CLAUDE.md

CLAUDE.md is a configuration file that gives Claude Code project-specific operating instructions: what conventions to follow, what tone to use, how to handle content, and what rules to enforce. It’s automatically loaded at the start of every session, so the AI behaves consistently across conversations without being re-told the same things. Think of it as an onboarding document that turns a general-purpose AI into one that knows your project’s house rules.

The Simple Version

If PLAN.md is the memory (what the project is and where it’s going), CLAUDE.md is the personality (how to work on it). It’s a file in your project that gives Claude Code operating instructions specific to your project.

Think of it like onboarding a new employee. You don’t just tell them what the company does (that’s the plan). You also tell them the house rules: “We use this tone in emails. We format reports this way. We always run tests before shipping.” CLAUDE.md is that onboarding document for the AI.

Why It Matters

Without operating instructions, an AI assistant makes reasonable but generic choices. It might add comments you don’t want, use a tone that doesn’t match your brand, or organize files differently than your convention. CLAUDE.md eliminates the guesswork. The AI knows your project’s rules before it writes a single line.

This is especially powerful for content-heavy projects. On this site, CLAUDE.md contains rules about linking strategy (all terms link to the glossary, not to external sites), blog tone (conversational, no em dashes, explain jargon inline), and content workflow (identify technical terms automatically, create glossary entries as you go).

How It’s Used on This Site

The CLAUDE.md file for scottkrukowski.com contains the full set of operating instructions: blog writing rules, linking strategy, design tokens, commit rules, the build audit system, and conventions for The Vibe-Coder’s Dictionary. When Claude Code writes a blog post, it automatically follows these rules: linking terms to glossary pages, weaving in soft CTAs, and flagging new terms that need definitions. The system runs itself because the instructions are baked in.

See It in Action

Every blog post on this site follows conventions defined in CLAUDE.md. The linking strategy where tool names like Astro or Vercel link to glossary pages instead of external websites? That’s a CLAUDE.md rule. The way each post includes exactly one soft CTA? Also a CLAUDE.md rule.