I Spent 100 Hours Learning Claude Code So You Don't Have To

Choose the right IDE: Use Google IDX or VS Code so AI features and Claude Code work smoothly and stay in your flow.

Install via Terminal: You can install Claude Code from the Extensions UI, but using the Terminal gives you better control and visibility (e.g. token usage) than the GUI.

Respect the context window limit: Claude has a limit of around 200,000 tokens (~150,000 words). Sending too much context hurts answer quality.

Use /context to check usage: Run /context regularly to see how many tokens you’ve used and what is taking up the AI’s memory.

Use a separate instance per project: For a new project or unrelated feature, open a new window so old context doesn’t confuse the AI.

Use Plan Mode first: Before writing code, switch to Plan Mode (Shift+Tab) so the AI can analyze structure first. This saves tokens and time.

Set goals in Plan Mode: In this mode the AI researches and plans without editing files, so you can review the approach before you "Build".

Create a local CLAUDE.md: Use /init to create CLAUDE.md as your project playbook (SOP) so the AI knows structure and rules.

Put your tech stack in CLAUDE.md: List languages and frameworks (e.g. Next.js 15, Tailwind) there to avoid the AI mixing stacks and breaking the project.

Don't put API keys or huge docs in CLAUDE.md: Link to main docs only; never put private API keys in the file.

Get comfortable with slash commands: Use /help (list commands), /cle (clear chat), and /model (switch AI model) to work faster.

Use Bypass Permissions with care: When you trust the plan, Bypass lets the AI edit files without asking every time, speeding up work.

Create Claude Skills for repeat tasks: Use Markdown "Skills" to store reusable workflows (e.g. scraping or email automation).

Reuse Skills across projects: Skills in the system folder are available in every project, so you don’t have to re-teach the AI.

Use a site URL as a template: Send a reference site URL so the AI can analyze the design and generate a new page with a similar structure.

Use Vercel to deploy the site: When the code is ready, ask Claude Code to prepare the repo for deployment on Vercel via GitHub so the site goes live in minutes.

Use Modal for backend and automation: For automations (e.g. an AI scraper), deploy on Modal for a simple setup and a useful free tier.

Connect GitHub for continuity: Linking the project to GitHub lets you push updates and have Vercel deploy automatically (vibe coding).

Learn by doing: The fastest way is to build something small (e.g. a one-page site or a simple bot) and then grow from there.

Keep CLAUDE.md up to date: When you add features, run /init or update the playbook so the AI has the latest context and doesn’t repeat mistakes.