claude-md-author_skill

Overview

This skill creates an effective CLAUDE.md that gives Claude the minimal, high-value project context it should load into every session. It focuses on WHAT (stack and structure), WHY (purpose and key decisions) and HOW (build/test/verify commands) to keep prompts compact and useful. Use this to initialize or refine project-specific Claude instructions so responses are accurate and actionable.

How this skill works

It inspects the repository for build/test commands, key folders, architectural decisions, and common patterns, then drafts a concise CLAUDE.md following the WHAT-WHY-HOW framework. It replaces code excerpts with file:line references, limits lines to a strict budget, and points to separate docs for task-specific procedures. The result is a focused, universal context file that Claude will include in every conversation.

When to use it

• Initializing Claude context for a new repo
• Adding or updating project context so Claude understands architecture
• Preparing a codebase for consistent automated code review or generation
• Refining instructions after major refactors or dependency changes
• Reducing noise when Claude is producing irrelevant or verbose output

Best practices

• Keep CLAUDE.md minimal: prioritize build/test commands, structure, and key decisions
• Follow WHAT-WHY-HOW and avoid task-specific step-by-step details in the core file
• Use file:line references instead of code snippets to avoid staleness
• Move one-off or detailed procedures into .claude/docs/ and reference them
• Aim for under 300 lines and prefer under ~60 lines to conserve instruction budget

Example use cases

• Create CLAUDE.md for a TypeScript Next.js app with Prisma and CI commands
• Update CLAUDE.md after introducing a new auth module and SSO flow
• Add a short commands section so Claude can run correct build/test steps
• Refactor an oversized CLAUDE.md by moving testing/deployment details to .claude/docs/
• Provide Claude quick pointers to architecture decisions using file:line references