Prompt Optimize Command

Command

/prompt-optimize

Description

Takes an existing LLM prompt (system, user, or tool prompt) and rewrites it for better performance. Analyzes for common anti-patterns, improves structure, reduces ambiguity, and generates optimized variants you can A/B test.

Behavior

1. Analyze the current prompt for anti-patterns and inefficiencies
2. Score against a prompt quality rubric
3. Rewrite with improved structure and clarity
4. Generate 2-3 variants for testing
5. Provide evaluation criteria for measuring improvement

Analysis Rubric

Criterion	Weight	Description
Clarity	25%	Unambiguous instructions, no conflicting rules
Specificity	20%	Concrete examples, not vague directives
Structure	20%	Logical ordering, clear sections, scannable
Efficiency	15%	Minimal tokens for maximum effect
Robustness	10%	Handles edge cases, prevents jailbreaks
Testability	10%	Success criteria are measurable

Common Anti-Patterns Detected

Copy
## Anti-Pattern: Vague Instructions
- BAD:  "Be helpful and provide good answers"
- GOOD: "Answer the user's question in 2-3 sentences. Include a code 
         example if the question is about implementation."

## Anti-Pattern: Contradicting Rules
- BAD:  "Always respond in JSON" + "For errors, explain in plain text"
- GOOD: "Always respond in JSON. For errors: {"error": "description"}"

## Anti-Pattern: Instruction Stuffing
- BAD:  30 rules in a flat list
- GOOD: Grouped rules under ## sections with priorities

## Anti-Pattern: No Examples
- BAD:  "Format output appropriately"
- GOOD: "Format output as:\n| Name | Value |\n|------|-------|\n| CPU | 45% |"

## Anti-Pattern: Missing Edge Cases
- BAD:  "Translate the text to Spanish"
- GOOD: "Translate to Spanish. If text is already Spanish, return it 
         unchanged. If text contains code blocks, preserve them."

Output Format

Copy
## Prompt Analysis

### Current Score: 52/100
| Criterion | Score | Issues |
|-----------|-------|--------|
| Clarity | 6/10 | 3 ambiguous instructions |
| Specificity | 4/10 | No examples provided |
| Structure | 5/10 | Flat list, no grouping |
| Efficiency | 7/10 | Some redundancy |
| Robustness | 4/10 | Missing edge cases |
| Testability | 3/10 | No success criteria |

### Optimized Prompt (Score: 87/100)
[Rewritten prompt here]

### Variant A: Concise (Score: 82/100)
[Shorter variant focusing on efficiency]

### Variant B: Comprehensive (Score: 85/100)
[Longer variant with more examples]

### A/B Test Plan
- **Metric**: Task completion accuracy on 50 sample inputs
- **Duration**: 100 requests per variant
- **Evaluation**: Compare output quality scores

Rules

1. Preserve intent - never change what the prompt is trying to do
2. Show reasoning - explain why each change improves the prompt
3. Measure improvements - provide before/after scores
4. Test-ready variants - outputs should be copy-paste ready
5. Token awareness - report token count changes

Examples

Copy
# Optimize a prompt from a file
/prompt-optimize src/prompts/summarizer.txt

# Optimize inline prompt
/prompt-optimize "You are a helpful coding assistant..."

# Optimize with specific focus
/prompt-optimize src/prompts/classifier.txt --focus efficiency