Working with Code

Duration: 45 min

Working with Code

Duration: 45 min

Overview

Use LLMs for code generation, review, debugging, and refactoring.

Code Generation Prompt

Write a Python function that:

  • Takes a list of dicts with 'name' and 'score' keys


  • Returns top N items sorted by score descending


  • Handles: empty list, N > list length, negative scores


  • Include type hints and docstring


  • Follow PEP 8

Code Review Prompt

Review this code for bugs, performance, and security issues.
For each issue: line number, severity (critical/warning/style), and fix.
python def get_user(id): query = f"SELECT * FROM users WHERE id = {id}" result = db.execute(query) return result `

Expected findings:

  • Line 2: CRITICAL - SQL injection (use parameterized query)
  • Line 1: WARNING - no type hint, no input validation
  • Line 3: STYLE - no error handling for missing user


Debugging Prompt
python prompt = f'''This code produces an error. Explain the root cause in one sentence, then show the fix.

Code: 

{code}

Error: {error_message} 



Refactoring Prompt
Refactor for readability. Keep the same API. Explain each change.

def p(d,t=0.05,min=100):
    r=[]
    for i in d:
        if i['a']>min:
            x=i['a']*t
            r.append({'n':i['n'],'t':x,'f':i['a']-x})
    return sorted(r,key=lambda x:x['t'],reverse=True)
`

Quiz

Q1: What is the key benefit of this technique?

  • A) Reduces API cost
  • B) Improves output quality and reliability for the specific use case ✓
  • C) Makes prompts shorter
  • D) Only works with GPT-4

Q2: When should you NOT use this technique?

  • A) For production applications
  • B) For simple tasks where basic prompting already works well ✓
  • C) With Claude
  • D) On weekends

Q3: What should you always test?

  • A) Only happy paths
  • B) Edge cases, failures, and adversarial inputs ✓
  • C) Nothing - trust the model
  • D) Only the system prompt

← Back ▶ Run in Colab