claude-onramp: Skill for Claude Code

Walk a non-technical user through Claude Code setup, step by step. Use AskUserQuestion for each decision point. Write all configuration to .claude/settings.local.json (personal, not shared).

Tone: Friendly, plain language, no jargon. Explain every concept before asking the user to make a choice. Keep explanations short — 2-3 sentences max per concept.

---

Step 1: Welcome and orientation

Print the following (adapt lightly if needed, but keep the substance):

Welcome to Claude Code!

Claude Code is different from the Claude you might use in a browser.
It runs on your machine, which means it can read your files, run commands,
and make changes to your project. That makes it very powerful — and it's
why we'll set up some safety guardrails now.

Here's what Claude Code can do that browser Claude can't:
- See and edit files in your project
- Run commands in your terminal
- Search the web for answers
- Track changes with git
- Remember things about your project across sessions

---

Step 2: Your safety net

Print the following:

Before we configure anything, here's the most important thing to know:

You can always undo. Claude Code automatically saves a snapshot of your
code before every change. If something goes wrong, press Esc twice (or
type /rewind) to go back to any previous point.

Think of it like infinite undo — you can restore your code, your
conversation, or both. So don't worry about getting anything wrong
during this setup. You can always change it later.

---

Step 2.5 Pause for user input

Print the following:

Any questions before we get started configuring your setup?

STOP HERE. Do not call any tools or produce any further output until the user sends a message. If they have questions, answer them. If they say they're ready (or anything like "no", "nope", "let's go", etc.), continue to Step 3.

Step 3: Permissions

Print:

Permissions control what Claude can do without asking you first. There
are actions Claude is blocked from entirely (like touching passwords or
force-pushing code), and actions where Claude will ask your permission
each time (like pushing code to GitHub or installing packages).

On top of those safety rules, you can choose how much Claude asks about
everyday work — editing files, running scripts, etc.

This project already has safety rules that can't be overridden. 

Claude is permanently blocked from:
- Accessing passwords, secret keys, and other sensitive files on your machine
- Actions that could cause serious damage, like overwriting code history or publishing your project to the internet

Separately, Claude will always ask your permission before uploading code to GitHub or downloading new software.

These rules apply no matter which option you choose next.

Make sense? If so, we can proceed.