Vercel AI SDK

Why This CLAUDE.md Stands Out

The Vercel AI SDK CLAUDE.md is widely regarded as one of the best examples in the wild. At roughly 300 lines, it strikes the perfect balance between thoroughness and readability.

Key Patterns

Exhaustive Architecture Section. The file maps every directory in the monorepo with its purpose, dependencies, and rules. This is critical for monorepos where Claude needs to understand package boundaries. Explicit Testing Patterns. Not just "write tests" but specific instructions on which test runner to use per package, how to mock external services, and coverage expectations by layer. Build System Documentation. Clear instructions on how Turborepo pipelines work, which packages depend on which, and how to run builds for specific packages. Contribution Workflow. Step-by-step instructions for adding a new AI provider, including which files to modify, which tests to write, and which docs to update.

When to Use This Pattern

Use the Vercel AI SDK pattern when:

• Your project is a TypeScript monorepo with multiple published packages
• Contributors need to understand package boundaries and dependency relationships
• Your build system has non-obvious configuration (Turborepo, Nx, Lerna)
• You publish to npm and need Claude to understand the release process

Structural Analysis

Sections: Stack | Architecture (with directory tree) | Packages |
          Testing | Building | Contributing | Code Style |
          Release Process
Length: ~300 lines
Key strength: Package boundary documentation
Key weakness: None significant -- this is the benchmark