prompt-engineering-patterns

To improve LLM output reliability, apply structured prompt engineering patterns like few-shot learning, reasoning scaffolds, and system prompt design to reduce ambiguity. This Skill provides reusable templates with validation and error recovery behaviors to ensure consistent, controllable results.

The best way to structure few-shot examples is to select and format them to reduce ambiguity, boosting task accuracy. This Skill helps you construct few-shot learning examples for classification and extraction workflows by applying consistent instruction hierarchies and variable-based templates.

To add chain-of-thought reasoning scaffolds, apply techniques like self-consistency, least-to-most, and tree-of-thought prompting. This Skill implements these reasoning patterns to guide the model step-by-step, improving output quality for complex code-generation and extraction workflows.

Prompt optimization does support measuring token usage and latency. This Skill allows you to iterate on prompt variants and evaluate quality, latency, and token consumption to build modular templates optimized for production performance.

To implement self-verification and error recovery, add structured validation steps and fallback instructions to your system prompt. This Skill defines roles and constraints while supporting verification behaviors to catch failures early and reduce error rates in production.