Configuration

Directory Structure


~/.claude/
├── CLAUDE.md              # Global instructions (loaded every session)
├── settings.json          # Permissions, hooks, allowed tools
├── settings.local.json    # Machine-specific overrides
├── commands/              # Slash commands (symlink -> vault)
├── agents/                # Agent definitions (symlink -> vault)
├── hooks/                 # Pre/post tool hooks (symlink -> vault)
├── scripts/               # Utility scripts (symlink -> vault)
├── skills/                # Multi-step skill packs (symlink -> vault)
├── cache/                 # Performance cache
├── projects/              # Per-project settings & history
└── todos/                 # Task lists

How CLAUDE.md Works

~/.claude/CLAUDE.md is loaded into every Claude Code session as system instructions. It defines:

Interaction style (direct, structured, no fluff)
Autonomous task protocols (validation loops, completion criteria)
Git sync rules (when to commit/push)
Handoff protocols (what to tell the human when automation ends)

Each project can also have its own CLAUDE.md at the project root for project-specific instructions.

Settings Overview

settings.json controls:

Setting	Purpose
`allowedTools`	Which tools Claude can use without asking (Bash, Edit, Write, etc.)
`hooks`	Scripts that run before/after tool use (context injection, validation)
`mcpServers`	MCP (Model Context Protocol) server connections

settings.local.json controls:

Machine-specific bash patterns to allow
Bypass mode for trusted environments

Hooks System

Hooks are scripts that execute automatically at specific points:

Hook	Trigger	Purpose
SessionStart	Every new session	Load context, display status
PostToolUse (Edit/Write)	After file edits	Validate changes, run checks
PreCompletionCheck	Before marking task done	Ensure validation passed