Claude Code Guide

What Claude Code Does

Claude Code brings Claude into software work with repository context and tool permissions. The official overview says it can read your codebase, edit files, run commands, and integrate with development tools. That makes it different from a chat assistant that only suggests snippets.

The high-value use cases are multi-file refactors, bug diagnosis, test creation, documentation updates, dependency audits, CI failure triage, code review, and repetitive engineering chores. It is especially useful when the task requires navigating local project conventions rather than answering a standalone programming question.

Where It Runs And How To Install It

The current documentation home is code.claude.com/docs/en/overview. It lists terminal, VS Code, desktop app, web, and JetBrains paths, plus native install options for macOS, Linux, WSL, Windows PowerShell, and Windows CMD. Most surfaces require a Claude subscription or Anthropic Console account; terminal and VS Code also support third-party providers.

Follow the install instructions on the docs site rather than a cached command from a blog post. Install paths and package names have changed more than once, and the quickstart page is the owner of that fact.

The browser and remote-control paths are useful for checking progress away from the workstation. IDE and desktop paths are useful when visual review or local app interaction matters. Terminal remains the most explicit surface for reproducible commands, diffs, tests, and commits.

Which Model Claude Code Runs

Model choice inside Claude Code follows the same September 2026 guidance as everywhere else: start with Opus 5, and move to Fable 5.1 for long-horizon agentic work or when Opus 5 at higher effort still fails your tests. Claude Code is exactly the setting where that escalation tends to pay off, because agent runs are long, tool-heavy, and expensive to get wrong.

Effort defaults differ by surface. Fable 5.1 defaults to High effort in Claude Code but Medium on Claude.ai and Cowork, so the same prompt can behave differently depending on where you run it. Fable 5.1 also prices cache reads at 2.5% of base input, which matters in Claude Code because agent loops replay large, stable context on nearly every turn.

Opus 5 adds a fast mode at roughly 2.5x speed for 2x base price, billed against platform and Claude Code usage credits. Use it when iteration speed matters more than token cost.

Automation And Agents

Claude Code can be scripted through the CLI, used in CI-style workflows, and expanded with memory, skills, hooks, subagents, background agents, and scheduled tasks. The docs frame it as composable: pipe logs into it, run recurring reviews, or use it for bulk operations across changed files.

The Agent SDK extends this into programmable agents. Use it when you need a controlled agent loop with explicit permissions, custom orchestration, or integration into a product or internal developer platform.

Safe Operating Pattern

Claude Code can change real files and run real commands. A pragmatic workflow is: isolate work in a branch or worktree, state the goal and non-goals, ask for a plan when the change is broad, run narrow tests frequently, require a final full build or test gate, inspect the diff, and commit only known-good changes.

For company use, pair Claude Code with repository permissions, secret scanning, dependency policies, and review rules. Treat prompt context as data exposure, especially when using third-party providers or remote work surfaces.