Tutor Notes.

How to start and stop your lessons.

Three short phrases run every tutor session. Worth saving them somewhere visible (a sticky note, your phone, a notes app).

To start your tutor session, open your terminal (PowerShell on Windows, Terminal on a Mac, your usual terminal on Linux) and type:

You'll see [FLT cage] appear, which means you're inside. Then type: (no flt on Linux, start here.)

Then once the tutor's running, type:

The tutor reads your progress and picks up where you left off.

To stop for the day, when you're done, type:

The tutor saves your progress, gives you your pick-up-next-time reminder, and tells you it's safe to close the window. (Capitalisation and punctuation are forgiving; "End lesson." or "end lesson" both work.)

60 seconds with your terminal.Worth a read before you start using the terminal.

The terminal doesn't quite behave like the rest of your computer. A few quick things will save you some "why isn't this working?!" moments:

That's it. Your tutor explains the rest as it comes up. (And if a key-press catches you out later, the Troubleshooting page has the common ones.)

Things to remember.Concepts and techniques from the curriculum, in order.

Quick reference for the things your tutor walks you through. Useful for refreshing the bits you've half-forgotten.

The prompt → action → review loop (step 1).

The shape of every interaction with the AI. You give an instruction (the prompt) → the AI takes an action in the real world → you review what it did before deciding what's next. The review beat is the one beginners skip, and the one that matters most.

Writing prompts the AI can act on (step 3).

Four principles:

• Be specific. Write it like a brief, not a hint. Say what you want, what it should do, what it should avoid. The memorable test: "show your prompt to a friend with minimal context; if they'd be confused, the AI will be too."
• Give context. Who's it for, what's it for, what are the constraints. Don't assume the AI knows your situation.
• Show, don't just tell. An example of what you want is worth a paragraph of describing it.
• Split big jobs. Don't ask for the whole castle in one prompt; build it room by room. The phrase to remember: "let's divide this up along the logical seams."

Pointing at files with @ (step 4).

When you want the AI to look at a specific file, put @ in front of the filename. For example:

The AI reads the file before doing anything, so it knows what it's working with.

In the cage setup: the tutor can only see inside the cage and your FLT folder, not your Downloads, Desktop, or Documents. To show it a file from elsewhere on your computer, drop the file into the FLT folder in File Explorer (Windows) or Finder (Mac) first, then @ it as usual.

The review muscle: predict-then-check (step 5).

Before approving any change the AI proposes, say (or think) what you expect it to do. Then watch. When your prediction doesn't match what happens, that's the moment to investigate, not just click through.

Save-points (step 6).

Every time the tutor narrates "I'll save a checkpoint here", it's making a git commit, a snapshot of your files you can return to. To undo something, just ask: "undo that" or "take us back to the last save-point." The tutor handles the rollback.

Commit vs push (step 7).

A commit is a save-point on your own computer. A push is sending those save-points up to GitHub (online). Commits protect your work from your edits going wrong. Pushes protect it from your laptop dying.

Plan before building (step 8).

When you want to build something (even small), the default move is "let's plan it first" before any code happens. Talk through what files would exist, what could go wrong, what order to do it in. Write the plan into a file (often called plan.md). The conversation is the lesson; the file is the artefact.

Troubleshooting prompts (step 9).

When something feels off, these phrasings work surprisingly often:

• "I'm seeing this: [paste the error]. What does it mean and how do I fix it?"
• "I expected X, but got Y. What might be causing that?"
• "This was working before you did X; what might have changed?"
• "What are 2 or 3 things that could be causing this? Don't change anything yet; let's think first." (The most important one: it stops the AI charging in to "fix" things wrong.)
• "@filename: have a look here and tell me if anything looks wrong."

Asking fresh AIs for things (step 10b closing).

Your tutor knows you. Other AIs (a fresh Claude in another folder, or any other AI) don't. When you talk to a fresh one, describe what you want as an outcome, like a stranger. The launcher alias guide on the Going further page shows the shape.

Tokens and context windows.Why long sessions sometimes get weird.

Every AI conversation has a memory limit, called the context window; what fills it up are tokens (roughly: words and bits of words). When a session gets very long, it slows down and the tutor can start to forget the beginning.

While you're learning this rarely bites: your tutor saves your progress in progress.md regardless, and shorter sessions stay sharper. When you want the fuller picture (and the handy commands for managing it), the Hints & Tips page covers it properly.

Things you can ask your tutor.

If you ever feel stuck on a concept or want a different angle, these phrases almost always work. The tutor will adapt.

• "Can you explain that in simpler terms?"
• "What did you just do?"
• "Why isn't this working how I expected?"
• "Can you show me an example?"
• "Slow down, I want to follow this step by step."
• "I'm lost, can we go back to before the last change?"

And if the tutor itself is the thing failing, message Becky or the WhatsApp group. That's by design, not a fallback.

© 2026 Field Level Tech Ltd Company #16126242 · Privacy policy