prompt-engineering-expert_skill

Overview

This skill packages advanced prompt engineering expertise for designing, testing, and optimizing prompts and custom instructions for AI agents. It focuses on practical techniques—including structure, few-shot examples, chain-of-thought, and context management—to improve reliability and control. The guidance suits builders who need repeatable, high-quality prompt workflows for production or research.

How this skill works

The skill inspects existing prompts and agent instructions, identifies weaknesses, and recommends concrete rewrites and templates. It applies evaluation criteria and test cases to measure performance, suggests token- and context-optimization tactics, and provides patterns for role-based and multimodal prompting. Outputs include improved prompt variants, success metrics, and reusable templates.

When to use it

• When a prompt returns inconsistent or low-quality answers
• Designing system prompts and custom instructions for new agents
• Scaling prompts across domains or teams with reusable templates
• Optimizing token use for large or multimodal context windows
• Creating evaluation tests and success criteria for prompt changes

Best practices

• Write clear, goal-oriented prompts with explicit output format and constraints
• Use few-shot examples and role framing for complex tasks
• Maintain concise system instructions separate from user-facing text
• Iteratively test changes with small A/B style experiments and metrics
• Monitor token usage and prioritize salient context to avoid window overflow

Example use cases

• Refine a vague customer-support prompt to produce consistent step-by-step troubleshooting
• Create a domain-specific system prompt for legal or medical draft generation
• Convert a free-form workflow into a structured template with XML-like tags for parsable outputs
• Design test cases to verify hallucination reduction after prompt changes
• Optimize prompts for multimodal inputs by specifying how to reference images and embeddings