Overview
Claude Code is an agentic coding tool from Anthropic that runs in your terminal. You describe what you want in plain language and it works inside your codebase: it reads and explains code, executes routine tasks, edits files and handles git workflows. Beyond the terminal it also runs inside your IDE, and on GitHub you can tag @claude on an issue or pull request to hand it work.
Installation is a one-line script per platform (a curl script on macOS and Linux, a PowerShell script on Windows), with Homebrew and WinGet packages as alternatives; the older npm install path is deprecated. After installing, you run the claude command inside a project directory and start giving it tasks. The full setup options, uninstall steps and troubleshooting live in Anthropic's setup documentation.
The GitHub repository is where the project's issues are tracked and where Anthropic publishes plugins that extend Claude Code with custom commands and agents. Usage data and feedback (including the /bug command) are collected under Anthropic's data-usage policies, with stated safeguards such as limited retention periods and a policy against using feedback for model training. Claude Code is a commercial product covered by Anthropic's Commercial Terms of Service.
What it does
- Works from natural-language commands in the terminal: explains code, executes routine tasks, edits files and runs git workflows
- Three surfaces: the terminal, your IDE, and GitHub where you tag @claude on issues and pull requests
- One-line installers for macOS/Linux (curl) and Windows (PowerShell), plus Homebrew and WinGet packages
- Plugins in the repository extend it with custom commands and agents
- In-product bug reporting with the /bug command, a public issue tracker and a Claude Developers Discord
- Documented data-usage policies with limited retention of sensitive data and no model training on feedback
Getting started
Install Claude Code with the script for your platform, then run it inside a project directory. The README notes that installing via npm is deprecated in favour of these installers.
Install on macOS or Linux
The recommended path is the install script. Homebrew is the alternative: brew install --cask claude-code.
curl -fsSL https://claude.ai/install.sh | bashInstall on Windows
Run the PowerShell installer, or use WinGet with winget install Anthropic.ClaudeCode.
irm https://claude.ai/install.ps1 | iexStart it in your project
Change into the project directory and launch claude. From here you give it tasks in plain language and it works inside that codebase.
cd your-project
claudeReport a bug from inside a session
The /bug command files feedback directly; the repository's GitHub issues are the other channel.
/bugCommands and code are distilled from the project's own documentation — always check the official repo for the latest.
When to use it
- Delegate routine engineering tasks — refactors, test fixes, migrations — to an agent that works directly in your repository
- Get an unfamiliar codebase explained in plain language before you change it
- Let it drive git workflows such as branching, committing and preparing pull requests
- Tag @claude on a GitHub issue or pull request to hand work to it without leaving GitHub
Version history
Every verified update to Claude Code that AI/TLDR tracked, newest first — each links to our coverage and the official changeset.
- 2026-10-19v2.1.304
A guided /auto-mode-setup command teaches auto mode about your project, and Code Review gains Quick, Standard and Deep levels. Pro and Max users are warned before resuming a conversation would use a large share of the 5-hour limit.
- 2026-10-16v2.1.302
Fork subagents are on by default in claude -p, the Agent SDK, the desktop app and VS Code, and /config gains a switch for idle compaction. CLAUDE_CODE_AUTO_MODE_MODEL picks the client-side auto mode classifier's model.
- 2026-10-14v2.1.300
Trace spans carry OpenTelemetry GenAI semantic-convention attributes, and the api_request event splits cache writes into 5-minute and 1-hour tokens. Mods gain virtualText, a prompt.change hook and $.ui.view, and /advisor arrives on Claude Platform on AWS.
- 2026-10-14v2.1.299
Idle compaction compacts a 200K+ token terminal conversation left idle for about 54 minutes, before its one-hour prompt cache expires. CLAUDE_CODE_DISABLE_FILE_MENTIONS gives Claude the paths of @-mentioned files instead of their contents.
- 2026-10-09v2.1.296
CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL runs every workflow agent on one model, and subagents can set their own autoCompactWindow. Managed PreToolUse hooks that deny with "continue": false now refuse the call without ending the turn.
- 2026-10-08v2.1.295
Command and HTTP hooks accept onFailure: "block", so a hook that can't start, times out or exits unexpectedly blocks the action instead of letting it through. Terminals that implement the Program Status Protocol (OSC 7501) can now show whether Claude Code is working, waiting or done.
