prompt-engineering-patterns

To structure an LLM system prompt for reliable JSON output, define clear role-based instructions and specify the exact output formatting required. This Skill provides patterns for generating structured JSON, XML, or tool calls to ensure robust downstream processing.

Preventing prompt injection attacks requires implementing specific defense patterns within your LLM prompts. This Skill provides comprehensive guidance on securing system prompts against malicious manipulation and ensuring robust interactions.

Chain of thought prompting improves complex LLM tasks by implementing step-by-step reasoning and self-consistency techniques. This Skill covers reasoning patterns like CoT and ReAct to enhance model performance on difficult operations.

You can optimize LLM API costs through prompt engineering by applying model routing and prompt caching strategies. This Skill addresses cost optimization techniques alongside context window management using sliding windows and summarization.

Managing the context window when designing prompts involves optimizing prompt length and content through techniques like sliding windows and summarization. This Skill provides patterns for context management to maintain optimal LLM performance.

Yes, you can use few-shot examples to improve LLM output accuracy by carefully selecting examples within your prompt. This Skill covers few-shot example selection techniques to guide the model toward desired responses.