Claude Code is an agentic CLI that lives in your terminal. It executes a continuous ReAct (Reasoning + Action) loop. You give it a task (e.g., "Refactor the login logic"), and it autonomously navigates your file system, reads code, runs tests, and applies fixes until the task is complete.
Unlike a standard copilot, Claude Code handles failures autonomously. If it runs npm run test and it fails, it ingests the entire error log, identifies the corrupted lines, and applies a fix without human intervention. It only stops when it reaches your goal or hits a safety wall.
The CLAUDE.md file in your project root serves as persistent memory across sessions. This markdown file contains project-specific guidance: coding standards, architecture decisions, dependency constraints, and context that Claude should always be aware of. Claude reads this file at session start and uses it to inform every decision it makes.
// Example CLAUDE.md # Project: InfinityStack - Framework: Next.js 15 with App Router - Deployment: Vercel CLI only, NEVER git push - Testing: vitest for unit, playwright for e2e - Style: Vanilla CSS, no Tailwind
Set up the Claude Code CLI from scratch, launch an interactive session, then run it headless — the exact workflow used in CI pipelines.