$ █

What is AI?

AI — specifically large language models (LLMs) — are tools that understand and generate text. They've read enormous amounts of code and documentation, so they can help you write, debug, and reason about software.

Think of it as a very well-read pair programmer. You describe what you want, it writes the code. You review, iterate, and ship.

Getting Started with Spec Kit

Spec-driven development means you write a short specification before you write code. The spec becomes the single source of truth — for you and for the AI.

Initialize a project with one command:

specify init --here --ai claude

Then follow the slash-command workflow inside Claude Code:

  1. /speckit.constitution — Establish project principles
  2. /speckit.specify — Create baseline specification
  3. /speckit.plan — Create implementation plan
  4. /speckit.tasks — Generate actionable tasks
  5. /speckit.implement — Execute implementation

Optional commands to improve quality and confidence:

  • /speckit.clarify — Ask structured questions to de-risk ambiguous areas before planning
  • /speckit.analyze — Cross-artifact consistency and alignment report
  • /speckit.checklist — Generate quality checklists to validate requirements

Claude Code Essentials

CLAUDE.md

A markdown file at the root of your project that tells Claude how your codebase works — conventions, stack details, do's and don'ts. It's the foundation of every session.

# CLAUDE.md Never use the `any` type. Always run tests before committing.

Context

Claude builds understanding from everything you give it. The more precise your context, the better the output. Here's how to feed it:

  • @file / @folder — Reference files and directories inline. Type @ and autocomplete will suggest paths
  • Images — Paste screenshots (Ctrl+V), drag and drop, or reference image paths directly
  • Web — Claude can fetch URLs and search the web for up-to-date documentation and references
  • IDE integration — When running inside VS Code or JetBrains, Claude gets access to your open files, selection, and editor state

Sub Agents / Teams

Claude can delegate work to specialized agents — Explore for codebase research, Plan for architecture, Bash for shell commands, and more. They run in parallel when tasks are independent.

Sub agents vs agent teams comparison

Skills

Slash commands like /commit and /review-pr invoke built-in or custom skills. You can create your own skills to automate repetitive workflows.

Check out the skills documentation or community collections like Expo's skills for inspiration.

MCPs

Model Context Protocol (MCP) is an open standard that lets Claude connect to external tools and data sources — databases, APIs, GitHub, Slack, browsers, and hundreds more. Each MCP server gives Claude new capabilities it can use during your session, like querying a database or creating a pull request.

Language server protocols (LSPs) for your code's language give Claude real-time diagnostics, type checking, and go-to-definition — making it far more accurate when editing code.

Hooks

Shell commands that auto-trigger on Claude Code events. For example, run a linter every time a file is saved, or run tests after every code edit. They keep quality guardrails in place without manual effort.

Modes

Press Shift+Tab to cycle through Claude Code's input modes: Normal for standard prompts, Plan to design an approach before coding, and Auto-accept to skip approval prompts and let Claude work autonomously.