Learning mode in the AI space

Why it exists#

Getting stuck is a normal part of learning. People used to search the web or open a textbook; now they ask an AI, and it almost always hands over a finished solution right away. The task is done, but you did not learn anything.

In Learning mode the agent leads you to the solution instead of handing it over.

Turn it on#

1. Open the AI space and start a session.
2. Next to the input, in the mode picker, choose Learning.
3. Ask as usual — for example, paste an error from a log or describe what is not working.

The session remembers the mode. To go back, use the same picker: With confirmation.

How the agent answers#

• Hints in steps. First a leading question or a direction. Still stuck — where to look: a log line, a place in a file, a section of the docs. Stuck again — an almost complete next step in words.
• Explains errors. It says what the log message means and which line it points to, and asks you to find the cause.
• Checks your solution. Suggest your own — in words or in code — and the agent says what is right, where the mistake is and why.
• Does not write the solution. Even "just give me the code" gets a hint.

What the agent cannot do in this mode#

A rule in the model prompt can be talked around, so SKILLOK itself enforces the important parts:

• The agent has no tools that write files or merge branches: proposing a change, merging a pull request, resolving a conflict, rewriting a challenge. Ask for it and it will refuse and tell you what to change yourself.
• If an answer still contains more than 8 lines of ready code and you have not shown any code of your own, the answer is rewritten into a hint before you see it.
• Reading files, searching the repository, looking at issues and searching the web work as usual.

Autonomous mode does not apply while learning: Learning mode wins.

For teachers: the mode for everyone#

The repository owner can set the mode for the repository:

1. Open the repository → Settings → AI.
2. Under Agents, check Learning mode for everyone and save.

From then on, every session with this repository attached runs in Learning mode. Instead of the picker, the input shows a Learning label that cannot be changed. This works well for course repositories and group challenges: see Challenges.

What it costs#

The same as a regular question: the mode uses the same model and the same limit. If an answer had to be rewritten into a hint, that is one more model call — you will see it in the cost of the turn.

Was this helpful?

Yes No