Meta Prompt

What it does

The Meta Prompt skill provides tools to improve AI response quality through evaluation and analysis of solution paths. It offers a framework for critiquing responses, assigning scores (0-10), and determining if a complete solution has been provided. Additionally, it guides the analysis of the reasoning steps taken by an agent towards answering questions or completing tasks.

When to use it

• When you need to assess the quality of an AI's response to ensure accuracy and completeness.
• To understand how an AI arrived at a solution, identifying potential areas for improvement in its reasoning process.
• When debugging an agent’s problem-solving approach by examining its observations, thoughts, and actions.
• For evaluating the clarity and conciseness of generated text.

Key capabilities

• Response Quality Evaluator: Provides a structured framework for assessing AI responses with fields for reflections, score (0-10), and solution status (found_solution).
• Evaluation Criteria: Offers specific criteria to guide response evaluation: accuracy, completeness, clarity, conciseness, and relevance.
• Question-Answering Trajectory Analyzer: Enables analysis of the steps taken by an AI agent, including observations, thoughts, and actions.
• Action Types: Defines specific action types like Search[entity] and Lookup[keyword] for analyzing agent behavior.

Example prompts

• "Evaluate this response: [AI Response]"
• "Analyze the trajectory of how you arrived at this answer."
• "Critique this solution based on accuracy, completeness, and clarity."

Tips & gotchas

• The skill is designed for evaluating existing AI responses or analyzing completed task trajectories. It doesn't directly generate content itself.
• Pay close attention to the evaluation criteria when providing feedback; thoughtful reflections are encouraged.
• found_solution should only be set to true if the response fully addresses the question or completes the task.