An "instruction" to Claude is often called a prompt. How good the answer turns out largely depends on how well the prompt is written - whether in Claude.ai or in Claude Code. This chapter shows you the key principles that will accompany you throughout the rest of this series.

Principle 1: clear, specific instructions

The more specific the instruction, the more targeted the answer. "Make this better" leaves a lot of room for interpretation; "Shorten this paragraph to three sentences and keep the formal tone" leaves almost no doubt. That applies to text just as much as to code: "Improve the function" is vague, "Change the function calculateSum so it handles an empty list without throwing an error" is precise.

Principle 2: give context

Claude knows neither your situation nor your history unless you tell it. Instead of "Write me an email", "Write a short, friendly email to a customer asking about a delayed delivery - we expect the goods in three days" provides all the information needed for a fitting answer. In Claude Code, giving context often means naming relevant files or error messages: "In utils.py, the function parse_date throws an error on an empty string - here's the exact error message: ...".

Principle 3: give concrete examples

An example often says more than a long description. If you want a specific output format, show it: "Format each result like this: Name - Price - Stock, for example Apple - $0.89 - 42." This drastically reduces follow-up questions and misunderstandings, because the desired result no longer needs to be interpreted, only matched.

Good vs. bad prompts compared

A few side-by-side comparisons that make the principle tangible:

• Bad: "Fix the bug." — Better: "Nothing happens when clicking the Save button on the contact page, but a success message should appear. Take a look at ContactForm.jsx."
• Bad: "Make the website better." — Better: "The mobile navigation overlaps with the logo below 400px width. Fix that without changing the desktop layout."
• Bad: "Write tests." — Better: "Write tests for the function validateEmail in validators.js, including edge cases like an empty string and a missing @ sign."

Working iteratively instead of demanding everything at once

A prompt doesn't have to be perfect on the first try. Especially for larger tasks, it's often more effective to work in small steps: ask for a rough version first, look at the result, then adjust in a targeted way ("This looks good, but shorten paragraph two" or "Use the same error handling as the neighbouring function instead"). That's not a sign of a bad first prompt - it's the normal, efficient way of working.

With these basics in hand, the next block looks at how Claude Code actually acts on these instructions: which tools it uses to read, change, and run code - and how you stay in control the whole time.