Claude Code is Anthropic's AI coding agent. It reads a codebase, writes and edits files, runs shell commands and tests, and works with git on a developer's behalf. Unlike a chat window that only sees pasted snippets, Claude Code operates directly on a project: it can understand how the code fits together, make changes across files, and verify them by running tests. It is available from the terminal, a desktop app, the browser, and inside code editors, and the same underlying agent powers every surface.

Where does Claude Code run?

Claude Code runs across several surfaces that share one engine. The browser version runs in a sandboxed cloud environment with no install. A desktop app for macOS and Windows offers visual diff review and scheduled tasks. The terminal command-line tool gives full control and the complete extension surface. Editor integrations bring it into Visual Studio Code and the JetBrains family of IDEs. Because the surfaces share configuration, project context and settings travel with the developer rather than being set up again on each one.

What is a CLAUDE.md file?

A CLAUDE.md file is a markdown document at the root of a repository that Claude Code reads at the start of every session. It works like the onboarding note a new developer would receive: what the project is, the tech stack, the conventions to follow, the things to avoid, and where things live. Because it loads automatically each session, it gives the agent persistent, project-level context instead of requiring the same background to be re-explained every time. Files can also be layered, with user-level, repository-level, and subdirectory-level instructions, where the more specific file takes precedence.

What are Claude Code skills, hooks, and plugins?

Skills, hooks, and plugins are how Claude Code is extended. A skill is a structured workflow invoked as a slash command: a markdown file that walks the agent through a repeatable, multi-step task. A hook is a script that runs automatically on an event, such as before or after a tool call, and enforces a project standard deterministically rather than relying on the model to remember it. A plugin bundles skills, hooks, and tool declarations into one distributable unit so a team can share a consistent set of capabilities. Together with CLAUDE.md, they cover persistent context, named workflows, and reactive automation.

How is Claude Code different from Claude Cowork?

Claude Code is the coding agent: it is built to read a codebase, edit files, run commands, and work with git. Claude Cowork is oriented toward delegated, asynchronous knowledge work rather than hands-on code editing. On the desktop app the two sit side by side as separate tabs, so a team can pick the right mode for the task. Lightbridge Labs covers the wider lineup in its Claude products compared guide at https://lightbridgelabs.ai/resources/claude-products-compared and explains Cowork at https://lightbridgelabs.ai/resources/what-is-claude-cowork.

Who is Claude Code for?

Claude Code is built primarily for software developers and engineering teams, but its surfaces widen the audience. The browser and desktop versions need no command-line knowledge, so technical product managers, analysts, and curious non-developers can use them to read a repository, draft a change, or review a diff. The terminal command-line tool and editor integrations suit developers who want full control and automation in continuous-integration pipelines. Teams typically adopt it by starting small on real tasks, writing a CLAUDE.md so context is consistent, then adding skills, hooks, and plugins as repeatable patterns emerge.

How do teams adopt Claude Code in production?

Teams usually begin with individual developers on everyday tasks, then standardize. A good first step is a CLAUDE.md that captures the stack, conventions, and known hazards so every session starts from the same context. Repeatable multi-step work becomes a skill; standards that must always hold become a hook; shared capabilities are packaged as a plugin. Moving from this per-developer practice to a governed, auditable deployment that a security team can review is exactly the engineering work Lightbridge Labs delivers through custom AI development at https://lightbridgelabs.ai/consulting/custom-ai-development.

RL Written by Robert LabardeeFounder and CEO

Last updated June 2026.

What is Claude Code?

Lightbridge Labs defines Claude Code as Anthropic's AI coding agent: a tool that reads a codebase, writes and edits files, and runs commands on a developer's behalf. It works from the terminal, a desktop app, the browser, and inside code editors, and it carries project context through a CLAUDE.md file.

Compare the Claude products Build custom Claude capabilities

Claude Code is Anthropic's AI coding agent that works directly on a codebase.

Claude Code is an agent built for software development. It reads the files in a project, writes and edits code, runs shell commands and tests, and works with git, all on a developer's behalf. The distinction that matters is direct access. A chat window only sees what gets pasted into it; Claude Code operates on the real project, so it can understand how modules fit together, make changes that span several files, and verify those changes by running the test suite.

That direct access is also why project context and guardrails matter. Within a single session Claude Code remembers everything that happened; across sessions, only what a project records for it persists. The mechanisms below, CLAUDE.md, skills, hooks, and plugins, are how a project tells the agent what it should know and how it should behave every time, rather than re-explaining the same context on each run.

