codebase-onboarding

Run a structured four-phase analysis: reconnaissance of manifests and configs, architecture mapping, convention detection, and artifact generation. This produces an onboarding guide with the tech stack, directory map, request lifecycle, and common tasks.

Detect the project's stack, naming conventions, test commands, and build scripts, then write a concise CLAUDE.md summarizing them. If a CLAUDE.md already exists, read it first and enhance it while preserving existing project-specific instructions.

Yes, the architecture mapping phase identifies whether a project is a monolith, monorepo, microservices, or serverless setup. It also detects the API style such as REST, GraphQL, gRPC, or tRPC from configuration and code signals.

When the repo has no commits or only a shallow history, the git conventions section is skipped and the guide notes that history is unavailable. Other phases like stack detection and architecture mapping still proceed normally.

The analysis relies on detectable signals like manifests and configs, so ambiguous conventions are flagged as unknown rather than guessed. It avoids reading every file, which means deep business logic details still require manual exploration.