- 2026-10-05v2.1.290
WebFetch no longer silently drops page text past 100,000 characters: it says how much was unread and takes an offset to read on. The interactive WebSearch budget now refills at 100 calls an hour, and scheduled /loop tasks keep firing after compaction and background hand-offs.
- 2026-10-03v2.1.289
Bash deny and ask rules now catch commands behind an environment variable prefix or bare assignment under sandbox auto-allow, and Read deny rules apply through IDE symlinks. Plugins gain agent.spawn for teammates, and a failing mod now fails alone instead of ending the session.
- 2026-10-02v2.1.288
Non-interactive sessions and subagents now continue from a partial reply after a mid-response API timeout, and a prompt cleared with Ctrl+C comes back when you press Up. /code-review gains --max-findings, and many resume, plugin and MCP bugs are fixed.
- 2026-10-01v2.1.287
Adds Claude Mods, plugins that can modify deeper behavior, and a built-in "You should know" mod where a side agent flags things you or Claude might miss. Opus 4.7+ and Fable now default to a 1M context window on Bedrock, Vertex, Foundry and the Claude apps gateway.
- 2026-09-30v2.1.286
Stacked permission prompts now show a count such as "2 of 5", and Claude Code retries once on the previous model of the same tier when the API refuses the default model. Fixes cover resumed sessions that lost turns after parallel tool calls, secret masking in logs, and the gateway spend meter's prompt-cache pricing.
- 2026-09-29v2.1.285
A new CLAUDE_CODE_DISABLE_WEB_FETCH variable turns off the WebFetch tool, claude --desktop opens the desktop app on the current session, and an allowedProviders admin setting limits API providers. Subagent, MCP server and artifact publishing bugs are fixed.
- 2026-09-28v2.1.284
Claude Sonnet 5.5 becomes the default Sonnet model, with a 1M-token context window at $2 / $10 per million input / output tokens. /usage and the status line now show spending in US dollars, /mcp reconnect all retries failed MCP servers, and damaged response streams are retried instead of shown as raw errors.
- 2026-09-25v2.1.283
Admins get two new managed settings: deniedModels blocks specific models, and availableModelsMatch set to "exact" pins exact model versions. A new /doctor prompt-audit command checks CLAUDE.md files, skills, agents and commands for patterns written for older models, and Amazon Bedrock's Mantle endpoint becomes an upstream provider.
- 2026-09-24v2.1.282
Project and local settings files now ignore OpenTelemetry variables that turn on export, set its endpoint or capture content, and a startup notice lists the telemetry variables that were ignored. Adds a maxProseWidth setting for wide terminals and makes auto mode use the server-side classifier by default when telemetry is off.
- 2026-09-23v2.1.281
A recursive rm aimed only at command-substitution output, such as rm -rf "$(pwd)", now asks for approval in auto mode and with --dangerously-skip-permissions, even under a Bash allow rule. Adds "attribution": false in settings.json to hide commit and PR attribution, and fixes resumed sessions that lost earlier reasoning or the prompt cache.
- 2026-09-22v2.1.280
Adds Claude Opus 5.5 (claude-opus-5-5) as the default Opus model, with a 1M-token context window at $4/$20 per MTok, and changes the default on Pro and Team Standard plans from Sonnet to Opus. Fixes writes through a symlinked path being judged by their in-tree spelling, so acceptEdits, allow rules and auto mode no longer approve a write landing outside the project.
- 2026-09-19v2.1.278
Auto mode's safety classifier moves to the server and stops being billed: the checks ride along with the session's own model requests, by default on Enterprise plans, Claude API accounts, Bedrock, Agent Platform and Foundry. A gateway that rewrites requests forces a fallback to billed classifier requests, and CLAUDE_CODE_AUTO_MODE_SERVER=0 opts out.
- 2026-09-18v2.1.277
Adds AGENTS.md support: in a project with no CLAUDE.md, Claude Code reads AGENTS.md as its project instructions, selectable under "Project instructions" in /config (not yet on Bedrock, Vertex or Foundry). Removes the deprecated TaskOutput tool, marks subagent results with a header so their text cannot pass as session instructions, and strips invisible Unicode from prompts.
