Memory & CLAUDE.md System

Bootstrap Your First CLAUDE.md

This generates a CLAUDE.md based on your project structure. Claude analyzes your codebase and creates relevant instructions.

What to Put in CLAUDE.md

Pro Tip: The #1 most useful thing in CLAUDE.md is your build/test commands. Claude references these constantly.

CLAUDE.local.md

For personal preferences that shouldn't be in git:

CLAUDE.local.md is automatically added to .gitignore.

🧠 Knowledge Check

Q: What's the difference between CLAUDE.md and CLAUDE.local.md?

CLAUDE.md is shared with your team (committed to git). It contains project-wide instructions like architecture, conventions, and commands.

CLAUDE.local.md is personal and gitignored. It contains your individual preferences like local URLs, test accounts, and personal workflow tweaks.

Modular Rules

For larger projects, organize instructions into focused files:

All .md files in .claude/rules/ are automatically loaded as project memory.

Path-Specific Rules

Rules can be scoped to specific files using YAML frontmatter:

Rules without a paths field load unconditionally.

🧠 Knowledge Check

Q: You have a rule file at .claude/rules/api.md with paths: ["src/api/**/*.ts"]. When does Claude load this rule?

Only when Claude is working with files that match the pattern src/api/**/*.ts. Path-specific rules load on-demand, not at session start. Rules without a paths field load unconditionally for every session.

CLAUDE.md Imports

CLAUDE.md files can import other files using @path/to/file syntax:

Import Rules

• Relative paths resolve from the file containing the import
• Max depth: 5 hops (imports can recursively import)
• Code spans/blocks are not evaluated: `@not/imported`
• To see all loaded files: /memory

✍️ Recall Exercise

Fill in the blanks:

• To bootstrap a CLAUDE.md for your project: /___
• To view all loaded memory files: /___
• Path-scoped rules go in: .claude/___/*.md

• /init
• /memory
• .claude/rules/*.md

Auto Memory

Auto memory is where Claude writes notes for itself based on what it discovers during sessions.

What Gets Remembered

• Project patterns: build commands, test conventions, code style
• Debugging insights: solutions to tricky problems, common errors
• Architecture notes: key files, module relationships
• Your preferences: communication style, workflow habits

Storage Location

How It Works

1. First 200 lines of MEMORY.md load into Claude's system prompt automatically
2. Topic files load on demand when Claude needs them
3. Claude reads and writes memory files during your session

Asking Claude to Remember

Tell Claude directly:

Managing Auto Memory

Control Auto Memory

Gotcha: Only the first 200 lines of MEMORY.md load automatically. Keep it concise! Topic files load on-demand.

🧠 Knowledge Check

Q: How many lines of MEMORY.md load automatically at session start?

200 lines. Only the first 200 lines of MEMORY.md are loaded into Claude's system prompt at startup. Topic files (like debugging.md) load on-demand when Claude needs that information.

How Memory Loads at Startup

Child directory CLAUDE.md files and auto memory topic files load on demand as Claude works with relevant files.

Compaction & Memory

When Claude's context fills up, it automatically compacts — summarizing the conversation to free space. This can lose important details.

Custom Compaction Instructions

You can also add instructions to your CLAUDE.md:

Pro Tip: Use a SessionStart hook with a compact matcher to re-inject critical context after compaction. We'll cover this in the Hooks lesson.

✍️ Recall Exercise

Fill in the blanks:

• Auto memory is stored at: ~/.claude/projects/<project>/___/
• To disable auto memory: export CLAUDE_CODE_DISABLE_AUTO_MEMORY=___
• CLAUDE.md files can import other files with: @___

• memory/
• 1
• @path/to/file

🔨 Try It Yourself

1. Navigate to a project and run /init to generate a CLAUDE.md
2. Open the generated file and customize it with your actual commands and conventions
3. Create a .claude/rules/ directory with one rule file (e.g., code-style.md)
4. Tell Claude: "Remember that we always use pnpm, not npm"
5. Run /memory to verify the auto memory was created
6. Start a new session and check if Claude remembers your instructions

Reflection: How does the generated CLAUDE.md compare to what you'd write manually?

🤔 What Would You Do?

Scenario: You have a monorepo with a React frontend and a Python backend. The frontend team wants Claude to use Prettier and ESLint. The backend team wants Claude to use Black and mypy. How would you organize memory?

Use path-specific rules in .claude/rules/:

The right rules load automatically based on which files Claude is working with.

📋 Key Takeaways (Review These Tomorrow)

Next up: Permissions & Security → — Learn the tiered permission system, sandboxing, and enterprise security.