/optimize-prompt Command

Metadata

name: optimize-prompt
version: 1.0.0
category: orchestration
status: active
priority: P0
derived_from: Claude Operating Preferences v6.0

Description

Optimize prompts for Claude Opus 4.5 by replacing aggressive language that may cause overtriggering. Converts imperatives like "MUST", "ALWAYS", "NEVER" to gentler alternatives that produce better results with Opus 4.5.

Usage

# Optimize a prompt string
/optimize-prompt "You MUST ALWAYS follow these rules EXACTLY"

# Optimize from file
/optimize-prompt --file prompts/system-prompt.txt

# Optimize with thinking-disabled mode
/optimize-prompt --no-thinking "Think carefully about each step"

# Show optimization report
/optimize-prompt --report "CRITICAL: NEVER skip validation"

# Batch optimize directory
/optimize-prompt --batch prompts/

# Preview changes without applying
/optimize-prompt --dry-run --file system-prompt.md

Arguments

Argument	Type	Default	Description
prompt	string	-	Prompt text to optimize (positional)
--file	path	-	Read prompt from file
--no-thinking	flag	false	Also replace "think" words (for when thinking is disabled)
--report	flag	false	Show detailed optimization report
--batch	path	-	Optimize all .md/.txt files in directory
--dry-run	flag	false	Preview changes without modifying files
--output	path	-	Write optimized prompt to file

Replacement Rules

Aggressive Language

Original	Replacement	Reason
MUST	Please	Imperative too strong
ALWAYS	Prefer to	Absolute may overtrigger
NEVER	Avoid	Absolute may overtrigger
CRITICAL	Important	Too urgent
ABSOLUTELY	(removed)	Unnecessary emphasis
REQUIRED	Recommended	Softer obligation
MANDATORY	Suggested	Softer obligation
VERY IMPORTANT	Note	Less urgent
EXTREMELY	Quite	Softer emphasis
ESSENTIAL	Helpful	Less demanding
CRUCIAL	Useful	Less demanding

Thinking Words (when --no-thinking)

Original	Replacement
think about	consider
think through	work through
think carefully	review carefully
think step by step	proceed step by step

Output Examples

Basic Optimization

$ /optimize-prompt "You MUST ALWAYS validate input. NEVER trust user data."

Original:
You MUST ALWAYS validate input. NEVER trust user data.

Optimized:
You Please Prefer to validate input. Avoid trust user data.

Changes: 3 replacements made

With Report

$ /optimize-prompt --report "CRITICAL: You MUST NEVER skip the validation step. This is EXTREMELY IMPORTANT!"

Optimization Report
-------------------
Original length:  72 characters
Optimized length: 62 characters
Reduction:        13.9%

Changes Made:
  1. CRITICAL → Important (Aggressive language)
  2. MUST → Please (Imperative too strong)
  3. NEVER → Avoid (Absolute may overtrigger)
  4. EXTREMELY → Quite (Softer emphasis)
  5. IMPORTANT → helpful (Less demanding)

Original:
CRITICAL: You MUST NEVER skip the validation step. This is EXTREMELY IMPORTANT!

Optimized:
Important: You Please Avoid skip the validation step. This is Quite helpful!

File Processing

$ /optimize-prompt --file system-prompt.md --dry-run

Processing: system-prompt.md
Lines: 45
Size: 2.3KB

Would make 12 replacements:
  - Line 3:  MUST → Please
  - Line 7:  ALWAYS → Prefer to
  - Line 12: NEVER → Avoid
  - Line 15: CRITICAL → Important
  ...

Use without --dry-run to apply changes.

System Prompt

When this command is invoked:

1. Parse input - Get prompt from argument, file, or stdin
2. Apply replacements - Use Opus45PromptOptimizer patterns
3. If --no-thinking - Also replace "think" words
4. If --report - Generate detailed change report
5. Output result - Print optimized prompt or write to file

Implementation

#!/usr/bin/env python3
"""optimize-prompt command implementation"""

import argparse
import re
from pathlib import Path

AGGRESSIVE_REPLACEMENTS = {
    "MUST": "Please",
    "ALWAYS": "Prefer to",
    "NEVER": "Avoid",
    "CRITICAL": "Important",
    "ABSOLUTELY": "",
    "REQUIRED": "Recommended",
    "MANDATORY": "Suggested",
    "VERY IMPORTANT": "Note",
    "EXTREMELY": "Quite",
    "ESSENTIAL": "Helpful",
    "CRUCIAL": "Useful",
}

