Claude Code
Overview
Claude Code is Anthropic's terminal-native agentic coding tool. Unlike chat-style assistants, it operates directly within your development environment — reading and writing files, running shell commands, managing git, and understanding your project autonomously.
Key Features
Terminal-Native
Launches with a single claude command. No editor or web app needed — interactive coding right in the terminal. Auto-detects project structure at repo root.
Model
The default model is Claude Opus 5, with extended thinking on by default, a flat 1M-token context window, and a 128K-token standard max output. Use /model or --model to switch to Sonnet 5 or another model.
Sub-agents and Parallel Execution
Multiple Claude Code instances work in parallel on shared tasks. A team lead coordinates and synthesizes results while teammates work in their own context windows, communicating over a mailbox system and a shared task list. Sessions running in different terminals can now reference each other's work and intermediate state, making it practical to divide one task across several Claude Code instances.
Built-in Browser
Claude Code can open external websites and read, click, and type on them from inside the terminal loop — verifying an app under development, reading official docs, or operating an authenticated admin panel without bolting on an MCP browser tool.
Claude Security Plugin
A multi-agent vulnerability scanner that runs inside the terminal, so a coding session can hunt for vulnerabilities in the same loop that generated the code.
/design Command
Create UI mockups directly in the terminal, folding a screen-design step that used to sit outside code generation into the Claude Code session itself.
Artifacts
Coding sessions can generate interactive web pages that use the full session context and auto-update when related changes happen, with version history so teams can share and iterate on a live page of in-progress work.
Auto Mode by Default
To protect developers from rubber-stamping approval prompts, auto-judgment is enabled out of the box and only genuinely risky commands are surfaced for human confirmation.
Pricing
Included in Claude subscriptions:
| Plan | Monthly | Highlights |
|---|---|---|
| Pro | $20 | Claude Code access (Sonnet 5 default, Opus 5 selectable) |
| Max (5x) | $100 | 5x usage, Opus 5 by default |
| Max (20x) | $200 | 20x usage, priority access |
API access is pay-per-use, with Claude Opus 5 at $5 per million input tokens and $25 per million output tokens. The full 1M context window carries no premium, the Batch API halves both rates, and cache reads drop input to $0.50 per million.
Typical Use Cases
- Multi-file refactoring
- Autonomous feature implementation, tests included
- Understanding and explaining large codebases
- Automated bug fixing
- CI/CD script generation
- Parallel frontend/backend/test work via sub-agents
Strengths
- ✅ Terminal-native fits dev workflows
- ✅ Sub-agents for multi-agent collaboration
- ✅ 1M token full-repo context
- ✅ Claude Opus 5 coding capability
- ✅ Long-running autonomous tasks
- ✅ Available from Claude Pro at $20/mo
Weaknesses
- ❌ No GUI — not ideal for editor-focused devs
- ❌ Multi-agent use consumes a lot of tokens
- ❌ Effectively locked into Anthropic models
- ❌ Windows requires WSL
Official Resources
使いこなし Tips
SNS・コミュニティで共有されている実践的な使い方
- In Claude Code, press Esc twice to rewind the session to a previous checkpoint — a free undo for risky changes (Source: https://x.com/svpino/status/2050563493708112190)
- In Claude Code, press Ctrl+V to paste a screenshot directly into the chat — point at UI issues visually instead of describing them (Source: https://x.com/svpino/status/2050563493708112190)
- Claude Code tip from Anthropic: run `/context` to audit what's loaded at session start, and `/compact` before stepping away so the next session starts lighter (Source: https://x.com/hqmank/status/2088635186938601544)
- In Claude Code, press Ctrl+R to reverse-search your prompt history instead of scrolling with arrow keys (Source: https://x.com/svpino/status/2048027253616058869)
- In Claude Code, when deciding between a custom slash command and a skill: pick a slash command when you want to invoke it yourself, and a skill when you want Claude to auto-load and call it (Source: https://x.com/housecor/status/2007591718196261300)
- In Claude Code, reference a file or directory directly with @filename.py or @src/classes/ to constrain the agent's work to that target (Source: https://x.com/svpino/status/2029645889980305777)
- In Claude Code, press Ctrl+G to open your editor and compose long prompts there instead of typing directly into the terminal (Source: https://x.com/svpino/status/2050563493708112190)
- Claude Code's Remote Control: run `claude remote-control` on your PC and that machine appears in the Claude mobile app, letting you start and control the session from anywhere (Source: https://x.com/Hyde_ai3/status/2092488756737581419)