A free guide by Pierre Sabbagh
Five Claude codes, and how to use each one
L99, OODA, artifact, scaffold and ultrathink: what each one does, and an example to copy for each.
Needs a Claude account. Free is enough for four of the five. You won’t write any code.
It picks one real decision and writes the first message to paste. Press send when it opens.
Not now? Send it to my computer.
Written November 21 from Anthropic’s own pages. The steps are below.
Let AI walk you through it
It reads this guide and tells you what to click, one step at a time.
You’ll need a Claude account. The fifth code, ultrathink, needs Claude Code, which comes with Pro, Max and Ultra, and with Team and Enterprise.
It opens in a new tab with the prompt typed in. Press send.
Or follow the guide yourself
Here’s the full list from my video, with what to type for each one and an example you can copy. None of them is a setting you switch on: you type the word with your question and send it. Four of them work in any Claude chat. The fifth, ultrathink, is built into Claude Code. You won’t write any code.
What you need
- A Claude account for codes 1 to 4. The free plan is enough. Artifacts (code 3) are on the free plan too.
- For code 5, Claude Code. It comes with every paid plan (Pro, Max, Ultra, Team and Enterprise) and isn’t in the free plan.
Why one word works
Anthropic’s prompting page says to think of Claude as “a brilliant but new employee who lacks context on your norms and workflows.” It adds: “The more precisely you explain what you want, the better the result.”
Each code is a short way of saying what kind of answer you want. Claude reads the word with the rest of your message and answers that way. Under each code there’s also a sentence you can add when you want even more control.
Code 1: L99
What it does: Claude stops hedging with “it depends” and gives you a straight answer.
Put it in front of your question:
L99. Should I raise my prices for new clients this month?
I tried it, and that’s what it does: one answer instead of a list of options. To make it even firmer, add a line:
L99. Give me one answer, not a list of options. If it really depends, pick the case that fits me best and tell me the reason in two lines.
[your question]
Code 2: OODA
What it does: Claude analyzes your situation with a military decision framework: what to watch, what matters, what to decide and what to do.
Add it to your message:
OODA. A competitor just cut their prices by 20%. What should I do?
OODA stands for observe, orient, decide, act. It was developed by US Air Force Colonel John Boyd in the early 1970s and is used in business and law too, according to Wikipedia’s page on the OODA loop. Claude knows the framework by name, so the word is enough. To shape the answer, say what you want under each step:
OODA. Under each of the four steps, tell me what you see in what I wrote, and what you would need to know that I haven't said.
[describe your situation]
Code 3: artifact
What it does: Claude builds you a dashboard, or an app, right in the chat.
Put it in your message, with what you want built:
Artifact: a simple dashboard to track my sales this month. Ask me what to include before you start.
“Artifact” is Anthropic’s own word for what Claude builds: a design, a document, a dashboard or a small interactive tool. It opens in its own window beside your conversation, and you can change it by talking to Claude or directly in the artifact. Everything you make is saved in the Artifacts tab in the sidebar.
Two things from Anthropic’s help center:
- Claude makes an artifact when the result is substantial, typically over 15 lines, and something you’re likely to edit or reuse. A very short answer won’t become one.
- Artifacts need Cloud code execution and file creation turned on in Settings > Capabilities (called Create and edit files in the updated Settings layout). If you don’t get one, check that setting first.
Code 4: scaffold
What it does: Claude breaks any project or task down into a full action plan, with every step mapped out for you.
Put it before your project:
Scaffold: launch my online course in six weeks.
To get more out of it, say what the plan must cover:
Scaffold: [your project]. The steps in order, what each step needs before it can start, and who or what could block it. Ask me anything you need first.
Code 5: ultrathink
What it does: Claude thinks much harder before it answers. It’s my favorite, and it’s the one built into Claude Code itself. From Claude Code’s model configuration page:
- It works in Claude Code, not in the regular chat.