THINK_REPLACEMENTS = {
    "think about": "consider",
    "think through": "work through",
    "think carefully": "review carefully",
    "think step by step": "proceed step by step",
}

def optimize_prompt(text: str, no_thinking: bool = False) -> tuple[str, list]:
    """Optimize prompt and return (result, changes)"""
    result = text
    changes = []

    for original, replacement in AGGRESSIVE_REPLACEMENTS.items():
        pattern = re.compile(re.escape(original), re.IGNORECASE)
        if pattern.search(result):
            changes.append((original, replacement, "Aggressive language"))
            result = pattern.sub(replacement, result)

    if no_thinking:
        for original, replacement in THINK_REPLACEMENTS.items():
            pattern = re.compile(re.escape(original), re.IGNORECASE)
            if pattern.search(result):
                changes.append((original, replacement, "Thinking word"))
                result = pattern.sub(replacement, result)

    # Remove excessive exclamation marks
    if re.search(r'!{2,}', result):
        changes.append(("!!", "!", "Excessive punctuation"))
        result = re.sub(r'!{2,}', '!', result)

    return result.strip(), changes

def print_report(original: str, optimized: str, changes: list):
    """Print detailed optimization report"""
    print("\nOptimization Report")
    print("-" * 40)
    print(f"Original length:  {len(original)} characters")
    print(f"Optimized length: {len(optimized)} characters")

    reduction = (1 - len(optimized) / len(original)) * 100 if original else 0
    print(f"Reduction:        {reduction:.1f}%")

    print(f"\nChanges Made ({len(changes)}):")
    for i, (orig, repl, reason) in enumerate(changes, 1):
        print(f"  {i}. {orig} → {repl} ({reason})")

    print(f"\nOriginal:\n{original}")
    print(f"\nOptimized:\n{optimized}")

def main():
    parser = argparse.ArgumentParser(description="Optimize prompts for Opus 4.5")
    parser.add_argument("prompt", nargs="?", help="Prompt to optimize")
    parser.add_argument("--file", type=Path, help="Read from file")
    parser.add_argument("--no-thinking", action="store_true")
    parser.add_argument("--report", action="store_true")
    parser.add_argument("--batch", type=Path)
    parser.add_argument("--dry-run", action="store_true")
    parser.add_argument("--output", type=Path)

    args = parser.parse_args()

    # Get input text
    if args.file:
        text = args.file.read_text()
    elif args.prompt:
        text = args.prompt
    else:
        print("Error: Provide prompt or --file")
        return

    # Optimize
    optimized, changes = optimize_prompt(text, args.no_thinking)

    # Output
    if args.report:
        print_report(text, optimized, changes)
    else:
        print(optimized)

    if args.output and not args.dry_run:
        args.output.write_text(optimized)
        print(f"\nWritten to: {args.output}")

if __name__ == "__main__":
    main()

Related Components

• skills/opus-optimization/ - Core optimization patterns
• agents/prompt-optimizer-opus45.md - Agent for batch optimization
• scripts/prompt-optimizer.py - Batch processing script

---

Action Policy

<default_behavior> This command optimizes prompts automatically. Proceeds with:

• Text replacement based on rules
• Report generation
• File output if specified

Provides before/after comparison. </default_behavior>

After execution, verify: - Replacements applied correctly - Original meaning preserved - Output file written (if specified)

Success Output

When optimize-prompt completes:

✅ COMMAND COMPLETE: /optimize-prompt
Input: <source>
Replacements: <N>
Output: <optimized text or file>
Reduction: <N>%

Completion Checklist

Before marking complete:

• Input parsed
• Replacements applied
• Report generated (if --report)
• Output written (if --output)

Failure Indicators

This command has FAILED if:

• ❌ No input provided
• ❌ File not found
• ❌ Write failed
• ❌ Invalid options

When NOT to Use

Do NOT use when:

• Prompt already optimized
• Need aggressive language
• Non-Opus target model

Anti-Patterns (Avoid)

Anti-Pattern	Problem	Solution
Over-optimize	Loses emphasis	Review output
Skip --report	Miss changes	Use --report first
Ignore context	Wrong replacements	Check semantic fit

Principles

This command embodies:

• #3 Complete Execution - Full optimization
• #6 Clear, Understandable - Clear report
• #9 Based on Facts - Rule-based changes

Full Standard: CODITECT-STANDARD-AUTOMATION.md