RL/BFree · No signup · Keep the page open while you workFor teams ↗RL/B · The free on-ramp
Twenty minutes, five rungs. You leave with prompts that actually work, a real agent installed on your machine, and your first automated job running. No signup, nothing to buy; this page is the product.
Rung zero was showing up. 1 of 6 already done.
Start the ladder ↓2 minutes · nothing to install
Four prompts that outperform almost everything people type into a chatbot.
Open claude.ai, chatgpt.com, or whichever assistant you already use. The difference between a toy and a tool is the shape of the request. Copy one of these and fill in the bracket.
I have a problem: [describe it in one sentence]. Ask me one question at a time until you understand it well enough to give advice worth paying for. Then give the advice, and tell me what you would do in my position.
Forces the model to interview you before advising you. The advice stops being generic the moment it knows your specifics.
Explain [topic] to me twice. First as if I'm smart but completely new to it. Then the way the sharpest person in the field actually thinks about it. Finish by telling me what the first explanation had to oversimplify.
One explanation for comfort, one for depth, and the gap between them is where real understanding lives.
Here is an email I received: [paste it]. Draft the reply I should send, not the one I want to send. Calm, brief, keeps my position strong. Then list what I should deliberately not say, and why.
Separates the message that protects your position from the one that vents. You keep the leverage and the relationship.
I'm about to decide: [the decision, and your reasons]. Make the strongest honest case that I'm wrong. Then tell me what evidence would change your mind, and what evidence should change mine.
You already argued yourself into the decision. This buys the strongest argument against it for the price of a paste.
5 minutes · one command
A real AI agent in your terminal. It reads files, writes files, and does work; chat was the demo.
The terminal sounds like a programmer thing. It is a text box that runs on obedience. You will type three things, total. This installs Claude Code, the agent this entire site is built and run with; you need a Claude Pro plan or above for it to work once installed.
Mac: press ⌘+Space, type Terminal, hit enter. Windows: open the Start menu, type PowerShell, hit enter.
$ curl -fsSL https://claude.ai/install.sh | bashWINDOWS POWERSHELL$ irm https://claude.ai/install.ps1 | iex$ claude→ your browser opens; log in with your Claude accountCheck it took: type claude --version in a new terminal. A version number means you are in. Prefer no terminal at all? The desktop app runs the same agent with buttons.
5 minutes · your first delegation
Stop asking questions. Start assigning work. This is the moment it clicks for most people.
Type claude in your terminal, then paste one of these. Each is phrased so the agent shows you its plan before it touches anything; you stay the approver.
Look at my Downloads folder. Propose an organisation scheme with named folders and show me exactly which files would go where. Don't move anything until I say go.
A zero-risk first job: it can only look and propose. You see how it thinks before you let it act.
Create a new folder called my-page and build me a clean single-page personal site in it: my name, three lines about what I do, one contact link. Then open it in my browser.
The fastest visible win there is. A thing that did not exist, existing, four minutes after you asked.
Find the 20 largest files in my home folder, list them biggest first with sizes and locations, and flag anything that looks safe to delete. Don't delete anything.
Genuinely useful on day one, and read-only. It hunts; you decide.
10 minutes · the real unlock
One instruction, many rounds. The agent checks its own work and goes again until the job is actually done.
Everything above was one command, one result. The unlock is giving it a finish line instead of a task, and letting it iterate. Two pieces make that work.
In any folder, ask it to write its own briefing file. From then on, every session in that folder starts already knowing how you work.
Create a CLAUDE.md file for this folder. Interview me first: ask me five questions about what I use this folder for, how I like things named and organised, and what you should never do without asking. Then write the file from my answers.
CLAUDE.md is read automatically at the start of every session. Preferences stop being repeated; they become policy.
Then hand it a goal with a built-in test. It works, checks, fixes, and repeats; you come back to a result instead of a conversation.
Research [question you actually care about] using the web. Write the findings to a file called report.md. Then re-read the report as a sceptical editor, list its three weakest claims, verify or fix each one, and repeat until you can't find a weak claim. Tell me when it's done.
The self-check clause is the whole trick. Without it you get a draft; with it you get a deliverable.
The last rung is a door
Everything above is the free floor, and it stays free. The ceiling is your whole team working like this, on your actual files, in one configured day.