- 2026-09-17v2.1.275
Terminal sessions signed in with a claude.ai account now sync the skills and plugins enabled there; syncClaudeAiSkills and syncClaudeAiPlugins turn each off. Also adds a send-now key that interrupts the turn and delivers every queued message at once. The follow-up v2.1.276 undoes a regression that made every request fail with a 400 when ANTHROPIC_BASE_URL points at a proxy or gateway.
- 2026-09-15v2.1.273
Fixes two permission-checker gaps: Bash commands the checker cannot fully analyze skipped the prompt under permissions.blockReadsOutsideWorkingDirectories, and a subshell could hide a dangerous rm in bypass mode. Also corrects a context meter that counted advisor-tool turns at roughly twice their real size, so auto-compact fired at about half the real window.
- 2026-09-14v2.1.271
Adds per-command allowed_domains to Bash, PowerShell and Monitor in auto mode with sandboxing, so the hosts a command needs are reviewed with it and opened for that command alone. Also brings fast mode to Claude Code Remote sessions, an omitClaudeMd agent flag that lets subagents skip CLAUDE.md files, and a modelPricing multiplier up to 10x for internal chargeback.
- 2026-09-11v2.1.269
Adds claude plugin eval, which runs a plugin against a suite of test cases and scores them, repeating each case without the plugin loaded so the report shows what the plugin contributed. Also adds /output-style for headless and cloud sessions, a diff of the files a Bash command changed, and a setting to raise the Workflow tool's concurrent agent limit.
- 2026-09-09v2.1.267
Adds the maxEffortLevel setting, which caps effort level on every provider including Bedrock, Vertex and Foundry, and the --system-prompt-snapshot off flag. A long run of fixes repairs prompt-cache reuse across model switches, MCP reconnects and resumed sessions.
- 2026-09-04v2.1.261
Adds /skill-doctor, which lists the loaded skills a session never used and what each costs in context, and new settings that raise inline command output to 128K characters.
Claude Code in the news
- 2026-10-19MAJORClaude Code 2.1.304 — guided auto mode setup and Code Review levels
- 2026-10-16MAJORClaude Code 2.1.302 — fork subagents on by default in the SDK and apps
- 2026-10-15NOTABLEHow the Claude Code team does code review — seven layers, few humans
- 2026-10-14MAJORClaude Code 2.1.300 — OpenTelemetry GenAI spans and new mod hooks
- 2026-10-14MAJORClaude Code 2.1.299 — idle compaction saves long sessions before the cache expires
- 2026-10-09MAJORClaude Code 2.1.296 — one model for workflow agents, earlier subagent compaction
- 2026-10-08MAJORClaude Code 2.1.295 — hooks can fail closed, terminals show agent status
- 2026-10-05MAJORClaude Code 2.1.290 — WebFetch reads past 100K characters, /loop survives compaction
From the AI/TLDR release feed — every item is source-verified when it ships.
How Claude Code compares
Claude Code alongside other open-source autonomous coding agents tools AI/TLDR tracks, ranked by GitHub stars.
| Tool | Stars | What it does |
|---|---|---|
| opencode | ★ 220k | OpenCode is an open source AI coding agent that runs in your terminal, with built-in build and plan agents and an optional desktop app. |
| Claude Code | ★ 153k | Anthropic's agentic coding tool that lives in your terminal, understands your codebase and handles routine tasks and git workflows from natural language |
| OpenAI Codex | ★ 130k | OpenAI's cloud and CLI coding agent that writes features, fixes bugs, and proposes code changes across a repo, running tasks in parallel. |
| Pi | ★ 117k | Minimal terminal coding agent harness with four built-in tools, extended through TypeScript extensions, skills and prompt templates instead of forks. |
| Gemini CLI | ★ 107k | An open-source command-line AI agent from Google that connects your terminal to Gemini models for reading code, editing files, running shell commands, and searching the web. |
| Orca | ★ 101k | Open-source desktop workspace that runs Claude Code, Codex, OpenCode and other CLI coding agents in parallel, each in its own git worktree, with a mobile companion app. |
| autoresearch | ★ 98.3k | Karpathy's minimal harness that hands a coding agent a single-GPU LLM training script and lets it run fixed five-minute experiments, keeping or discarding each change on its own. |
| OpenHands | ★ 91.5k | An open-source AI software-development agent that plans tasks, edits files, runs commands, and tests code, usable from a terminal CLI, a local web GUI, or a Python SDK. |