跳转至

Prompt Engineering Best Practices | 提示词工程最佳实践

The Golden Rule Garbage In, Garbage Out. If you give the AI a vague request, you get a vague answer. 黄金法则 垃圾进，垃圾出。 如果你给 AI 一个模糊的请求，你就会得到一个模糊的答案。

1. The "CO-STAR" Framework | CO-STAR 框架

A proven formula to write perfect prompts every time. 一个经过验证的完美提示词公式。

  1. C (Context): What is the background? (我是谁？我在做什么？)
  2. O (Objective): What is the task? (你要做什么？)
  3. S (Style): Who should the AI act like? (像鲁迅？像乔布斯？)
  4. T (Tone): What is the emotional tone? (正式？幽默？激进？)
  5. A (Audience): Who is reading this? (小学生？博士？)
  6. R (Response): What is the output format? (JSON? Markdown? Table?)

2. Before & After: Real Examples | 实战对比

Scenario 1: Coding (写代码)

❌ Bad Prompt:

"Write a Python script to scrape a website." (Result: Generic code, probably breaks, no error handling.)

✅ Good Prompt:

Context: I am building a data pipeline to track prices on an e-commerce site. Objective: Write a Python script using BeautifulSoup and requests. Constraints: - Handle HTTP 403/404 errors with retries. - Add comments explaining each step. - Do NOT use Selenium (I need speed). Format: Return a single code block with a main() function.

Scenario 2: Writing Email (写邮件)

❌ Bad Prompt:

"Write an email to my boss asking for a raise." (Result: Generic, maybe too aggressive or too weak.)

✅ Good Prompt:

Context: I have successfully led the "Project X" which saved the company $50k last quarter. Objective: Write an email to my manager requesting a salary review. Tone: Professional, confident, but polite. Not demanding. Audience: My manager is busy and likes concise emails. Format: Subject line + Body. Keep it under 200 words.

3. The "Mega-Prompt" Template | “超级提示词”模板

Copy and paste this structure for complex tasks. 复制粘贴此结构用于复杂任务。

# Role
You are an expert [Role Name], specializing in [Skill 1] and [Skill 2].

# Context
I am currently [Current Situation]. My goal is to [Goal].

# Task
Please [Action Verb] a [Output Type] that [Specific Requirement].

# Constraints
- The output must be in [Language].
- Do not use [Forbidden Thing].
- Ensure [Quality Standard].

# Output Format
Please provide the result in [Format, e.g., Markdown Table, JSON].

# Example (Few-Shot)
Input: [Example Input]
Output: [Example Output]

4. Pro Tip: "Iterative Refinement" | 专家技巧：“迭代精炼”

Don't expect perfection on the first try. Treat the AI like an intern. 别指望一次就完美。把 AI 当作实习生。

  1. Prompt: "Write code."
  2. AI: (Gives buggy code).
  3. Refine: "Good, but it fails on large files. Optimize it for memory."
  4. AI: (Gives better code).
  5. Refine: "Now add docstrings."

5. Practice Mission: The Socrates Challenge | 练习任务：苏格拉底挑战

Objective (目标): Write a prompt that forces the AI to teach you, not just tell you. 目标：编写一个提示词，强迫 AI 教会你，而不仅仅是告诉你答案。

Scenario (场景): You are a student trying to understand "Quantum Physics" (or any topic you like). If the AI just gives you a lecture, you will fall asleep. You want the AI to act like Socrates (苏格拉底). 你是想学习“量子物理”（或任何你喜欢的主题）的学生。 如果 AI 只是给你讲课，你会睡着。你希望 AI 扮演苏格拉底。

Task (任务): Write a prompt using the CO-STAR framework. 1. Role: Socrates (The philosopher who asks questions). 2. Objective: Help me understand [Topic] by asking me guiding questions. 3. Constraint: Do NOT give me the answer directly. Wait for my response after each question.

Try it on ChatGPT/Claude:

Paste your prompt and see if it works.

Example Answer (参考答案): 

# Role
You are Socrates, the wise philosopher.

# Task
I want to understand "How Gravity Works".
Instead of explaining it to me, ask me a series of simple questions to help me derive the answer myself.

# Constraints
- Ask only ONE question at a time.
- Wait for my answer before moving on.
- If I am wrong, gently guide me with a hint, but don't give the answer.
- Use a curious and humble tone.