What it does Claude Code is a terminal-based agentic coding tool from Anthropic. You type natural language commands; it explores your repo, explains code, executes shell commands, and handles git workflows. It also works inside IDEs and via GitHub @-mentions.

The interesting bit The installation method tells the story: curl | bash is the recommended path, npm is explicitly deprecated. This is infrastructure, not a library. The tool collects usage data—accept/reject signals, conversation history, /bug reports—with safeguards around retention and training exclusions.

Key highlights

• Cross-platform: macOS/Linux via Homebrew or shell script; Windows via PowerShell or WinGet
• Plugin system with custom commands and agents (see plugins/)
• Built-in feedback loop: /bug reports go straight to Anthropic
• Data policies are documented, not buried—limited retention, restricted access, no training on feedback
• 129k+ stars suggests this is not a niche experiment

Caveats

• Requires Node.js 18+ despite the Python repo classification (the repo appears to be documentation/plugins; the tool itself is distributed differently)
• Data collection is opt-out-by-participation; review the Commercial Terms if your code is sensitive

Verdict Worth a look if you want an agent that actually lives in your shell and can push commits. Skip if you need air-gapped or fully offline tooling.