Claude Code runs across browser, desktop, terminal, and editor surfaces from one engine.

Claude Code is the same agent presented through several surfaces. Most people only need one or two, chosen by role and task. The browser and desktop options need no command-line knowledge; the terminal and editor options suit developers who want full control. Because the surfaces share configuration, a session can even move between them.

Browser (web)

A browser-based version that runs in a sandboxed cloud environment with no local install. It clones a repository from a connected git host, makes changes there, and returns a branch or pull request. The friendliest first try, and useful from a phone or tablet.

Desktop app

A standalone application for macOS and Windows that puts coding sessions in one window alongside chat and delegated work. It supports visual diff review before changes are accepted, multiple sessions side by side, and recurring scheduled tasks.

Terminal CLI

The full command-line tool. It reads and edits files in the current directory, runs shell commands, integrates with git, and supports the complete extension surface of hooks, skills, and connected tools. The choice for developers who already live in a terminal.

VS Code extension

Adds Claude Code to Visual Studio Code and its forks, with inline diff viewing, plan review in a side pane, and conversation history alongside the code being changed.

JetBrains plugin

A native plugin for IntelliJ IDEA, PyCharm, WebStorm, and the rest of the JetBrains family, giving diff viewing and selection-context sharing inside the IDE. Availability and feature maturity move quickly, so confirm current status before relying on it.

Claude Code carries project context through a CLAUDE.md file at the repository root.

CLAUDE.md is a markdown file at the root of a repository that Claude Code reads at the start of every session. It is the primary way to give the agent persistent, project-level context: a one-paragraph description of what the project is, the tech stack, a brief map of the main directories, the coding conventions to replicate, and the things that would look plausible but are wrong for this project. Useful additions include how to run the tests, the preferred git workflow, and the gotchas that keep biting people.

Context can be layered. A user-level file applies to all of a developer's projects, a repository-level file applies to one project, and a subdirectory file narrows further, with the more specific file taking precedence. The repository-level file is the most common. The discipline that keeps it valuable is keeping it current: a stale CLAUDE.md hands the agent wrong instructions with high confidence, so the convention is to update it in the same change that fixes whatever made it drift.

Claude Code is extended with skills, hooks, and plugins on top of CLAUDE.md context.

Beyond persistent context, Claude Code offers three extension mechanisms. Skills capture named workflows, hooks add reactive automation, and plugins package the pieces for a team to share. Together with CLAUDE.md they cover the customization surface most projects need: long-lived context, repeatable workflows, deterministic checks, and distributable bundles.

CLAUDE.md

A markdown file at the repository root that the agent reads at the start of every session. It carries the project description, tech stack, conventions, and known hazards, so the same context applies every time rather than being re-explained per session.

Skills

Structured, repeatable workflows invoked as slash commands. A skill is a markdown file that tells the agent exactly what to do step by step, turning a multi-step task such as a dependency audit or a changelog into a single named invocation.

Hooks

Scripts that run automatically on events such as before or after a tool call, or when a response finishes. Hooks enforce a project standard deterministically, for example rejecting a file write that contains a forbidden pattern, without relying on the model to remember the rule.

Plugins

Packaged bundles that group skills, hooks, and tool declarations into one distributable unit, so a team can share a consistent set of capabilities rather than configuring each piece by hand.

Skills, hooks, and plugins are software. Taking them from a single developer's machine to a governed, auditable, fleet-wide deployment is an engineering exercise, and it is the work Lightbridge Labs delivers through custom AI development: production-grade custom Claude skills, plugins, connectors, and hooks built as reviewable artifacts.

Claude Code is the coding agent, distinct from Claude Cowork and the wider Claude lineup.

Claude Code is purpose-built for hands-on code: reading a repository, editing files, running commands, and working with git. Claude Cowork is oriented toward delegated, asynchronous knowledge work rather than direct code editing. On the desktop app the two appear as separate tabs alongside chat, so a team can pick the mode that fits the task at hand. The point is fit, not hierarchy.

For the full picture, the Claude products compared guide sets each tool against the others, and what is Claude Cowork explains the delegated-work surface in depth. Developers who want a structured path into agentic coding can start with the AI for developers program from Lightbridge Labs.

Claude Code suits developers first, and teams adopt it by standardizing context before automation.

Claude Code is built primarily for software developers and engineering teams, yet its browser and desktop surfaces invite a wider audience, including technical product managers and analysts who want to read a repository or review a diff without touching a terminal. Adoption tends to follow a rhythm: orient on the task, plan the approach, let the agent make changes, verify by running tests, then commit. Short loops with human review beat long autonomous runs on complex work.

