Overview
Claude Code is a command-line AI programming tool launched by Anthropic, allowing Claude to work directly in your terminal—reading and writing files, executing commands, searching code, and running tests. Unlike IDE plugins, Claude Code uses the command line as its interface, making it more suitable for system-level development and backend engineers' workflows.
**⚠️ Important Security Notice (2026/7/8 MIIT Notification + Full Incident Overview)**: On 2026/6/30, a user discovered through reverse engineering that Claude Code had been embedding a covert detection mechanism since **2026/3 (versions 2.1.91 to 2.1.196)**—reading the system timezone (identifying Chinese timezones like Asia/Shanghai) and comparing against a list of **147 domains** (involving Chinese companies such as Baidu, Alibaba, ByteDance, Moonshot, MiniMax, Stepfun, and Claude API relay services). The detection results were covertly marked using **Unicode steganographic encoding** (replacing date format symbols and apostrophe characters in system prompts) and reported with each request, without any mention in release notes. **Anthropic team members acknowledged on 7/1** that this mechanism was an experimental measure to "prevent account resale and model distillation attacks," and **promised to fully roll back and remove it in the 7/2 update**. **On 2026/7/8, the MIIT National Vulnerability Database (NVDB) officially notified**, rating the hazard level as "severe." **Anthropic has completed the fix in version 2.1.204 and above**. Alibaba has added it to its internal high-risk software list and fully banned it since 7/10, recommending Qoder as an alternative. **Official recommendation**: Check your current installed version; if it falls within the affected range, immediately uninstall or upgrade to the latest secure version.
**2026/5/28**: The flagship model was upgraded to **Claude Opus 4.8**, with programming and agent task capabilities surpassing GPT-5.5. **2026/6/30 filled a key gap**—**Claude Sonnet 5** simultaneously landed on Claude Code, positioned as "thinking ability close to Opus 4.8 + Sonnet-level execution speed," and is called the long-missing "best executor" for Claude Code: Opus 4.8 handles complex planning and reasoning, while Sonnet 5 handles high-frequency, low-cost execution. The combination significantly reduces the overall cost of long tasks. Anthropic officially launched the **Claude Code Plugin Directory** on GitHub, building a high-quality AI programming extension ecosystem; the enterprise version introduced **Managed Agents** self-hosted sandboxes (paired with Project Glasswing).
The core advantage of Claude Code is based on Claude's powerful code understanding ability, combined with direct terminal operation permissions, enabling file-level code refactoring, multi-file modifications, and complex development tasks.
Key Features
- Terminal Native: Directly converse with AI for programming in the command line, no IDE needed
- File Read/Write Permissions: AI can directly read and modify your code files
- Command Execution: Can execute shell commands, run tests, and install dependencies
- Codebase Understanding: Able to understand the entire project's code structure and dependencies
- Multi-file Refactoring: Modify multiple files in a single conversation, handling cross-file dependencies
- Git Integration: Understands git history, can commit code and create branches
Use Cases
- Terminal programming for backend and system developers
- Refactoring and modifying large codebases
- Automation of DevOps and operations tasks
- Experienced developers who prefer command-line workflows
- Automated code modifications in CI/CD pipelines
Pros
- Terminal Native: The most natural AI programming method for command-line users
- Claude Code Ability: The underlying model performs top-tier in programming
- Broad Operation Permissions: Can directly read/write files and execute commands
- Strong Multi-file Operations: Performs well in complex refactoring scenarios
- No IDE Lock-in: Works in any terminal environment
Pricing
Claude Code is free to install, but usage is billed via the Anthropic API. **Claude Opus 4.8** (recommended for complex planning): Input $15/million tokens, Output $75/million tokens. **Claude Sonnet 5** (recommended for daily execution, launched 2026/6/30, priced at 60% of the previous Sonnet 4.6): Cost-effective tier, suitable for high-frequency execution tasks. A typical programming session costs approximately $0.5-$5; routing to Sonnet 5 can further reduce costs. Usage can also be obtained through Claude Max subscription ($100/month or $200/month).
Summary
Claude Code is the 'AI programming tool for terminal geeks'—if you're used to working in the terminal, it fits your workflow better than any IDE plugin. The Sonnet 5 launched on 2026/6/30 fills the long-missing 'high-frequency executor' role for Claude Code: complex tasks are planned by Opus 4.8, daily changes are executed by Sonnet 5, balancing quality and cost. **However, the security backdoor issue notified by MIIT on 2026/7/8 requires attention**—ensure your current version is not within the affected range of 2.1.91-2.1.196 (upgrade to 2.1.204 or above). For classified/national enterprise scenarios, evaluate carefully or choose alternatives like Qoder. Recommended for advanced developers who have confirmed a secure version and are accustomed to CLI. Frontend developers may prefer Cursor.
Version History
- MIIT Notification + Official Fix Confirmed (2.1.204+) (2026/07/08): MIIT NVDB officially notified the security backdoor in versions 2.1.91-2.1.196 as 'severe'; Anthropic had previously acknowledged on 7/1 that the mechanism was for 'preventing account resale + model distillation,' rolled back and removed in the 7/2 update, fully fixed from version 2.1.204; Alibaba fully banned it internally from 7/10 and recommended Qoder as an alternative