Skill יסודות הנדסת Prompt ל-Claude

מתי להשתמש

"How to prompt", "Prompt engineering", "Better AI results", "Prompt template".

הוראות עבודה

1. Anatomy of Great Prompt

[Role/Context] You are an expert at...
[Task] Your task is to...
[Input] Here's the input: ...
[Constraints] - Must include X, exclude Y
[Examples] Example 1: input → output
[Output Format] Respond in JSON: {...}

2. 5 Core Principles

א. Specificity

• ❌ "Write about marketing"
• ✅ "Write 300-word LinkedIn post for B2B CMOs about CAC optimization"

ב. Context

• ❌ "Improve this code"
• ✅ "I'm a junior dev. This Python function calculates user retention but is slow with >10K records. Optimize for speed, keep readable: [code]"

ג. Constraints

• "Max 200 words"
• "No bullet points"
• "Hebrew only"
• "Tone: professional but warm"

ד. Examples (Few-shot) — see separate skill

• 1-3 examples dramatically improve quality.

ה. Output Format

• "Respond in JSON: {category, summary, action_items}"
• "Use markdown headings"
• "Number each step"

3. Iterative Refinement

1. First prompt → Result.
2. Identify what's wrong/missing.
3. Add specificity / constraints.
4. Try again.
5. Repeat 3-5 times typically.

4. Common Patterns

Role Assignment

• "You are an expert [role]..."
• Boosts quality 20-40%.

Step-by-Step

• "Think step by step"
• "First analyze, then..."
• See chain-of-thought skill.

Structured Output

• JSON / YAML / Markdown table.
• Reliable parsing for workflows.

Negative Constraints

• "Do NOT include..."
• "Avoid..."

5. Prompt Length

• Too short → vague.
• Too long → diluted, costly.
• Sweet spot: 100-500 words for most tasks.
• System prompts: 200-2,000 words OK.

6. Tone Control

• "Professional", "casual", "friendly", "formal".
• "Match this style: [example]".
• "Voice of [brand]: bold, direct, no fluff".

7. Hebrew Prompts

• Work natively in Claude/GPT-4.
• Same principles apply.
• Mix Hebrew + English OK.

8. Common Mistakes

❌ Vague task → vague output. ❌ No examples for complex tasks. ❌ No output format → unreliable parsing. ❌ Conflicting instructions. ❌ Asking AI to do too much in one prompt.

9. Test Your Prompt

• Run 5-10 variations of input.
• Check consistency.
• Edge cases.
• Refine.

10. אסיים בהמלצה.

פרומפט לדוגמה

תכתוב לי prompt לסיכום מאמרים מקצועיים.

Prompt for code review. What sections to include?

Why does my prompt return inconsistent results?

© 2026 AI Expert Pro | גרסה 1.0.0