prompt-optimizer

name	prompt-optimizer
description	Use when the user wants to improve prompt effectiveness, apply best practices from Anthropic's research, or transform a basic prompt into a high-performance structured prompt using XML tags, examples, chain of thought, role prompting, and other proven techniques.
version	1.0.0
author	Abe Diaz (@abe238)
license	MIT
metadata	{"hermes":{"tags":["prompting","prompt-engineering","claude","optimization","xml","chain-of-thought","examples"],"related_skills":["adversarial-prompting","rapid-domain-mastery"]}}

Prompt Optimizer

Overview

Transform basic prompts into high-performance, structured prompts using proven techniques from Anthropic's research. Apply the compound effect of multiple techniques (XML structure + examples + chain of thought + role prompting) for multiplicative quality improvements of up to 95% over baseline prompts.

When to Use This Skill

Use this skill when:

• Users provide a basic prompt that needs optimization
• Output quality or consistency is poor with current prompts
• Format requirements are inconsistent or unclear
• Complex reasoning tasks need better structure
• Domain expertise needs to be incorporated
• Multi-step workflows require better organization
• Users want to apply Claude 4.5 best practices

Three-Pass Optimization Method

Pass 1: Analysis and Coverage

Analyze the original prompt and apply core techniques focusing on coverage and structure:

1. Clarity Assessment: Apply "The Golden Rule" - would a colleague understand this prompt without context?

2. Core Technique Selection: Apply Anthropic's priority order based on prompt needs:

   • Be Clear & Direct (always) - Add context, audience, workflow, success criteria
   • Use Examples (format consistency) - Add 3-5 diverse, relevant examples when output format matters
   • Let Claude Think (complex reasoning) - Add structured chain of thought for multi-step problems
   • Use XML Tags (multi-part prompts) - Structure with cognitive containers for organization
   • Give Claude a Role (domain expertise) - Add specific expert persona when specialized knowledge needed
   • Prefill Response (output control) - Guide response format for specific requirements
   • Chain Prompts (multi-step workflows) - Break into sequential steps for complex dependent tasks

3. Advanced Strategy Integration: Consider metacognitive prompting, constraint satisfaction, effort control

Pass 2: Structure and Flow Review

Improve clarity, flow, and evidence while marking uncertain areas:

1. XML Organization: Structure with meaningful cognitive containers that separate different prompt parts
2. Example Quality: Ensure examples are relevant, diverse, clear, and progressive - covering edge cases
3. Chain of Thought: Add structured thinking for complex tasks (thinking → alternatives → evaluation → answer)
4. Context Layering: Organize immediate, background, and success contexts clearly
5. Constraint Hierarchy: Define hard constraints (must satisfy), soft constraints (preferred), anti-constraints (avoid)
6. Flow Optimization: Ensure logical progression and clear connections between sections
7. Mark Uncertainties: Flag areas that need verification with [VERIFY] tags

Pass 3: Final Polish and Optimization

Apply Claude 4.5 optimizations and quality assurance:

1. Claude 4.5 Enhancements:

   • Add explicit instructions and context explanations (why, not just what)
   • Enable parallel tool calling where appropriate
   • Prevent overengineering with focused constraints

2. Quality Assurance: Apply validation checklist using quality_checker.py

3. Token Efficiency: Optimize for performance without losing quality

4. Error Prevention: Add hallucination guards and uncertainty tracking

5. Output Format Control: Reduce markdown overuse, ensure clear prose

6. Remove [VERIFY] tags: After fact-checking and validation

Optimization Framework

Use this systematic approach for consistent results:

<prompt_analysis>
1. <original_prompt>: [User's original prompt]
2. <clarity_issues>: [What's unclear or missing - context, audience, success criteria]
3. <technique_recommendations>: [Which core techniques to apply based on task type]
4. <complexity_assessment>: [Simple/Medium/Complex reasoning required]
5. <domain_expertise_needed>: [What specialized knowledge is required]
</prompt_analysis>

<optimization_plan>
1. <structure_improvements>: [XML organization needed for cognitive containers]
2. <examples_needed>: [What examples would help - format, edge cases, variations]
3. <thinking_requirements>: [Chain of thought structure for reasoning]
4. <constraints_to_add>: [Hard/soft/anti-constraints for clarity]
5. <claude45_enhancements>: [Specific 4.5 optimizations - explicit instructions, context explanations]
</optimization_plan>

<optimized_prompt>
[Final optimized prompt with all techniques applied systematically]
</optimized_prompt>

Quality Validation

After optimization, validate using the built-in quality checker:

python scripts/quality_checker.py "your optimized prompt here"

This checks for:

• Clear task definition
• Sufficient context
• Relevant examples
• XML structure
• Defined constraints
• Output format specification
• Ambiguity avoidance
• Appropriate complexity matching

Resources

references/

• core_techniques.md - Detailed reference for all 7 core prompting techniques with examples
• advanced_strategies.md - Metacognitive prompting, constraint satisfaction, optimization patterns
• claude45_optimizations.md - Claude 4.5 specific enhancements and behavioral changes
• templates.md - Ready-to-use templates for common prompt patterns

scripts/

• prompt_analyzer.py - Automated prompt analysis and technique recommendation script
• quality_checker.py - Validation script for optimized prompts

Load references as needed to inform optimization decisions. Use scripts for automated analysis when helpful.