meta-prompt-engineering_skill

Overview

This skill transforms vague or unreliable prompts into structured, constraint-aware prompts that produce consistent, high-quality outputs. It provides a repeatable framework for defining roles, decomposing tasks, enforcing constraints, and building quality checks to reduce hallucination and variability. Use it to create reusable prompt templates and safety guardrails for production workflows.

How this skill works

The skill inspects an existing prompt, identifies failure modes (vagueness, missing constraints, format drift), and generates an engineered prompt with explicit role, step-by-step task decomposition, and format requirements. It adds length/tone/content constraints and a self-evaluation checklist, then recommends tests and iterations to measure consistency and fix recurring issues. The result is a prompt designed for reliable, auditable outputs and easier reuse.

When to use it

• Outputs vary across runs or quality is unpredictable
• Tasks require multi-step reasoning or clear subtask breakdowns
• You need strict output formats (JSON, tables, sections) or length limits
• Safety, legal, or domain-specific constraints must be enforced
• You are building prompt templates for production or reuse

Best practices

• Define a clear role/persona and precise success criteria up front
• Decompose complex tasks into numbered steps with explicit deliverables
• Specify format, length, tone, and prohibited content as constraints
• Include self-checks (e.g., 'If unsure, say "I don't know"') and evaluation criteria
• Run multiple iterations and test edge cases; refine for common failure modes

Example use cases

• Convert a vague content request into a reusable blog-post template with word limits, headings, and references
• Create a structured reviewer prompt for security audits that lists checks and format for findings
• Design a medical-answer prompt that enforces citation, uncertainty expression, and content restrictions
• Build an API-facing prompt that outputs strict JSON with keys validated by a quality checklist
• Optimize prompts for a research assistant to decompose multi-step analyses and summarize results