- It can go anywhere in your prompt. Before your question is fine. It asks for deeper reasoning on that one turn, and your usual setting stays as it was.
- On supported setups it raises that turn to
xhigheffort. There’s one level above it,max, which you set with/effort(below). - Other phrases such as “think”, “think hard” and “think more” are read as ordinary words. Only “ultrathink” is the keyword.
How to use it
- Download the Claude desktop app from claude.ai/download, install it and sign in.
- Click Code at the top of the sidebar.
- Make a new folder to work in, for example “Ultrathink test” in your Documents.
- In the row under the message box, choose Local, then click the folder control, choose Open folder… and pick that folder.
- Write your prompt with the word in it, for example:
Ultrathink. I'm choosing between [option A] and [option B] for [your situation]. Tell me which one and why.
The page says that in the desktop app’s Code tab, in a local session, the word can raise the turn to xhigh effort. This needs Claude Code version 2.1.316 or later. In a terminal it needs 2.1.311 or later, and in VS Code’s chat box 2.1.316 or later. In those places the word shows in rainbow colors when it can raise the turn.
If the word isn’t rainbow when you start a new turn, the page’s fixes are:
- Retype it outside quotes, brackets, a path or a command, in a prompt that doesn’t start with
/. - If you’ve set
CLAUDE_CODE_EFFORT_LEVELbelowxhigh, unset it. Or raise your ownmaxEffortLevel, or ask an organization admin to raise your organization’s limit. - Run
/modeland switch to Opus 5.5, Sonnet 5.5, Haiku 5.5, Fable 5.2 or Fable 5.1.
The word can’t raise a turn on Amazon Bedrock, Google Cloud’s Agent Platform or a Claude apps gateway. There, set the effort level instead (next section).
To turn the keyword off, set CLAUDE_CODE_DISABLE_ULTRATHINK to 1 (Claude Code 2.1.316 or later).
For a lasting change: /effort
ultrathink is for one turn. To change how hard Claude thinks for more than that, Claude Code has /effort:
/effort
On its own it opens a slider. You can also type a level after it, for example /effort high. The levels are low, medium, high, xhigh and max on current models. Some older models skip xhigh.
- Press
Enterin the slider, or type a level after/effort, to save the level as your default for later sessions. - Press
sin the slider to apply it to this session only (Claude Code 2.1.257 or later). maxis the deepest level, and Claude Code applies it to the current session only.
Anthropic says higher levels spend more tokens. The default is medium on Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5. About max, the page says it “may show diminishing returns and is prone to overthinking, so test before adopting it broadly.”
What to take from the list
| Code | What it is |
|---|---|
| L99 | One straight answer instead of “it depends”. Any Claude chat. |
| OODA | Your situation through observe, orient, decide, act. Any Claude chat. |
| artifact | A dashboard or small app built beside the chat. Any Claude chat. |
| scaffold | A full action plan, step by step. Any Claude chat. |
| ultrathink | Deeper reasoning for one turn. Claude Code. |
Written on 21 November 2026 from Claude Code’s model configuration page, Anthropic’s Claude prompting best-practices page, Claude’s help center article on artifacts, Claude’s pricing page and Wikipedia’s page on the OODA loop. Pierre tried L99 himself; the Claude Code steps were not run while writing this guide. Claude changes fast: if a menu has moved, look for the same words in the app.
The prompt
This is what the buttons send. Copy it and paste it into the AI app you use.
Walk me through this guide, one step at a time: https://pierresabbagh.com/resources/five-claude-codes Read the whole page first. If you can't open it, tell me so I can paste the text in, and don't guess the steps. I'll do each step myself on my computer, so tell me exactly what to click or type, and stop after each step so I can check it.
Send this guide to your inbox.
Next guide
Get your business ready for AI shoppers
A short checklist from OpenAI’s own pages: let ChatGPT’s agents onto your site, keep your checkout, and send them clear product data.
At work
Doing this at a company?
With a team, the hard part is the task itself, the people who do it and who checks the result.