learn-with-santosh Prompt Optimizer 💎
Turn your basic instructions into high-performance prompts. This skill uses advanced engineering techniques (Chain-of-Thought, Delimiters, Few-Shot, and Persona) to ensure your AI interactions are consistent, accurate, and powerful.
When to use this skill
Use this skill when:
- You have a simple prompt that is producing vague or incorrect results.
- You want to turn a task description into a reusable system prompt.
- You need to add constraints or structural requirements to an AI output.
- You want to "AI-ify" a business process.
The Optimization Framework
To optimize a prompt, Claude applies these layers:
1. The Persona Layer
Define who Claude should be. "Instead of 'Write a script', use 'You are a Senior Python Developer with 10 years of experience in automation...'"
2. The Objective Layer
Clarify exactly what the goal is. "Define the 'Primary Mission' and the 'Success Criteria' for the task."
3. The Constraint Layer
Add the "Boundaries."
- Tone (Professional, Humorous, Academic)
- Length (Max 200 words, exactly 5 bullet points)
- Format (JSON, Markdown, CSV)
4. The Engineering Layer
Add advanced logic:
- Chain-of-Thought: Add
<thought>blocks. - Delimiters: Use tags like
<context>or<instructions>. - Few-Shot: Request specific examples to be added.
Output Format
🛠 Analysis
What was wrong with the original prompt?
✨ The Optimized Prompt
The ready-to-use, engineered version.
💡 Why it's better
Explanation of the techniques applied.
Examples
Example: Transforming a simple request
Original: "Help me write a blog post about AI." Optimized: "Persona: You are a Tech Journalist and SEO Expert. Objective: Write a 1,000-word deep-dive article about the impact of Claude Code on developer productivity. Instructions:
- Lead with a 'Pattern Interrupt' hook.
- Use H2 headers for each major section.
- Include a 'Key Takeaways' summary at the end. Constraints: Avoid buzzwords. Use a conversational but professional tone."
Best Practices
- Iterate: If the first optimization isn't perfect, tell Claude what's missing.
- Use Prompt Pro: For extremely complex prompts, use the
prompt-engineer-proskill first to define the strategy, thenprompt-optimizerto draft it. - Test: Always run the optimized prompt through its paces with different inputs.