8 min · +120 XP
Ask for the whole job at once
What a complete request contains, and which chat habits to drop.
By the end of this lesson
You will be able to write one request that lets Claude finish a small tool without coming back with questions.
Claude Code does not need special words. It needs the same things a capable assistant would need to finish a job without phoning you every ten minutes. If something important is missing from your message, Claude will either guess or stop to ask.
The five parts of a complete request
- 1Goal: the result you want, described as an outcome. "A page where I add expenses and see the monthly total."
- 2Context: who it is for, what already exists, which files or folder to use, any limits. "For me and my wife, on this laptop, no internet needed."
- 3Example: one picture or sample of what good looks like. A screenshot, a paper form you use now, a site you like.
- 4What "done" means: how you will know it is finished. "I can add, edit and delete an expense, and the data is still there after I close the page."
- 5Freedom: how much Claude may decide alone. "Choose the details yourself, build the whole thing, then tell me how to open it."
The first request gives Claude almost nothing to work with, so you get a generic page or a list of questions. The second one can be finished in one go. It took a minute longer to write and saves several rounds of back and forth.
“Build a small budget tool for me and my family. We add an expense with a name, an amount and a category, and see the total for the current month and a simple chart by category. Keep everything on this computer, no accounts. Large text, calm colours. It is done when we can add, edit and delete an expense and the data stays after the page is closed. Decide the details yourself, build it completely, check that it works, then tell me how to open it.”
Change the details to your own situation. Claude builds the whole tool in one go and tells you how to open it.
›How to write a long request in the terminal
Pressing Enter sends the message. To start a new line inside the same message, press Ctrl+J (works in every terminal) or Shift+Enter (works in most modern terminals). You can also write the request in any text editor, copy it, and paste it into the terminal in one piece.
Four chat habits to drop
- 1Asking in small pieces. In a chat you ask small questions and assemble the answers yourself. Here you describe the whole task and Claude does the assembling.
- 2A new chat for every question. Within one project, stay in one session, because it remembers what you already agreed. Your files are safe either way, but a fresh session has to be told everything again.
- 3Regenerating instead of steering. If you do not like the result, say exactly what to change: "the total should be at the top and larger". Claude changes that and keeps the rest.
- 4Copying results out. The result is real files in your folder that already work. You open them; you do not paste them anywhere.
This is a safe practice terminal, nothing here touches your computer. Follow the tasks; for the free-text steps any wording is accepted, so write the way you would really write.
1Start with the weak request. Type: make me a budget app
The docs' best-practices page puts three things first. Give Claude a check it can run (tests, a build, a screenshot to compare) and ask for evidence, not a claim of success. Name the specific files, the scenario and the constraints instead of "fix the login bug". For uncertain or multi-file changes, separate the phases: explore and plan in plan mode, then implement, then commit. You can also ask Claude to interview you about the task before it writes a spec.
Claude built your tool, but the buttons are too small and the chart is monthly instead of weekly. What is the best next message?
Which part is missing from this request: "Build a recipe list for me. Large text. Decide the details yourself."?
Facts checked against the official Claude Code docs October 24, 2026 · source
Author
Evgenii Arsentev
PhD · Chief Executive Officer, digital health