prompt-engineer

Guide the creation, testing, and refinement of LLM prompts with structured workflows.

3|2|Updated Jun 9, 2026

Search Tags:#chain-of-thought #evaluation #rubric #few-shot #prompt-engineering #structured-output #system-prompt

Ready to run

npx skills add https://github.com/JayRHa/AgentSkills --skill prompt-engineer-jayrha

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: prompt-engineer
Source: https://github.com/JayRHa/AgentSkills/tree/main/prompt-engineer
Command: npx skills add https://github.com/JayRHa/AgentSkills --skill prompt-engineer-jayrha

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Designs, refines, and systematically evaluates prompts for LLMs using structure, role framing, few-shot examples, explicit output contracts, and reasoning scaffolds. It also provides a rigorous evaluation approach with golden sets, rubrics, and baselines to compare prompt variants.

Core Features & Use Cases

Clear prompt anatomy and system framing
Few-shot example selection, construction, and evaluation
Enforce JSON/structured output via output contracts
LLM-as-judge rubrics and evaluation frameworks
Pattern-driven workflow: extraction, classification, generation, transformation
Debug prompts, reduce hallucinations and refusals, and compare variants

Quick Start

Install and run the evaluation harness to compare prompt variants against a golden set.

Frequently Asked Questions about prompt-engineer

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I evaluate and compare LLM prompt variants against a golden set?▼

To evaluate prompt variants against a golden set, use LLM-as-judge rubrics and baseline comparisons. This approach systematically scores output accuracy across classification, extraction, and generation tasks to identify the most effective prompt structure.

What is the best way to enforce structured JSON output from large language models?▼

The best way to enforce structured JSON output is by defining explicit output contracts within your prompt. This technique constrains the model's response format, ensuring downstream parsing succeeds without unexpected formatting errors or missing data fields.

How do I write few-shot examples for prompt engineering tasks?▼

Writing few-shot examples involves selecting representative exemplars and assembling them into your prompt. This guides the model's behavior by demonstrating the desired input-output mapping, significantly improving accuracy for complex reasoning and transformation tasks.

Why does my LLM prompt produce hallucinations and refusals, and how can I debug it?▼

Hallucinations and refusals often occur due to ambiguous system prompts or lack of reasoning scaffolds. You can debug these issues by applying a pattern-driven workflow: clarify the job, draft canonical sections, add output contracts, and run automated evaluation.

When do I need a chain-of-thought reasoning scaffold for my LLM prompts?▼

You need a chain-of-thought reasoning scaffold when tackling complex extraction, classification, or generation tasks. This mechanism forces the model to articulate intermediate reasoning steps, drastically reducing logical errors and improving overall output reliability.

Can I use an automated evaluation harness to debug system prompts for extraction tasks?▼

Yes, you can use an automated evaluation harness to debug system prompts for extraction tasks. By running your prompts against a golden set with defined rubrics, you isolate failure points and systematically refine the prompt pattern.