Hands-on: Claude Code

Claude Code is different. It doesn't live in the editor — it lives in the terminal. You tell it what you want, it works autonomously: opens files, modifies them, runs tests, opens PRs. It's more like a colleague you ask to do something, not an autocomplete.

Quick setup

# Instalare globală
npm install -g @anthropic-ai/claude-code

# În folderul proiectului
cd ~/Desktop/proiect
claude

# Apoi îți cere să te autentifici cu contul Anthropic.

Slash commands — shortcuts

Type / in the prompt and see a list of predefined commands. They're shortcuts for repetitive tasks.

• /init — generates a CLAUDE.md with a project summary.
• /review — review of the current PR.
• /security-review — vulnerability audit.
• Custom: you can create your own slash commands in .claude/commands/.

Sub-agents — delegation

When you have a task that would pollute context (e.g. search across 200 files where a function is used), delegate it to a sub-agent. The sub-agent has separate context, works in parallel, and returns ONLY a short report.

Tu: „Lansează un agent Explore care să găsească toate locurile
     unde se face validare email și să-mi dea o listă scurtă."

Claude Code:
  ▶ Spawning Explore agent...
  ▶ Agent finished. 4 locations found:
    - src/auth/signup.ts:45
    - src/auth/login.ts:23
    - src/profile/update.ts:78
    - src/admin/invite.ts:12

Tu: continui cu context-ul curat, fără 200 de fișiere citite.

Hooks — automation

Hooks are shell commands that run automatically on certain events: after each Edit, before each Bash, when a session ends. Useful for: auto-format after edit, audit logs, alerts.

// .claude/settings.json
{
  "hooks": {
    "PostToolUse": {
      "Edit": "prettier --write \"$CLAUDE_TOOL_FILE_PATH\""
    }
  }
}

Transcript — work audit

Every Claude Code session is automatically saved as a transcript (in ~/.claude/...). You can re-read, share, or ask another agent to analyze it. Useful as a BA for: auditing what was done, reusing a successful session as a template.