creating-claude-md

To generate a CLAUDE.md for a new repository, run the bootstrap workflow to scan manifests and configs, answer a six-question interview, and output a lean file under 80 lines at the repo root.

The best way to bootstrap a CLAUDE.md from scratch is using an automated workflow that combines mechanical repo scanning with a focused user interview to capture project WHY, gotchas, and testing philosophy into a high-signal document.

Yes, CLAUDE.md generation works with existing documentation by scanning mechanical content like manifests, repo structure, configs, and git history to consolidate existing project details into a single high-signal file.

To keep CLAUDE.md under 80 lines for complex repos, the bootstrap workflow enforces a 120-line hard cap and automatically splits oversized sections into auxiliary agent_docs/*.md files while retaining core context in the main file.

You should not use an automated CLAUDE.md bootstrap workflow for auditing existing files or making incremental improvements, as this process is strictly designed for generating a fresh CLAUDE.md from scratch.

A CLAUDE.md should contain mechanical repo details from scanning plus judgment content captured via a six-question interview, covering project WHY, gotchas, out-of-scope areas, approval gates, and testing philosophy to effectively guide AI agents.