Claude Code Onboarding
SkillAI & modelsInteractive getting-started walkthrough for new Claude Code users. Guides you through environment setup, your first session, and best practices.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the Claude Code Onboarding skill
What this skill tells your AI
The instructions your AI receives, as published by orinachum/claude-code-guide in skills/onboard/SKILL.md and read by ahel’s review.
You are helping a developer get started with Claude Code. Walk them through the essentials interactively — don't dump information, guide them step by step.
The onboarding flow
Step 1: Environment setup
Help the user set up their environment:
- CLAUDE.md — Create a project-level
CLAUDE.mdfile with context about their codebase: tech stack, conventions, build commands, test commands, and any project-specific rules. This is the single most impactful thing they can do. - Permissions — Explain the three permission modes (Ask, Auto-edit, YOLO/Dangerously skip permissions) and help them pick the right one for their workflow. Most users should start with the default (Ask) and move to Auto-edit once comfortable.
- Model selection — Opus 4.8 is the default and most capable for everyday work. Sonnet 4.6 is faster and cheaper for routine tasks. Fable 5 is also available for long autonomous tasks (
/model fable). They can switch with/modelanytime. Suggest starting with Opus.
Step 2: First session
Guide them through their first real task:
- Start with something concrete — a bug fix, a small feature, or a refactor they already understand. This builds intuition faster than toy examples.
- The explore-plan-implement pattern — Teach them to: (1) let Claude explore the codebase first, (2) ask it to make a plan, (3) approve the plan, (4) let it implement. This produces dramatically better results than jumping straight to "fix this."
- Plan Mode — Show them how to use Plan Mode (Shift+Tab to toggle) when they want Claude to think before acting. Useful for complex tasks.
- Reading what Claude does — Encourage them to read Claude's tool calls and diffs, not just the final output. This builds understanding and catches mistakes early.
Step 3: Best practices
Share the practices that make the biggest difference:
- Be specific in prompts — "Add input validation to the signup form that checks email format and password length" beats "improve the signup form."
- Use git as a safety net — Commit before asking Claude to make big changes. If things go wrong,
git diffshows exactly what changed andgit checkout .reverts everything. - Break big tasks into steps — Instead of "build a REST API with auth," break it into: create the routes, add the middleware, write the tests. Claude handles focused tasks much better than vague ones.
- Self-testing loops — Ask Claude to write tests, run them, and fix failures in a loop. This is one of Claude Code's strongest patterns.
- Context management — For long sessions, use
/compactto summarize the conversation and free up context. Start new sessions for unrelated tasks. - Quick asides with /btw — Need to ask something unrelated mid-task?
/btwsends a one-off request that doesn't enter context. Great for quick lookups without losing your place.
After onboarding
Once you've walked them through the basics, let them know:
- Run
/guide:onboardanytime to repeat this walkthrough - Run
/guide:askto ask specific questions about any Claude Code feature — automation, hooks, skills, sub agents, plugins, model selection, and more
Signals
- GitHub stars
- 122
- Forks
- 4
- Last commit
- Jun 2026
- Hacker News mentions
- 20
Advanced
- Catalog kind
- skill
- Gateway key
onboard-orinachum- Source
- github.com/orinachum/claude-code-guide