CLAUDE.md as Infrastructure is the concept that a @Claude Code configuration file is not just a settings document — it is the foundational layer of an @Augmentation Stack, the simplest form of persistent AI memory, and the entry point for building a @human-AI augmentation practice.

Beyond Configuration

@Anthropic designed CLAUDE.md as a project-level instruction file: coding conventions, file structure notes, deployment rules. That's the floor. In practice, CLAUDE.md becomes the institutional memory of a project — the document where architecture decisions accumulate, where collaboration patterns are encoded, and where the AI's understanding of your work compounds across sessions. A well-maintained CLAUDE.md means every Claude Code session starts with context instead of cold prompting. The AI knows your branching strategy, your test requirements, your deployment pipeline, and the decisions you've already made. This isn't configuration. It's the Memory layer of an Augmentation Stack — and for many practitioners, it's the first infrastructure they build.

How to Think About It

1. Layer your CLAUDE.md files. Claude Code reads CLAUDE.md at three levels: global (~/.claude/CLAUDE.md), project root, and subdirectories. Global holds collaboration patterns and workflow principles. Project root holds architecture decisions and deployment rules. Subdirectories hold domain-specific context. This layering mirrors how institutional knowledge actually works — general principles at the top, specifics at the edges
2. Encode decisions, not just rules. "Use Tailwind CSS" is a rule. "We chose Tailwind over styled-components because the team values utility-first patterns and our design system maps to utility classes" is a decision. Decisions give the AI judgment. Rules give it compliance. An AI with judgment produces better work
3. Let it evolve. The best CLAUDE.md files are living documents — updated after corrections, enriched with lessons learned, and pruned when context becomes stale. @MythOS takes this further by syncing CLAUDE.md from a MythOS memo, making the configuration file accessible from any device and editable through any @MCP-connected client
4. Connect it to richer memory. CLAUDE.md is the entry point, not the ceiling. As your augmentation practice matures, the CLAUDE.md file points to deeper context — MCP servers for knowledge systems, augmentation memos for identity and voice, project memos for strategy and status. The file becomes an index into a larger memory architecture

Why It Matters

The difference between "using Claude Code" and "building with Claude Code" is often a single file. Practitioners with a well-maintained CLAUDE.md report fewer corrections, more consistent output, and sessions that feel like continuing a conversation rather than starting one. This is @Collaborative Augmentation at its simplest — persistent context that makes every interaction better than the last. My CLAUDE.md is a MythOS memo. It syncs to every device automatically. When I correct Claude — "never push to main without asking," "use the API, not direct SQLite writes" — that correction goes into the MythOS memo, propagates to the CLAUDE.md file, and is loaded into every future session across every machine. I haven't repeated a correction in months. That's the shift. Most people treat CLAUDE.md as a settings file you write once. I treat it as the foundation of a memory system. The file itself is 50 lines. The system behind it is 17,000+ memos, 57 agents, and a year of accumulated collaboration context. CLAUDE.md is the door. What's behind it is the infrastructure.

Contexts

• #agentic-augmentation
• #claude-code
• #collaborating-with-claude