/optimize-prompt

Optimize Prompt for Token Efficiency

Takes an input prompt and returns ONLY a token-optimized version that preserves meaning while minimizing token count. Based on LLM tokenization principles: common words tokenize more efficiently, unusual words break into more tokens, and conciseness reduces cost.

**Output Format**: Return only the optimized prompt text with no additional commentary, analysis, or explanation.

Instructions

1. **Analyze Input Prompt**

• Extract the prompt from `$ARGUMENTS` using the `--prompt` flag (format: `--prompt "your prompt text"`)
• Identify current token inefficiencies:
• Verbose or redundant phrasing
• Unusual/rare words that break into multiple tokens
• Unnecessary qualifiers and filler words
• Repetitive statements
• Overly formal language when casual works
• Estimate current token count based on word complexity and frequency

2. **Apply Token Optimization Techniques**

Apply these optimization strategies systematically:

**Conciseness**

• Remove filler words (just, really, very, actually, basically)
• Eliminate redundant phrases
• Use active voice instead of passive
• Combine related ideas into single statements

**Word Choice**

• Replace rare/unusual words with common alternatives
• Use shorter synonyms when meaning is preserved
• Prefer common technical terms over obscure ones
• Avoid unnecessarily complex vocabulary

**Structure**

• Remove unnecessary introductions/conclusions
• Use direct imperatives instead of questions
• Eliminate meta-commentary about the prompt itself
• Consolidate multi-sentence ideas when possible

**Examples**

• Before: "Could you please help me understand how to..."
• After: "Explain how to..."

• Before: "I would really appreciate it if you could analyze..."
• After: "Analyze..."

• Before: "utilizing sophisticated methodology"
• After: "using advanced methods"

3. **Preserve Critical Elements**

While optimizing, ensure you maintain:

• **Core meaning and intent** - Don't change what's being asked
• **Essential context** - Keep details needed for accurate response
• **Technical specificity** - Preserve domain-specific terms when needed
• **Clarity** - Optimized version must be unambiguous

4. **Return Optimized Prompt**

Return ONLY the optimized prompt text with no analysis, summary, or explanation.

Do not include:

• Analysis of inefficiencies
• Optimization summary
• Changes made
• Token savings estimates
• Key improvements
• Any other commentary

Simply return the token-optimized version of the prompt.

5. **Validation**

Before returning, verify the optimized prompt:

• ✅ Preserves original intent and meaning
• ✅ Maintains necessary context and details
• ✅ Uses more common/efficient word choices
• ✅ Eliminates redundancy and verbosity
• ✅ Remains clear and unambiguous

**IMPORTANT**: After validation, return ONLY the optimized prompt text. Do not include this checklist, analysis, or any explanatory text in your response.

Token Optimization Principles

Based on how LLM tokenizers work:

1. **Common words = fewer tokens** - Frequent words in training data tokenize more efficiently 2. **Rare words = more tokens** - Unusual words break into multiple sub-word tokens 3. **Languages matter** - Less common languages/dialects use more tokens 4. **Code languages differ** - JavaScript tokenizes more efficiently than Haskell 5. **Token count = cost** - Every token in input and output costs money

Usage

/dev:optimize-prompt --prompt "Please could you help me to understand the various different methodologies and approaches that could potentially be utilized when implementing a comprehensive authentication system"

Returns:

Explain methods for implementing authentication systems

Example 2:

/dev:optimize-prompt --prompt "I need you to create commits and branches that culminate into a sensible release. I also need you to add model:inherit to agent files that do not have a model decleared in the frontmatter yaml"

Returns:

Create commits and branches for release. Add model:inherit to agent frontmatter missing model declaration.

Notes

• This command focuses on reducing token count while preserving meaning
• For creative writing, may want to preserve style over efficiency
• Technical prompts often benefit most from optimization
• Test optimized prompts to ensure they produce desired results