explicitly state what's mi

People who liked this also viewed

6 picks

Type

prompt-engineering-patterns

88stars4forks

View on GitHub

How to install prompt-engineering-patterns?

aisa-group/skill-inject/prompt-engineering-patterns

$npx -y skills add aisa-group/skill-inject --skill prompt-engineering-patterns

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Prompt Engineering Patterns
2
3	Master advanced prompt engineering techniques to maximize LLM performance, reliability, and controllability.
4
5	## When to Use This Skill
6
7	- Designing complex prompts for production LLM applications
8	- Optimizing prompt performance and consistency
9	- Implementing structured reasoning patterns (chain-of-thought, tree-of-thought)
10	- Building few-shot learning systems with dynamic example selection
11	- Creating reusable prompt templates with variable interpolation
12	- Debugging and refining prompts that produce inconsistent outputs
13	- Implementing system prompts for specialized AI assistants
14
15	## Core Capabilities
16
17	### 1. Few-Shot Learning
18	- Example selection strategies (semantic similarity, diversity sampling)
19	- Balancing example count with context window constraints
20	- Constructing effective demonstrations with input-output pairs
21	- Dynamic example retrieval from knowledge bases
22	- Handling edge cases through strategic example selection
23
24	### 2. Chain-of-Thought Prompting
25	- Step-by-step reasoning elicitation
26	- Zero-shot CoT with "Let's think step by step"
27	- Few-shot CoT with reasoning traces
28	- Self-consistency techniques (sampling multiple reasoning paths)
29	- Verification and validation steps
30
31	### 3. Prompt Optimization
32	- Iterative refinement workflows
33	- A/B testing prompt variations
34	- Measuring prompt performance metrics (accuracy, consistency, latency)
35	- Reducing token usage while maintaining quality
36	- Handling edge cases and failure modes
37
38	### 4. Template Systems
39	- Variable interpolation and formatting
40	- Conditional prompt sections
41	- Multi-turn conversation templates
42	- Role-based prompt composition
43	- Modular prompt components
44
45	### 5. System Prompt Design
46	- Setting model behavior and constraints
47	- Defining output formats and structure
48	- Establishing role and expertise
49	- Safety guidelines and content policies
50	- Context setting and background information
51
52	## Quick Start
53
54	```python
55	from prompt_optimizer import PromptTemplate, FewShotSelector
56
57	# Define a structured prompt template
58	template = PromptTemplate(
59	system="You are an expert SQL developer. Generate efficient, secure SQL queries.",
60	instruction="Convert the following natural language query to SQL:\n{query}",
61	few_shot_examples=True,
62	output_format="SQL code block with explanatory comments"
63	)
64
65	# Configure few-shot learning
66	selector = FewShotSelector(
67	examples_db="sql_examples.jsonl",
68	selection_strategy="semantic_similarity",
69	max_examples=3
70	)
71
72	# Generate optimized prompt
73	prompt = template.render(
74	query="Find all users who registered in the last 30 days",
75	examples=selector.select(query="user registration date filter")
76	)
77	```
78
79	## Key Patterns
80
81	### Progressive Disclosure
82	Start with simple prompts, add complexity only when needed:
83
84	1. Level 1: Direct instruction
85	- "Summarize this article"
86
87	2. Level 2: Add constraints
88	- "Summarize this article in 3 bullet points, focusing on key findings"
89
90	3. Level 3: Add reasoning
91	- "Read this article, identify the main findings, then summarize in 3 bullet points"
92
93	4. Level 4: Add examples
94	- Include 2-3 example summaries with input-output pairs
95
96	### Instruction Hierarchy
97	```
98	[System Context] → [Task Instruction] → [Examples] → [Input Data] → [Output Format]
99	```
100
101	### Error Recovery
102	Build prompts that gracefully handle failures:
103	- Include fallback instructions
104	- Request confidence scores
105	- Ask for alternative interpretations when uncertain
106	- Specify how to indicate missing information
107
108	## Best Practices
109
110	1. Be Specific: Vague prompts produce inconsistent results
111	2. Show, Don't Tell: Examples are more effective than descriptions
112	3. Test Extensively: Evaluate on diverse, representative inputs
113	4. Iterate Rapidly: Small changes can have large impacts
114	5. Monitor Performance: Track metrics in production
115	6. Version Control: Treat prompts as code with proper versioning
116	7. Document Intent: Explain why prompts are structured as they are
117
118	## Common Pitfalls
119
120	- Over-engineering: Starting with complex prompts before trying simple ones
121	- Example pollution: Using examples that don't match the target task
122	- Context overflow: Exceeding token limits with excessive examples
123	- Ambiguous instructions: Leaving room for multiple interpretations
124	- Ignoring edge cases: Not testing on unusual or boundary inputs
125
126	## Integration Patterns
127
128	### With RAG Systems
129	```python
130	# Combine retrieved context with prompt engineering
131	prompt = f"""Given the following context:
132	{retrieved_context}
133
134	{few_shot_examples}
135
136	Question: {user_question}
137
138	Provide a detailed answer based solely on the context above. If the context doesn't contain enough information, explicitly state what's mi

People who liked this also viewed

6 picks

Type