create-or-audit-claude-md

To generate a CLAUDE.md from your codebase, scan configuration files, CI/CD pipelines, git history, and code patterns to extract verified project conventions. This ensures the generated file reflects actual enforced tooling, build commands, and directory structure rather than manual assumptions.

AI coding assistants give incorrect suggestions when your CLAUDE.md file is outdated or contains generic rules that don't match actual project conventions. Auditing the file against real codebase evidence fixes gaps and removes stale content that misleads the AI.

Yes, you can audit an existing CLAUDE.md for inaccuracies by comparing it against verified codebase evidence like configuration files and CI/CD pipelines. This produces a prioritized list of missing conventions, suggested revisions, and companion tooling recommendations.

A CLAUDE.md for Claude Code onboarding should include universal build/test commands, directory structure, and non-standard code conventions extracted from real codebase evidence. It must avoid duplicating content already enforced by linters, formatters, or CI checks.

Yes, this works for generating AI context files for a monorepo by scanning the full project to capture actual build commands, directory structure, and code quality rules. It replaces manually writing an incomplete file with a verified, project-specific context.

You should not use a CLAUDE.md file to document conventions that are already enforced by linters, formatters, or CI checks. Duplicating automated checks creates stale content; the file should only capture non-standard rules and universal build commands not enforced elsewhere.