Claude Code Goes GA: What Changed and When to Reach for It

I've lost count of how many times I've pasted a function into a browser tab, gotten a fix back and then spent five minutes adjusting it because it didn't match the way the rest of the file was written. That round trip is the tax most AI coding tools charge you. Claude Code skips it by living in the terminal and reading your actual project before it touches anything.

It's been running in research preview for a while. Now it's generally available, with native installers on every major platform and a star count that's stopped looking like early traction and started looking like infrastructure. The anthropics/claude-code repository is the place to see it firsthand.

What Happened

Claude Code reached general availability after its earlier preview period. The repository sits at 135k stars and 21.8k forks across 56 contributors, with the CHANGELOG and plugin manifests updated as recently as this week.

The plugin format was bumped to version 1.1.0 in mid-June, and the frontend-design skill plugin picked up the update as well. That's a small detail, but it tells you the plugin system isn't an afterthought. People are actively building on it.

Anthropic also changed install paths. npm install -g @anthropic-ai/claude-code is deprecated. The recommended route now is a native installer: a shell script for macOS and Linux, PowerShell for Windows, or Homebrew and WinGet if you'd rather use a package manager.

Key Features

• Terminal-native agent: Run claude from your project directory and describe what you want in plain English. No browser tab, no copy-paste loop.
• Codebase awareness: The tool reads your actual files before making any changes, so a fix or refactor fits your existing patterns rather than looking like it was dropped in from a generic snippet.
• Git workflow handling: Stage, commit, and manage branches through conversation. Useful for the git steps that don't need a human to think carefully about syntax.
• GitHub integration: Tag @claude directly on an issue or pull request, and the agent joins your existing review process. No separate tooling, no context switch out of GitHub.
• Plugin system: Custom commands and agents extend the base tool. The plugins directory is active and clearly maintained, not a one-time feature dump.
• Cross-platform installers: Native paths for macOS, Linux, and Windows now exist side by side, which matters if your team isn't all on the same OS. Full options are in the setup documentation.

Why It Matters

The real shift is where the work happens. A chat assistant makes you describe a problem, copy the answer, and fix what doesn't fit. Claude Code reads your files directly and edits them in place, so there's nothing to paste and nothing that drifts from your actual codebase.

For teams, the GitHub integration is the part I'd actually build a workflow around. Tagging @claude on an issue keeps AI-assisted changes inside the same pull request and review flow your team already uses, instead of off in a separate tool nobody checks.

Deprecating the npm install path is a quieter signal, but it's a real one. Anthropic is treating Claude Code as a CLI tool in its own right, not a Node package that happens to live in your global modules. That distinction matters if you're deciding whether to standardize on it for your team.

Example Use Case

Say you inherit a Python service with a flaky test suite and zero documentation. You open the project and run claude, then ask it to explain what a specific module does. It reads the actual source rather than guessing from whatever snippet you'd normally paste into a chat window.

Next, you ask it to fix a failing test. The agent looks at the test file, the code under test, and the error output, then proposes a change. You approve it, then ask it to stage the fix and write a commit message. All of that happens in one terminal session, no tab-switching required.

Worth noting: the repository itself is 79.7% Python, so the team building this tool works in the same language as a lot of its users.

Competitive Context

Naming matters here, and it trips people up. Claude is Anthropic's model and chat interface. Claude Code is the agentic CLI built on top of it. If you're doing research, drafting, or asking a one-off code question, the chat interface is plenty. If you want something that edits files, runs commands, and manages git inside a real repository, that's Claude Code's job.

Compared with other terminal- and editor-based assistants, the differentiator is the combination of codebase reading, direct git control, and the @claude GitHub hook, all in one tool. A lot of assistants do autocomplete or chat well. Fewer act as an agent across your whole repo and your review workflow at the same time.

One thing worth checking before rolling this out at a team level: Anthropic collects usage data, including code acceptance and rejection signals, plus conversation data. The stated policy is limited retention and no use of feedback for model training. Worth reading the actual data usage policy yourself before you commit a whole team to it.

My Take

General availability after a long preview usually means a tool has stabilized enough that the team building it trusts it for daily use, not just early adopters willing to deal with rough edges. 56 contributors and an active plugin ecosystem back that up.

What I'd actually test before standardizing on this for a team: whether the GitHub integration holds up under real review volume, not just a demo PR. A tagged @claude comment on one issue is easy to show off. A team running it across dozens of PRs a week is the actual proof point, and that's not something a star count tells you. Start with the official documentation if you want the full picture before testing it yourself.