Elective: learn from Claude Code itself

01The idea

1. 01/powerup — short interactive lessons on Claude Code’s own features, with animated demos. A lesson counts as finished when you press Enter on its card.
2. 02Output styles change how Claude answers. The Explanatory style adds short Insight blocks explaining the choices behind its code. The Learning style adds those and leaves a few real decisions for you, marked TODO(human), finishing with a "Learn by Doing" nudge. Switch with /output-style learning (or explanatory), back with /output-style default. The choice is saved to .claude/settings.local.json for this project.
3. 03claude-code-guide — a built-in helper for questions about Claude Code’s own features. Name it in the prompt — "Use the claude-code-guide agent: …" — or Claude may answer from general knowledge; when it hands over, the transcript shows a claude-code-guide(...) row.

Why leave decisions to yourself on purpose? An Anthropic study found that developers who learned a new library with an AI assistant scored lower, on average, on a quiz about it than developers who coded by hand — and the ones who scored well used the assistant to ask questions and get explanations, not just to produce code. The Learning style builds exactly that habit: it makes the session a teacher rather than a vending machine.

We are an AI-delivery company and we still drill this. The engineers who thrive here are the ones who can explain every line their agents ship; the style is a small, mechanical way to keep that muscle working while the agent does the typing.

02The walkthrough

1. 01/powerup — pick a lesson, follow it, press Enter on its card to mark it done.
2. 02/output-style learning.
3. 03Ask for a change that leaves a decision to you: "Add a function that counts external links, with a test. Leave the decision about repeated links to me." Read the Insight blocks; when you reach the TODO(human) comment, open the file, write your lines, tell Claude you are done, and let it review your code. A style is an instruction, not a guarantee — if Claude writes everything anyway, ask again.
4. 04Set the exercise aside: git stash push --include-untracked.
5. 05Ask the helper: "Use the claude-code-guide agent: how do I make my own slash command in Claude Code?" Look for the claude-code-guide(...) row in the transcript; the answer should point you to skills — a SKILL.md under .claude/skills/ (the older .claude/commands/ folder still works). That is a preview of the next installment.
6. 06/output-style default — end where you started.

03Your turn

One more /powerup lesson. Then /output-style explanatory, an explanation of a module you have not touched, and a read of its Insight blocks. Then one claude-code-guide question about a feature you have not used — watch for the handover row. Back to default.