Agent Skills
› NeverSight/learn-skills.dev
› ai-prompting
ai-prompting
GitHub提供AI提示词工程与LLM交互模式指南，涵盖基础结构、少样本学习、思维链及系统提示设计等技巧，旨在优化用户与大模型的互动效果。
Trigger Scenarios
编写或优化提示词
设计系统提示
使用少样本或思维链技术
探索LLM交互最佳实践
Install
npx skills add NeverSight/learn-skills.dev --skill ai-prompting -g -y
SKILL.md
Frontmatter
{
"name": "ai-prompting",
"description": "AI prompt engineering and LLM interaction patterns. Use when user asks to \"write prompts\", \"optimize prompts\", \"design system prompts\", \"few-shot prompting\", \"chain-of-thought\", \"prompt techniques\", \"LLM patterns\", \"prompt best practices\", \"model interactions\", \"AI assistant design\", or mentions prompt optimization, LLM interactions, or generative AI patterns."
}
AI Prompting & LLM Patterns
Prompt engineering techniques and LLM interaction patterns for Claude, GPT, and other AI models.
Prompt Techniques
Basic Structure
Role: [Specify role/expertise]
Task: [Clear, specific task]
Context: [Relevant background]
Format: [Desired output format]
Constraints: [Any limitations]
Few-Shot Prompting
Provide 2-3 examples of input-output pairs before the actual request to guide model behavior.
Chain-of-Thought
Ask model to "explain your reasoning step by step" for complex tasks. Improves accuracy on logical problems.
System Prompts
Design detailed system prompts that establish persona, expertise level, and behavioral guidelines.
Key Strategies
- Clarity - Be specific and unambiguous
- Examples - Provide concrete examples
- Constraints - Define output format and limits
- Role Definition - Specify expertise and perspective
- Iterative Refinement - Test and improve prompts
Common Patterns
- Classification & Categorization
- Text Generation & Creative Writing
- Code Generation & Debugging
- Data Extraction & Parsing
- Analysis & Reasoning
- Summarization & Synthesis
- Translation & Conversion
- Problem Solving & Ideation
References
- OpenAI Prompt Engineering Guide
- Anthropic Claude Best Practices
- Prompt Engineering Institute
Version History
- e0220ca Current 2026-07-05 20:41
