Prompt Generator

Overview

This skill implements the Evaluator-Optimizer workflow to create high-quality prompts through iterative improvement. Instead of generating a prompt in one shot, this approach:

1. Generates an initial optimized prompt from user requirements
2. Evaluates the generated prompt against best practices
3. Optimizes based on evaluation feedback

This three-step process produces more reliable, well-structured prompts that follow LLM best practices.

Workflow

Step 1: Generate Initial Prompt

Read generator-template.md and use it to transform the user’s requirements into an optimized prompt.

Key principles to apply:

• Define clear Persona (role)
• Provide Context (background)
• Specify Task (objective) with verb-driven instructions
• Set Constraints (what to do/not do)
• Define Format (output structure)
• Include Chain of Thought for complex tasks
• Variablize reusable elements with {Variable} syntax

Output: Present the generated prompt to the user.

Step 2: Evaluate the Generated Prompt

Read evaluator-template.md and use it to critically assess the generated prompt.

Evaluation dimensions:

• Clarity: Are instructions unambiguous?
• Structural Completeness: Does it include all key elements?
• Example Quality: Are there sufficient examples?
• Reasoning: Does it require step-by-step thinking for complex tasks?
• Constraints: Are instructions primarily positive (what to do)?

Output: Provide scores (1-5) and specific improvement suggestions in JSON format.

Step 3: Optimize Based on Evaluation

Incorporate the evaluation feedback to refine the prompt:

1. Review the suggestions from Step 2
2. Identify the highest-impact improvements
3. Revise the prompt to address the feedback
4. Present the final optimized prompt to the user

Output: The final, improved prompt with explanations of key changes made.

Usage Notes

• When to use this skill: Trigger when users want to create or improve prompts, not when they want to use an existing prompt
• Iteration: If the user requests further refinement, repeat Steps 2-3
• Customization: Adapt the templates based on the specific LLM target (Claude, GPT, Gemini) if the user specifies one
• Best practices: The templates incorporate Anthropic’s prompt engineering guidelines and industry best practices

Resources

• references/generator-template.md: Detailed template for generating optimized prompts
• references/evaluator-template.md: Comprehensive evaluation criteria and scoring rubric