Prompt Optimizer

Overview

Evaluate prompt quality, provide targeted improvement suggestions, and generate optimized versions using 58 proven prompting techniques. This skill systematically analyzes prompts across multiple quality dimensions and applies evidence-based optimization patterns.

Quick Start

For most optimization tasks, follow this workflow:

1. Analyze the current prompt – Read and understand what the user wants to achieve
2. Evaluate quality – Assess across clarity, specificity, structure, completeness
3. Load relevant techniques – Read references/prompt-techniques.md for applicable methods
4. Generate suggestions – Use evaluation results and techniques to propose improvements
5. Create optimized version – Apply chosen techniques to produce an enhanced prompt

Evaluation Workflow

When a user asks to optimize or evaluate a prompt:

Step 1: Load Quality Framework

Read references/quality-framework.md to understand evaluation dimensions:

• Clarity – Is the prompt unambiguous and easy to understand?
• Specificity – Are requirements and constraints clearly defined?
• Structure – Does it follow logical organization?
• Completeness – Does it include all necessary context and instructions?
• Tone – Is the voice appropriate for the task?
• Constraints – Are boundaries and limitations clear?

Step 2: Perform Quality Assessment

Evaluate the prompt against each dimension:

For each quality dimension:
1. Identify strengths (what works well)
2. Identify weaknesses (what's missing or unclear)
3. Rate quality (Poor/Fair/Good/Excellent)
4. Note specific improvement opportunities

Step 3: Identify Applicable Techniques

Load references/prompt-techniques.md and identify techniques that address the identified weaknesses.

Example mapping:

• Weak: “Be creative” → Apply: Role-play or Creative Persona
• Weak: “Write an essay” → Apply: Chain of Thought or Step-by-Step
• Weak: “Summarize this” → Apply: Few-shot Learning with examples

Step 4: Generate Optimization Plan

Create a structured optimization plan:

1. Priority improvements – High-impact changes that address multiple weaknesses
2. Optional enhancements – Nice-to-have techniques that boost performance
3. Technique combinations – Suggest technique pairings for specific use cases

Step 5: Generate Optimized Prompt

Apply the selected techniques to create an improved version:

• Preserve original intent and requirements
• Add structure and clarity where missing
• Embed examples, constraints, or guidance as needed
• Maintain appropriate tone and voice

Optimization Patterns

For common optimization scenarios, use these proven patterns:

Ambiguous Requests → Structured Breakdown

When prompt lacks clarity:

1. Add explicit task definition
2. Break into sub-tasks with numbered steps
3. Include output format specification
4. Add completion criteria

Generic Tasks → Technique Enhancement

When prompt is too broad:

1. Apply relevant technique from references/prompt-techniques.md
2. Add examples (few-shot) or reasoning steps (CoT)
3. Include role or persona guidance
4. Specify evaluation criteria

Missing Context → Scenario Framing

When prompt lacks background:

1. Add user intent/goal statement
2. Include target audience specification
3. Define success metrics
4. Add relevant constraints or boundaries

Weak Instructions → Actionable Steps

When prompt provides vague guidance:

1. Convert abstract concepts to concrete actions
2. Add step-by-step instructions
3. Include quality checkpoints
4. Specify expected output format

Script Usage

Quality Evaluation

For consistent, repeatable evaluation:

python3 scripts/evaluate.py "Your prompt here"

This provides:

• Dimension scores (clarity, specificity, structure, completeness)
• Overall quality rating
• Detailed weakness analysis
• Suggested improvement areas

Prompt Optimization

For automatic optimization generation:

python3 scripts/optimize.py "Your prompt here" --techniques "few-shot,coT"

This generates:

• Multiple optimized prompt versions
• Explanation of applied techniques
• Comparison with original prompt

Note: Scripts should be used for automation or when you need deterministic results. For complex optimization tasks, use the manual workflow for more nuanced analysis.

Reference Files

references/prompt-techniques.md

Complete catalog of 58 prompting techniques including:

• Reasoning techniques (CoT, Tree of Thoughts, Decomposition)
• Context techniques (Few-shot, Self-Consistency, Reflection)
• Creative techniques (Role-play, Scenario, Persona)
• Structural techniques (Template, Framework, Checklists)
• And 50+ more with usage examples

Load this when you need to identify applicable techniques for a specific optimization task.

references/quality-framework.md

Detailed evaluation framework with:

• Dimension-specific criteria and rubrics
• Scoring guidelines
• Common anti-patterns to avoid
• Quality benchmarks for different prompt types

Load this before any evaluation task to ensure consistent assessment.

references/optimization-patterns.md

Collection of proven optimization patterns including:

• Pattern → Technique mappings
• Before/after examples
• Technique combination guidelines
• Use-case specific templates

Load this when optimizing common prompt types (essays, code generation, analysis, etc.).

Best Practices

1. Preserve user intent – Never change what the user wants, only how they ask for it
2. Add incrementally – Apply one technique at a time and evaluate impact
3. Test iteratively – After optimization, test the prompt and refine further if needed
4. Document choices – Explain which techniques you applied and why
5. Provide options – Offer multiple optimization versions when appropriate

When This Skill Should Trigger

This skill should be activated when:

• User explicitly asks to “optimize,” “improve,” or “evaluate” a prompt
• User asks if a prompt is “good” or “clear”
• User wants to “fix” or “enhance” a prompt that isn’t working well
• User requests “better versions” of a prompt
• User asks about prompt engineering techniques or best practices
• User wants to analyze why a prompt is producing poor results