Teams scale that practice by standardizing. They write a CLAUDE.md so every session shares the same context, turn repeatable multi-step tasks into skills, encode standards that must always hold as hooks, and package shared capabilities as plugins. The harder step, making that deployment consistent, auditable, and defensible to a security review, is where Lightbridge Labs builds custom Claude skills, plugins, and hooks as production software.

Vendor note: Claude Code's surfaces and capabilities change quickly. Verify the current feature set, supported platforms, and any subscription requirements at docs.anthropic.com before you plan around a specific detail.

What is Claude Code: frequently asked questions

What is Claude Code?: Claude Code is Anthropic's AI coding agent. It reads a codebase, writes and edits files, runs shell commands and tests, and works with git on a developer's behalf. Unlike a chat window that only sees pasted snippets, Claude Code operates directly on a project: it can understand how the code fits together, make changes across files, and verify them by running tests. It is available from the terminal, a desktop app, the browser, and inside code editors, and the same underlying agent powers every surface.
Where does Claude Code run?: Claude Code runs across several surfaces that share one engine. The browser version runs in a sandboxed cloud environment with no install. A desktop app for macOS and Windows offers visual diff review and scheduled tasks. The terminal command-line tool gives full control and the complete extension surface. Editor integrations bring it into Visual Studio Code and the JetBrains family of IDEs. Because the surfaces share configuration, project context and settings travel with the developer rather than being set up again on each one.
What is a CLAUDE.md file?: A CLAUDE.md file is a markdown document at the root of a repository that Claude Code reads at the start of every session. It works like the onboarding note a new developer would receive: what the project is, the tech stack, the conventions to follow, the things to avoid, and where things live. Because it loads automatically each session, it gives the agent persistent, project-level context instead of requiring the same background to be re-explained every time. Files can also be layered, with user-level, repository-level, and subdirectory-level instructions, where the more specific file takes precedence.
What are Claude Code skills, hooks, and plugins?: Skills, hooks, and plugins are how Claude Code is extended. A skill is a structured workflow invoked as a slash command: a markdown file that walks the agent through a repeatable, multi-step task. A hook is a script that runs automatically on an event, such as before or after a tool call, and enforces a project standard deterministically rather than relying on the model to remember it. A plugin bundles skills, hooks, and tool declarations into one distributable unit so a team can share a consistent set of capabilities. Together with CLAUDE.md, they cover persistent context, named workflows, and reactive automation.
How is Claude Code different from Claude Cowork?: Claude Code is the coding agent: it is built to read a codebase, edit files, run commands, and work with git. Claude Cowork is oriented toward delegated, asynchronous knowledge work rather than hands-on code editing. On the desktop app the two sit side by side as separate tabs, so a team can pick the right mode for the task. Lightbridge Labs covers the wider lineup in its Claude products compared guide at https://lightbridgelabs.ai/resources/claude-products-compared and explains Cowork at https://lightbridgelabs.ai/resources/what-is-claude-cowork.
Who is Claude Code for?: Claude Code is built primarily for software developers and engineering teams, but its surfaces widen the audience. The browser and desktop versions need no command-line knowledge, so technical product managers, analysts, and curious non-developers can use them to read a repository, draft a change, or review a diff. The terminal command-line tool and editor integrations suit developers who want full control and automation in continuous-integration pipelines. Teams typically adopt it by starting small on real tasks, writing a CLAUDE.md so context is consistent, then adding skills, hooks, and plugins as repeatable patterns emerge.
How do teams adopt Claude Code in production?: Teams usually begin with individual developers on everyday tasks, then standardize. A good first step is a CLAUDE.md that captures the stack, conventions, and known hazards so every session starts from the same context. Repeatable multi-step work becomes a skill; standards that must always hold become a hook; shared capabilities are packaged as a plugin. Moving from this per-developer practice to a governed, auditable deployment that a security team can review is exactly the engineering work Lightbridge Labs delivers through custom AI development at https://lightbridgelabs.ai/consulting/custom-ai-development.

This guide is independent, general educational information published by Lightbridge Labs. Claude, Claude Code, and Anthropic are trademarks of Anthropic, PBC. Lightbridge Labs is not affiliated with, endorsed by, or a partner of Anthropic, PBC.

From understanding Claude Code to running it in production.

When the question shifts from what Claude Code is to how your team runs it safely at scale, Lightbridge Labs builds the custom skills, plugins, and hooks that make a Claude capability auditable and centrally controlled.

Talk to Lightbridge Labs See custom AI development