Example CLAUDE.md files are useful when they show the commands and failure cases of a real repository. The examples below come from Anthropic’s public projects: a Bun-based GitHub Action and a quickstarts collection with Python and React applications. Use their structure to write instructions for your own checkout.

Applies to: Claude Code documentation checked October 9, 2026. Repository files checked the same day: claude-code-action, file commit a8cb0db (September 30, 2026); claude-quickstarts, file commit 09bdac6 (September 22, 2026).

On this page

A Bun project puts its commands first

The CLAUDE.md in anthropics/claude-code-action begins with the commands a contributor needs. This excerpt is copied from its Commands section:

bun test                # Run tests
bun run typecheck       # TypeScript type checking
bun run format          # Format with prettier
bun run format:check    # Check formatting

The file then explains the action’s entrypoint and execution phases. Its warnings describe failures a contributor could introduce, including moving token cleanup into a process that might crash before cleanup runs. It also identifies strict TypeScript checks and the difference between this repository’s unit tests and integration tests maintained elsewhere.

For a Bun application, borrow the ordering: commands near the top, followed by the conventions and failure cases that affect edits. Replace every command with one your own repository defines. Copying the action’s internal paths would give Claude instructions for the wrong project.

A Python project names its quality tools

Anthropic’s claude-quickstarts CLAUDE.md separates instructions by application. The Computer-Use Demo section documents Python quality checks, including these exact commands:

ruff check .
ruff format .
pyright
pytest

The same section gives Python naming conventions, type-annotation requirements and the project’s error-handling convention. A contributor can see which checks belong to this demo without mistaking them for the checks of another quickstart.

For your Python project, use the example to review what an instruction file needs to resolve: the chosen test runner, the formatter, and any project-specific handling of errors or types. Confirm those tools in your checkout before putting their commands into CLAUDE.md. The excerpt is a reference for Anthropic’s demo, not an installation recipe for an arbitrary Python application.

A React project keeps application commands together

The Customer Support Agent section of the same quickstarts file groups its development commands with its TypeScript and component conventions. Its documented commands include:

npm install
npm run dev
npm run lint
npm run build

The file also distinguishes UI variants and directs contributors to package.json for build variants. The React instructions specify function components with hooks and the UI component library used by that application.

A useful instruction for your frontend names the convention your team follows and the script that checks it. An unrelated UI library or a nonexistent npm script would make the borrowed file misleading. Keep the source application’s details only when they match the project you are documenting.

Adapt the examples into one project file

Compare the source file with your repository before adding instructions. Check commands against its package scripts or documented development setup. Replace source-specific paths, remove conventions your team does not follow, and retain the failure cases that a new contributor would otherwise discover through a rejected change.

For a repository with several applications, the quickstarts file demonstrates one approach: separate sections inside a root instruction file. Decide whether those instructions belong in every session. Anthropic’s memory documentation also supports nested CLAUDE.md files, which load when Claude works in their subdirectories.

Store team instructions at ./CLAUDE.md or ./.claude/CLAUDE.md. Anthropic documents /init as a way to generate a starting file from a codebase, and /context as the check for loaded memory files:

/init
/context

Review the generated instructions before relying on them. For file locations and imports, use our CLAUDE.md best practices guide. Personal preferences across repositories belong in the global CLAUDE.md setup.

Questions

Can I copy an example CLAUDE.md unchanged?

Only if its commands and conventions describe your project. The examples here belong to Anthropic’s repositories; adapt their structure and verify their details against your checkout.

Does a CLAUDE.md instruction enforce a security rule?

Anthropic describes the file as context rather than enforced configuration. Use the documented permission or hook mechanism for a boundary that must hold; see our Claude Code hooks guide.

Should every application share one CLAUDE.md?

A root file can group application instructions, as the quickstarts repository does. Nested files are another documented option when instructions should load as Claude works in a particular subdirectory.

Sources